diff --git a/packages/types/src/global-settings.ts b/packages/types/src/global-settings.ts index 95f246dbe7..1e57533b1f 100644 --- a/packages/types/src/global-settings.ts +++ b/packages/types/src/global-settings.ts @@ -12,15 +12,18 @@ import { } from "./provider-settings.js" import { telemetrySettingsSchema } from "./telemetry.js" import { toolNamesSchema } from "./tool.js" +import { changeCardDetailSchema, type ChangeCardDetail } from "./message.js" import { type Keys } from "./type-fu.js" import { languagesSchema } from "./vscode.js" /** * Default delay in milliseconds after writes to allow diagnostics to detect potential problems. - * This delay is particularly important for Go and other languages where tools like goimports - * need time to automatically clean up unused imports. + * Defaults to 0: the write path adds no artificial pacing by default, and post-save + * diagnostics are reported after the (zero) delay. Users who rely on auto-formatters that + * settle asynchronously (e.g. goimports for Go) can raise this setting to give formatters + * time to settle before diagnostics are captured. */ -export const DEFAULT_WRITE_DELAY_MS = 1000 +export const DEFAULT_WRITE_DELAY_MS = 0 /** * Default values for the "auto-close files Zoo opened" settings. @@ -99,6 +102,21 @@ export const MAX_CHECKPOINT_TIMEOUT_SECONDS = 60 */ export const DEFAULT_CHECKPOINT_TIMEOUT_SECONDS = 15 +/** + * Whether per-write checkpoints and task-start baseline are enabled by default. + * Master switch for the B cluster of checkpoint features. + * @default true + */ +export const DEFAULT_PER_WRITE_CHECKPOINTS = true + +/** + * Default detail level for per-step change cards (B3a). + * "summary" keeps cards compact (file list with +/− counts; the UI fetches + * diffs lazily); "full" carries the unified diff inline per file. + * @default "summary" + */ +export const DEFAULT_CHANGE_CARD_DETAIL: ChangeCardDetail = "summary" + /** * GlobalSettings */ @@ -200,6 +218,19 @@ export const globalSettingsSchema = z.object({ .min(MIN_CHECKPOINT_TIMEOUT_SECONDS) .max(MAX_CHECKPOINT_TIMEOUT_SECONDS) .optional(), + /** + * Whether to record a shadow-git checkpoint after every successful write_to_file, + * edit_file, and apply_patch (per-write checkpoints), plus a task-start baseline. + * @default true + */ + perWriteCheckpoints: z.boolean().optional(), + /** + * Detail level for per-step change cards: "full" includes the unified diff + * inline for every changed file, "summary" carries only the file list with + * +/− counts (diffs are fetched lazily by the UI). + * @default "summary" + */ + changeCardDetail: changeCardDetailSchema.optional(), ttsEnabled: z.boolean().optional(), ttsSpeed: z.number().optional(), diff --git a/packages/types/src/message.ts b/packages/types/src/message.ts index 28d5af82ac..b7940d00a4 100644 --- a/packages/types/src/message.ts +++ b/packages/types/src/message.ts @@ -134,6 +134,7 @@ export function isNonBlockingAsk(ask: ClineAsk): ask is NonBlockingAsk { * - `mcp_server_response`: Response received from MCP server * - `subtask_result`: Result of a completed subtask * - `checkpoint_saved`: Indicates a checkpoint has been saved + * - `change_card`: Per-step change card summarizing the files a completed tool step wrote (B3a) * - `rooignore_error`: Error related to .rooignore file processing * - `diff_error`: Error occurred while applying a diff/patch * - `condense_context`: Context condensation/summarization has started @@ -162,6 +163,7 @@ export const clineSays = [ "mcp_server_response", "subtask_result", "checkpoint_saved", + "change_card", "rooignore_error", "diff_error", "condense_context", @@ -235,6 +237,49 @@ export const contextTruncationSchema = z.object({ export type ContextTruncation = z.infer +/** + * ChangeCard + * + * Payload of the per-step change card (B3a). The extension host emits one + * `say: "change_card"` message per completed tool write step, keyed by the + * shadow-git checkpoint the step produced. The JSON payload (see + * {@link ChangeCardData}) is carried in the message `text` field, the same + * way tool approval messages carry their serialized ClineSayTool. + * + * `detail: "full"` carries the unified diff inline for every file so the UI + * can render it directly; `detail: "summary"` carries only the file list with + * +/− counts and the UI fetches diffs lazily (B3b). Auto-approved steps are + * always emitted with `detail: "summary"` regardless of the user setting. + */ +export const changeCardDetailSchema = z.enum(["full", "summary"]) + +export type ChangeCardDetail = z.infer + +export const changeCardFileSchema = z.object({ + path: z.string(), + additions: z.number(), + deletions: z.number(), + /** + * Unified diff for this file. Only present when the card was emitted with + * `detail: "full"`; summary cards leave it out to stay compact. + */ + diff: z.string().optional(), +}) + +export type ChangeCardFile = z.infer + +export const changeCardSchema = z.object({ + /** Opaque step identifier, reserved for future tool-step tracking. */ + stepId: z.string().optional(), + /** Checkpoint commit SHAs produced by the step (one per per-write checkpoint). */ + checkpointIds: z.array(z.string()), + files: z.array(changeCardFileSchema), + totalFiles: z.number(), + detail: changeCardDetailSchema, +}) + +export type ChangeCardData = z.infer + /** * ClineMessage * diff --git a/packages/types/src/vscode-extension-host.ts b/packages/types/src/vscode-extension-host.ts index 5f6b579779..5b3d739683 100644 --- a/packages/types/src/vscode-extension-host.ts +++ b/packages/types/src/vscode-extension-host.ts @@ -5,7 +5,7 @@ import type { ProviderSettings, ProviderSettingsEntry } from "./provider-setting import type { HistoryItem } from "./history.js" import type { ModeConfig, PromptComponent } from "./mode.js" import type { Experiments } from "./experiment.js" -import type { ClineMessage, QueuedMessage } from "./message.js" +import type { ChangeCardDetail, ClineMessage, QueuedMessage } from "./message.js" import type { MarketplaceItem, MarketplaceInstalledMetadata, InstallMarketplaceItemOptions } from "./marketplace.js" import type { TodoItem } from "./todo.js" import type { CloudUserInfo, CloudOrganizationMembership, OrganizationAllowList, ShareVisibility } from "./cloud.js" @@ -108,6 +108,7 @@ export interface ExtensionMessage { | "fileContent" | "rooHistoryImportProgress" | "themeFixtureProbeRequest" + | "checkpointRollbackResult" text?: string /** For fileContent: { path, content, error? } */ fileContent?: { path: string; content: string | null; error?: string } @@ -254,6 +255,34 @@ export interface ExtensionMessage { copyProgressItemName?: string // folderSelected path?: string + /** For checkpointRollbackResult: outcome of a change-card rollback request (B3b). */ + checkpointRollbackResult?: CheckpointRollbackResult +} + +/** + * CheckpointRollbackResult + * + * Outcome of a change-card restore request (B3b), posted back to the webview + * that sent `checkpointRollbackFile` / `checkpointRollbackStep` / + * `checkpointRestoreLatestFile`. `cardTs` echoes the change-card message + * timestamp so the requesting card can correlate the result: per-file results + * carry `filePath`, per-step results carry the per-file outcomes in + * `files`, and `kind` tells a per-file result which control it belongs to + * (absent = rollback, so results posted before `kind` existed still route). + */ +export interface CheckpointRollbackResult { + /** The `ts` of the change_card message the result belongs to. */ + cardTs: number + /** Per-file scope: the file that was restored. */ + filePath?: string + success: boolean + error?: string + /** Per-step scope: the per-file outcomes. */ + files?: { filePath: string; success: boolean; error?: string }[] + /** Per-file scope: which control the result belongs to. Absent = rollback. */ + kind?: "rollback" | "restore-latest" + /** Per-file scope: true when a restore-latest found no recorded write and left the file as-is. */ + noOp?: boolean } export interface OpenAiCodexRateLimitsMessage { @@ -348,6 +377,8 @@ export type ExtensionState = Pick< enableCheckpoints: boolean checkpointTimeout: number // Timeout for checkpoint initialization in seconds (default: 15) + perWriteCheckpoints: boolean + changeCardDetail: ChangeCardDetail maxOpenTabsContext: number // Maximum number of VSCode open tabs to include in context (0-500) maxWorkspaceFiles: number // Maximum number of files to include in current working directory details (0-500) showRooIgnoredFiles: boolean // Whether to show .rooignore'd files in listings @@ -544,6 +575,9 @@ export interface WebviewMessage { | "openCustomModesSettings" | "checkpointDiff" | "checkpointRestore" + | "checkpointRollbackFile" + | "checkpointRollbackStep" + | "checkpointRestoreLatestFile" | "completionCheckpointDiff" | "completionCheckpointRestore" | "deleteMcpServer" @@ -786,6 +820,46 @@ export const checkoutRestorePayloadSchema = z.object({ export type CheckpointRestorePayload = z.infer +/** + * Payload of the `checkpointRollbackFile` webview message (B3b): restore one + * change-card file to the checkpoint commit the card was keyed by. + */ +export const checkpointRollbackFilePayloadSchema = z.object({ + /** The `ts` of the change_card message the request comes from (echoed on the result). */ + cardTs: z.number(), + checkpointId: z.string(), + filePath: z.string(), +}) + +export type CheckpointRollbackFilePayload = z.infer + +/** + * Payload of the `checkpointRollbackStep` webview message (B3b): restore + * every file of a change-card step to the step's checkpoint. + */ +export const checkpointRollbackStepPayloadSchema = z.object({ + cardTs: z.number(), + /** The step's checkpoint commit (the card's first checkpointId); optional. */ + checkpointId: z.string().optional(), + filePaths: z.array(z.string()).min(1), +}) + +export type CheckpointRollbackStepPayload = z.infer + +/** + * Payload of the `checkpointRestoreLatestFile` webview message (B3b): restore + * one change-card file to the latest recorded version of that file (the + * content of its most recent write checkpoint — the forward direction to a + * rollback). + */ +export const checkpointRestoreLatestFilePayloadSchema = z.object({ + /** The `ts` of the change_card message the request comes from (echoed on the result). */ + cardTs: z.number(), + filePath: z.string(), +}) + +export type CheckpointRestoreLatestFilePayload = z.infer + export interface IndexingStatusPayload { state: "Standby" | "Indexing" | "Indexed" | "Error" | "Stopping" message: string @@ -799,6 +873,9 @@ export interface IndexClearedPayload { export type WebViewMessagePayload = | CheckpointDiffPayload | CheckpointRestorePayload + | CheckpointRollbackFilePayload + | CheckpointRollbackStepPayload + | CheckpointRestoreLatestFilePayload | IndexingStatusPayload | IndexClearedPayload | UpdateTodoListPayload diff --git a/src/core/checkpoints/__tests__/changeCard.spec.ts b/src/core/checkpoints/__tests__/changeCard.spec.ts new file mode 100644 index 0000000000..82190638a1 --- /dev/null +++ b/src/core/checkpoints/__tests__/changeCard.spec.ts @@ -0,0 +1,108 @@ +import { describe, expect, it } from "vitest" + +import type { ChangeCardData } from "@roo-code/types" + +import { + buildChangeCard, + buildChangeCardPayload, + isAutoApprovedStep, + resolveChangeCardDetail, + type ChangeCardWrite, +} from "../changeCard" + +describe("changeCard (B3a)", () => { + function write(overrides: Partial = {}): ChangeCardWrite { + return { + path: "src/a.ts", + diffStats: { additions: 2, deletions: 1 }, + diff: "--- a/src/a.ts\n+++ b/src/a.ts\n@@ -1 +1,2 @@\n-old\n+new-1\n+new-2", + ...overrides, + } + } + + describe("isAutoApprovedStep", () => { + it("returns false for an empty step", () => { + expect(isAutoApprovedStep([])).toBe(false) + }) + + it("returns true only when every write was auto-approved", () => { + expect(isAutoApprovedStep([write({ autoApproved: true }), write({ autoApproved: true })])).toBe(true) + expect(isAutoApprovedStep([write({ autoApproved: true }), write()])).toBe(false) + expect(isAutoApprovedStep([write()])).toBe(false) + }) + }) + + describe("resolveChangeCardDetail", () => { + it("forces summary for auto-approved steps even when the setting is full", () => { + const writes = [write({ autoApproved: true })] + expect(resolveChangeCardDetail(writes, "full")).toBe("summary") + expect(resolveChangeCardDetail(writes, undefined)).toBe("summary") + }) + + it("follows the setting for interactive steps, defaulting to summary when unset", () => { + const writes = [write()] + expect(resolveChangeCardDetail(writes, "full")).toBe("full") + expect(resolveChangeCardDetail(writes, "summary")).toBe("summary") + expect(resolveChangeCardDetail(writes, undefined)).toBe("summary") + }) + }) + + describe("buildChangeCard", () => { + it("carries the inline diff per file for full detail on a multi-file step", () => { + const card = buildChangeCard( + "sha-1", + [write(), write({ path: "src/b.ts", diffStats: { additions: 1, deletions: 0 }, diff: "+b" })], + "full", + ) + + expect(card).toEqual({ + checkpointIds: ["sha-1"], + files: [ + { + path: "src/a.ts", + additions: 2, + deletions: 1, + diff: "--- a/src/a.ts\n+++ b/src/a.ts\n@@ -1 +1,2 @@\n-old\n+new-1\n+new-2", + }, + { path: "src/b.ts", additions: 1, deletions: 0, diff: "+b" }, + ], + totalFiles: 2, + detail: "full", + }) + }) + + it("omits the diff per file for summary detail (lazy fetch is B3b)", () => { + const card = buildChangeCard("sha-1", [write()], "summary") + + expect(card.files).toEqual([{ path: "src/a.ts", additions: 2, deletions: 1 }]) + expect(card.files[0]).not.toHaveProperty("diff") + expect(card.detail).toBe("summary") + expect(card.totalFiles).toBe(1) + }) + + it("defaults missing diffStats to zero counts and keeps full detail without diff for a write without one", () => { + const card = buildChangeCard("sha-1", [write({ diffStats: undefined, diff: undefined })], "full") + + expect(card.files[0]).toEqual({ path: "src/a.ts", additions: 0, deletions: 0 }) + }) + }) + + describe("buildChangeCardPayload", () => { + it("resolves the detail level and builds the payload in one call", () => { + // The expectations are typed against the shared ChangeCardData + // contract in @roo-code/types, so the builder's output is checked + // against the same single source of truth the webview consumes. + // Interactive step with the full setting: diff inline. + const full: ChangeCardData = buildChangeCardPayload("sha-1", [write()], "full") + expect(full.detail).toBe("full") + expect(full.files[0].diff).toBe("--- a/src/a.ts\n+++ b/src/a.ts\n@@ -1 +1,2 @@\n-old\n+new-1\n+new-2") + + // Auto-approved step with the full setting: compact summary, no diff. + const compact: ChangeCardData = buildChangeCardPayload("sha-1", [write({ autoApproved: true })], "full") + expect(compact.detail).toBe("summary") + expect(compact.files[0]).not.toHaveProperty("diff") + expect(compact.checkpointIds).toEqual(["sha-1"]) + expect(compact.totalFiles).toBe(1) + }) + }) +}) diff --git a/src/core/checkpoints/__tests__/changeJournal.spec.ts b/src/core/checkpoints/__tests__/changeJournal.spec.ts new file mode 100644 index 0000000000..20f53551aa --- /dev/null +++ b/src/core/checkpoints/__tests__/changeJournal.spec.ts @@ -0,0 +1,150 @@ +import fs from "fs/promises" +import os from "os" +import path from "path" + +import { afterEach, beforeEach, describe, expect, it } from "vitest" + +import { appendChange, journalPath, loadChanges, type ChangeJournalEntry } from "../changeJournal" + +describe("changeJournal", () => { + const taskId = "test-task" + + let tmpRoot: string + + beforeEach(async () => { + tmpRoot = await fs.mkdtemp(path.join(os.tmpdir(), "b2-journal-")) + }) + + afterEach(async () => { + await fs.rm(tmpRoot, { recursive: true, force: true }) + }) + + function entry(overrides: Partial = {}): ChangeJournalEntry { + return { + path: "src/foo.ts", + operation: "create", + checkpointId: "abc12345", + ...overrides, + } + } + + describe("appendChange", () => { + it("writes one JSON line per entry with the documented field shape", async () => { + await appendChange(tmpRoot, taskId, entry({ operation: "create", checkpointId: "aaa" })) + + const raw = await fs.readFile(journalPath(tmpRoot, taskId), "utf8") + const lines = raw.split("\n").filter((line) => line !== "") + expect(lines).toHaveLength(1) + const parsed = JSON.parse(lines[0]) as ChangeJournalEntry + expect(parsed.path).toBe("src/foo.ts") + expect(parsed.operation).toBe("create") + expect(parsed.checkpointId).toBe("aaa") + }) + + it("appends multiple entries sequentially", async () => { + await appendChange(tmpRoot, taskId, entry({ checkpointId: "a" })) + await appendChange(tmpRoot, taskId, entry({ checkpointId: "b" })) + + const raw = await fs.readFile(journalPath(tmpRoot, taskId), "utf8") + expect(raw.split("\n").filter((line) => line !== "")).toHaveLength(2) + }) + }) + + describe("loadChanges", () => { + it("returns [] for an absent journal file", async () => { + expect(await loadChanges(tmpRoot, taskId)).toEqual([]) + }) + + it("returns [] for an empty journal file", async () => { + await fs.mkdir(path.dirname(journalPath(tmpRoot, taskId)), { recursive: true }) + await fs.writeFile(journalPath(tmpRoot, taskId), "") + + expect(await loadChanges(tmpRoot, taskId)).toEqual([]) + }) + + it("propagates non-ENOENT read failures instead of reporting an empty journal", async () => { + // A directory at the journal path makes readFile fail with EISDIR — + // a stand-in for any permission or I/O failure (EACCES etc.). Such a + // failure must not be swallowed into "no changes": it would let a + // rollback report a no-op success without reading the history. + await fs.mkdir(journalPath(tmpRoot, taskId), { recursive: true }) + + await expect(loadChanges(tmpRoot, taskId)).rejects.toMatchObject({ code: "EISDIR" }) + }) + + it("parses all entries in order with a clean tail", async () => { + await appendChange(tmpRoot, taskId, entry({ checkpointId: "x" })) + await appendChange(tmpRoot, taskId, entry({ checkpointId: "y" })) + await appendChange(tmpRoot, taskId, entry({ checkpointId: "z" })) + + const result = await loadChanges(tmpRoot, taskId) + expect(result).toHaveLength(3) + expect(result[0].checkpointId).toBe("x") + expect(result[1].checkpointId).toBe("y") + expect(result[2].checkpointId).toBe("z") + }) + + it("parses a journal whose final line has no trailing newline", async () => { + await appendChange(tmpRoot, taskId, entry({ checkpointId: "ok" })) + + // Rewrite the file without the trailing newline of the last line. + const filePath = journalPath(tmpRoot, taskId) + const content = (await fs.readFile(filePath, "utf8")).replace(/\n$/, "") + await fs.writeFile(filePath, content) + + const result = await loadChanges(tmpRoot, taskId) + expect(result).toHaveLength(1) + expect(result[0].checkpointId).toBe("ok") + }) + + it("discards a torn final line and returns the complete entries", async () => { + await appendChange(tmpRoot, taskId, entry({ checkpointId: "ok" })) + + // Append a second line truncated mid-content, with no trailing newline. + await fs.appendFile(journalPath(tmpRoot, taskId), '{"path":"src/half.ts","operation":"upd') + + const result = await loadChanges(tmpRoot, taskId) + expect(result).toHaveLength(1) + expect(result[0].checkpointId).toBe("ok") + }) + + it("skips a corrupt middle line and still loads the later valid entries", async () => { + await appendChange(tmpRoot, taskId, entry({ checkpointId: "ok" })) + + // Corrupt the first line in place, then append a valid entry after it. + await fs.writeFile( + journalPath(tmpRoot, taskId), + '"{"path":"src/corrupt.ts","operation":"update"\n' + + JSON.stringify(entry({ checkpointId: "after" })) + + "\n", + ) + + const result = await loadChanges(tmpRoot, taskId) + expect(result).toHaveLength(1) + expect(result[0].checkpointId).toBe("after") + }) + + it("does not throw when the entire journal is torn", async () => { + await appendChange(tmpRoot, taskId, entry({ checkpointId: "first" })) + + // Truncate to a single character — definitely invalid JSON. + await fs.writeFile(journalPath(tmpRoot, taskId), "{") + + expect(await loadChanges(tmpRoot, taskId)).toEqual([]) + }) + + it("includes diffStats when present", async () => { + await appendChange(tmpRoot, taskId, entry({ checkpointId: "s", diffStats: { additions: 5, deletions: 2 } })) + + const result = await loadChanges(tmpRoot, taskId) + expect(result[0].diffStats).toEqual({ additions: 5, deletions: 2 }) + }) + + it("omits diffStats when not provided", async () => { + await appendChange(tmpRoot, taskId, entry({ checkpointId: "n" })) + + const result = await loadChanges(tmpRoot, taskId) + expect(result[0].diffStats).toBeUndefined() + }) + }) +}) diff --git a/src/core/checkpoints/__tests__/checkpointJournal.test.ts b/src/core/checkpoints/__tests__/checkpointJournal.test.ts new file mode 100644 index 0000000000..8671bb491c --- /dev/null +++ b/src/core/checkpoints/__tests__/checkpointJournal.test.ts @@ -0,0 +1,218 @@ +import fs from "fs/promises" +import os from "os" +import path from "path" + +import { afterEach, beforeEach, describe, expect, it, vi, type Mock } from "vitest" + +import type { Task } from "../../task/Task" +import { loadChanges } from "../changeJournal" +import { checkpointSave, type CheckpointWriteInfo } from "../index" + +// Mock the VS Code API surface (index.ts imports vscode at module level). +vi.mock("vscode", () => ({ + window: { + showErrorMessage: vi.fn(), + showInformationMessage: vi.fn(), + createTextEditorDecorationType: vi.fn(() => ({})), + }, + Uri: { + file: vi.fn((p: string) => ({ fsPath: p })), + parse: vi.fn((uri: string) => ({ with: vi.fn(() => ({})) })), + }, + commands: { + executeCommand: vi.fn(), + }, +})) + +vi.mock("@roo-code/telemetry", () => ({ + TelemetryService: { + instance: { + captureCheckpointCreated: vi.fn(), + captureCheckpointRestored: vi.fn(), + captureCheckpointDiffed: vi.fn(), + }, + }, +})) + +vi.mock("../../../utils/path", () => ({ + getWorkspacePath: vi.fn(() => "/test/workspace"), +})) + +vi.mock("../../../utils/git", () => ({ + checkGitInstalled: vi.fn().mockResolvedValue(true), +})) + +vi.mock("../../../i18n", () => ({ + t: vi.fn((key: string) => key), +})) + +vi.mock("p-wait-for", () => ({ + default: vi.fn(), +})) + +// The real service would require a git repo; the checkpointSave under test +// only needs the pre-initialized service on the task. +vi.mock("../../../services/checkpoints") + +const TASK_ID = "journal-test-task" +const COMMIT = "test-commit-hash" + +interface ServiceLike { + isInitialized: boolean + saveCheckpoint: (...args: unknown[]) => Promise +} + +interface ProviderLike { + context: { globalStorageUri: { fsPath: string } } + log: (...args: unknown[]) => void + postMessageToWebview: (...args: unknown[]) => void + getState: () => Promise> +} + +interface TaskLike { + taskId: string + enableCheckpoints: boolean + checkpointService: ServiceLike + checkpointServiceInitializing: boolean + providerRef: { deref: () => ProviderLike | undefined } + say: (...args: unknown[]) => Promise +} + +describe("checkpointSave change-journal wiring (B2)", () => { + let tmpStorageDir: string + let saveCheckpointSpy: Mock + let mockProvider: ProviderLike + let mockTask: TaskLike + const write: CheckpointWriteInfo = { + path: "src/foo.ts", + operation: "create", + diffStats: { additions: 3, deletions: 0 }, + } + + beforeEach(async () => { + tmpStorageDir = await fs.mkdtemp(path.join(os.tmpdir(), "b2-journal-wiring-")) + saveCheckpointSpy = vi.fn().mockResolvedValue({ commit: COMMIT }) + mockProvider = { + context: { globalStorageUri: { fsPath: tmpStorageDir } }, + log: vi.fn(), + postMessageToWebview: vi.fn(), + // B3a: the card emission reads the live settings through getState. + getState: vi.fn().mockResolvedValue({}), + } + // Structural test double for Task (the class is not instantiated at + // this unit layer); the cast is safe because the fields checkpointSave + // reads are exactly these. + mockTask = { + taskId: TASK_ID, + enableCheckpoints: true, + checkpointService: { isInitialized: true, saveCheckpoint: saveCheckpointSpy }, + checkpointServiceInitializing: false, + providerRef: { deref: () => mockProvider }, + // B3a: the card emission calls task.say; a resolved double keeps the + // test double complete instead of letting the emission take the + // error path. + say: vi.fn().mockResolvedValue(undefined), + } + }) + + afterEach(async () => { + vi.restoreAllMocks() + await fs.rm(tmpStorageDir, { recursive: true, force: true }) + }) + + it("appends exactly one journal line referencing the B1 checkpoint id for a per-write save", async () => { + await checkpointSave(mockTask as Task, false, true, write) + + const entries = await loadChanges(tmpStorageDir, TASK_ID) + expect(entries).toHaveLength(1) + expect(entries[0]).toEqual({ + path: "src/foo.ts", + operation: "create", + checkpointId: COMMIT, + diffStats: { additions: 3, deletions: 0 }, + }) + + // The raw file holds exactly one JSON line. + const journalFile = path.join(tmpStorageDir, "tasks", TASK_ID, "checkpoints", "changes.jsonl") + const raw = await fs.readFile(journalFile, "utf8") + expect(raw.split("\n").filter((line) => line !== "")).toHaveLength(1) + }) + + it("omits diffStats in the journal entry when not provided", async () => { + await checkpointSave(mockTask as Task, false, true, { path: "src/bar.ts", operation: "update" }) + + const entries = await loadChanges(tmpStorageDir, TASK_ID) + expect(entries).toHaveLength(1) + expect(entries[0].path).toBe("src/bar.ts") + expect(entries[0].operation).toBe("update") + expect(entries[0].checkpointId).toBe(COMMIT) + expect(entries[0].diffStats).toBeUndefined() + }) + + it("does not write a journal entry for non-write checkpoint saves (task-start baseline)", async () => { + await checkpointSave(mockTask as Task) + + const journalFile = path.join(tmpStorageDir, "tasks", TASK_ID, "checkpoints", "changes.jsonl") + await expect(fs.stat(journalFile)).rejects.toThrow() + expect(await loadChanges(tmpStorageDir, TASK_ID)).toEqual([]) + }) + + it("appends one entry per file change for a multi-file write (apply-patch shape)", async () => { + await checkpointSave(mockTask as Task, false, true, [ + { path: "src/a.ts", operation: "create" }, + { path: "src/b.ts", operation: "update" }, + { path: "src/c.ts", operation: "delete" }, + ]) + + const entries = await loadChanges(tmpStorageDir, TASK_ID) + expect(entries).toHaveLength(3) + // Every entry references the single checkpoint of the whole patch. + expect(entries.map((entry) => entry.checkpointId)).toEqual([COMMIT, COMMIT, COMMIT]) + expect(entries.map((entry) => entry.path)).toEqual(["src/a.ts", "src/b.ts", "src/c.ts"]) + expect(entries.map((entry) => entry.operation)).toEqual(["create", "update", "delete"]) + }) + + it("keeps the existing error-swallowing behavior and skips the journal on save failure", async () => { + saveCheckpointSpy.mockRejectedValueOnce(new Error("git exploded")) + + await expect(checkpointSave(mockTask as Task, false, true, write)).resolves.toBeUndefined() + expect(mockTask.enableCheckpoints).toBe(false) + expect(await loadChanges(tmpStorageDir, TASK_ID)).toEqual([]) + }) + + it("does not write a journal entry when the checkpoint save is a no-op (empty commit)", async () => { + saveCheckpointSpy.mockResolvedValueOnce(undefined) + + await checkpointSave(mockTask as Task, false, true, write) + + expect(await loadChanges(tmpStorageDir, TASK_ID)).toEqual([]) + expect(mockTask.enableCheckpoints).toBe(true) + }) + + it("does not crash when the provider has no globalStorageDir", async () => { + mockTask.providerRef = { deref: () => undefined } + + await expect(checkpointSave(mockTask as Task, false, true, write)).resolves.toMatchObject({ commit: COMMIT }) + expect(mockTask.enableCheckpoints).toBe(true) + expect(await loadChanges(tmpStorageDir, TASK_ID)).toEqual([]) + }) + + it("logs and continues when the journal cannot be written (checkpoints stay enabled)", async () => { + // Block the per-task checkpoint dir so the journal mkdir/append fails. + const taskDir = path.join(tmpStorageDir, "tasks", TASK_ID) + await fs.mkdir(taskDir, { recursive: true }) + await fs.writeFile(path.join(taskDir, "checkpoints"), "blocker") + + const consoleErrorSpy = vi.spyOn(console, "error").mockImplementation(() => undefined) + + // The journal failure is swallowed (logged, not rethrown), so the + // checkpoint result still resolves exactly as without journaling. + await expect(checkpointSave(mockTask as Task, false, true, write)).resolves.toMatchObject({ commit: COMMIT }) + + expect(mockTask.enableCheckpoints).toBe(true) + expect(consoleErrorSpy).toHaveBeenCalledWith( + expect.stringContaining("failed to append change journal entry"), + expect.anything(), + ) + }) +}) diff --git a/src/core/checkpoints/__tests__/checkpointSave.spec.ts b/src/core/checkpoints/__tests__/checkpointSave.spec.ts new file mode 100644 index 0000000000..25afe4cad7 --- /dev/null +++ b/src/core/checkpoints/__tests__/checkpointSave.spec.ts @@ -0,0 +1,223 @@ +import fs from "fs/promises" +import os from "os" +import path from "path" + +import { afterEach, beforeEach, describe, expect, it, vi } from "vitest" + +import type { Task } from "../../task/Task" +import { journalPath } from "../changeJournal" +import { checkpointSave } from "../index" + +vi.mock("@roo-code/telemetry", () => ({ + TelemetryService: { + instance: { + captureCheckpointCreated: vi.fn(), + captureCheckpointRestored: vi.fn(), + captureCheckpointDiffed: vi.fn(), + }, + }, +})) + +/** + * Minimal Task double for the checkpointSave wiring tests. Only the members + * touched by getCheckpointService + checkpointSave are provided: a pre-set + * checkpointService (so no git installation happens), the provider context + * (journal dir + setting state), and say. Structural cast at the boundary, + * matching the documented test-double style of the tool specs. + */ +function makeTask(options: { saveCheckpoint?: unknown; state?: Record; enableCheckpoints?: boolean }) { + const say = vi.fn().mockResolvedValue(undefined) + // An explicit `saveCheckpoint: undefined` (a checkpoint that produced no + // commit) must be preserved as-is; only an omitted option falls back to + // the default commit result. + const saveCheckpoint = vi + .fn() + .mockResolvedValue("saveCheckpoint" in options ? options.saveCheckpoint : { commit: "sha-card-1" }) + const providerDeref = { + context: { globalStorageUri: { fsPath: globalStorageDir } }, + getState: vi.fn().mockResolvedValue(options.state ?? {}), + } + + const task = { + taskId: "task-card", + cwd: "/workspace", + enableCheckpoints: options.enableCheckpoints ?? true, + checkpointService: { + saveCheckpoint, + isInitialized: true, + }, + providerRef: { deref: vi.fn().mockReturnValue(providerDeref) }, + say, + } as unknown as Task + + return { task, say, saveCheckpoint } +} + +let globalStorageDir: string + +beforeEach(async () => { + globalStorageDir = await fs.mkdtemp(path.join(os.tmpdir(), "b3a-save-")) +}) + +afterEach(async () => { + await fs.rm(globalStorageDir, { recursive: true, force: true }) +}) + +describe("checkpointSave change-card emission (B3a)", () => { + it("emits a full-detail change card after a successful per-write checkpoint and still appends the journal", async () => { + const { task, say, saveCheckpoint } = makeTask({ state: { changeCardDetail: "full" } }) + + await checkpointSave(task, false, true, { + path: "src/a.ts", + operation: "create", + diffStats: { additions: 2, deletions: 1 }, + diff: "+line-a\n+line-b\n-old", + }) + + expect(saveCheckpoint).toHaveBeenCalledWith(expect.stringContaining("task-card"), expect.any(Object)) + + const cardCalls = say.mock.calls.filter(([type]) => type === "change_card") + expect(cardCalls).toHaveLength(1) + const [type, text, images, partial, sayOptions, _progress, options] = cardCalls[0] as unknown as [ + string, + string, + undefined, + undefined, + undefined, + undefined, + { isNonInteractive?: boolean }, + ] + expect(type).toBe("change_card") + expect(images).toBeUndefined() + expect(options).toEqual({ isNonInteractive: true }) + const card = JSON.parse(text as string) as { + checkpointIds: string[] + files: Array<{ path: string; additions: number; deletions: number; diff?: string }> + totalFiles: number + detail: string + } + expect(card.checkpointIds).toEqual(["sha-card-1"]) + expect(card.totalFiles).toBe(1) + expect(card.detail).toBe("full") + expect(card.files).toEqual([{ path: "src/a.ts", additions: 2, deletions: 1, diff: "+line-a\n+line-b\n-old" }]) + + // B2 regression: the journal entry is still appended with the commit id. + const journalRaw = await fs.readFile(journalPath(globalStorageDir, "task-card"), "utf8") + const entries = journalRaw + .split("\n") + .filter((line) => line !== "") + .map((line) => JSON.parse(line)) + expect(entries).toHaveLength(1) + expect(entries[0]).toMatchObject({ path: "src/a.ts", operation: "create", checkpointId: "sha-card-1" }) + }) + + it("emits a summary card without diffs for the default (summary) setting", async () => { + const { task, say } = makeTask({}) + + await checkpointSave(task, false, true, { + path: "src/a.ts", + operation: "update", + diffStats: { additions: 1, deletions: 0 }, + diff: "+x", + }) + + const cardCall = say.mock.calls.find(([type]) => type === "change_card") + expect(cardCall).toBeDefined() + const card = JSON.parse((cardCall as unknown as [string, string])[1]) as { + files: Array> + detail: string + } + expect(card.detail).toBe("summary") + expect(card.files[0]).not.toHaveProperty("diff") + }) + + it("emits a compact card for auto-approved steps even when the setting is full", async () => { + const { task, say } = makeTask({ state: { changeCardDetail: "full" } }) + + await checkpointSave(task, false, true, { + path: "src/a.ts", + operation: "create", + diffStats: { additions: 1, deletions: 0 }, + diff: "+x", + autoApproved: true, + }) + + const cardCall = say.mock.calls.find(([type]) => type === "change_card") + expect(cardCall).toBeDefined() + const card = JSON.parse((cardCall as unknown as [string, string])[1]) as { + files: Array> + detail: string + } + expect(card.detail).toBe("summary") + expect(card.files[0]).not.toHaveProperty("diff") + }) + + it("emits one card with all writes for a multi-file step", async () => { + const { task, say } = makeTask({ state: { changeCardDetail: "full" } }) + + await checkpointSave(task, false, true, [ + { path: "src/a.ts", operation: "create", diffStats: { additions: 2, deletions: 0 }, diff: "+a1\n+a2" }, + { path: "src/b.ts", operation: "delete", diffStats: { additions: 0, deletions: 3 } }, + ]) + + const cardCalls = say.mock.calls.filter(([type]) => type === "change_card") + expect(cardCalls).toHaveLength(1) + const card = JSON.parse((cardCalls[0] as unknown as [string, string])[1]) as { + files: Array<{ path: string; diff?: string }> + totalFiles: number + } + expect(card.totalFiles).toBe(2) + expect(card.files.map((file) => file.path)).toEqual(["src/a.ts", "src/b.ts"]) + expect(card.files[0].diff).toBe("+a1\n+a2") + expect(card.files[1]).not.toHaveProperty("diff") + }) + + it("emits no change card for baseline checkpoints without write info", async () => { + const { task, say } = makeTask({}) + + await checkpointSave(task) + + // `say` is invoked with seven arguments, so a three-argument + // `toHaveBeenCalledWith` negative assertion can never fail; filter the + // recorded calls by type instead. + const cardCalls = say.mock.calls.filter(([type]) => type === "change_card") + expect(cardCalls).toHaveLength(0) + }) + + it("emits no change card when the checkpoint produced no commit", async () => { + const { task, say } = makeTask({ saveCheckpoint: undefined }) + + await checkpointSave(task, false, true, { path: "src/a.ts", operation: "create" }) + + const cardCalls = say.mock.calls.filter(([type]) => type === "change_card") + expect(cardCalls).toHaveLength(0) + }) + + it("emits no change card when checkpoints are disabled for the task", async () => { + const { task, say } = makeTask({ enableCheckpoints: false }) + + await checkpointSave(task, false, true, { path: "src/a.ts", operation: "create", diff: "+x" }) + + expect(say).not.toHaveBeenCalled() + }) + + it("keeps the journal append when a card emission failure occurs", async () => { + const { task, say } = makeTask({}) + say.mockImplementation(async (type: string) => { + if (type === "change_card") { + throw new Error("task aborted") + } + }) + + await checkpointSave(task, false, true, { + path: "src/a.ts", + operation: "create", + diffStats: { additions: 1, deletions: 0 }, + diff: "+x", + }) + + // The say failure is contained: the journal is still written. + const journalRaw = await fs.readFile(journalPath(globalStorageDir, "task-card"), "utf8") + expect(journalRaw).toContain("src/a.ts") + }) +}) diff --git a/src/core/checkpoints/__tests__/rollback.spec.ts b/src/core/checkpoints/__tests__/rollback.spec.ts new file mode 100644 index 0000000000..0af6384113 --- /dev/null +++ b/src/core/checkpoints/__tests__/rollback.spec.ts @@ -0,0 +1,455 @@ +import fs from "fs/promises" +import os from "os" +import path from "path" + +import { afterEach, beforeEach, describe, expect, it, vi } from "vitest" + +import type { Task } from "../../task/Task" +import type { ChangeJournalEntry } from "../changeJournal" +import { getCheckpointService } from "../index" +import * as changeJournal from "../changeJournal" +import { appendChange, journalPath } from "../changeJournal" +import { restoreLatestFile, rollbackFile, rollbackStep } from "../rollback" + +vi.mock("../index", () => ({ + getCheckpointService: vi.fn(), + checkpointSave: vi.fn(), + checkpointRestore: vi.fn(), + checkpointDiff: vi.fn(), +})) + +const mockedGetCheckpointService = getCheckpointService as unknown as ReturnType + +function makeTask(): Task { + return { + taskId: "task-rollback", + providerRef: { + deref: vi.fn().mockReturnValue({ context: { globalStorageUri: { fsPath: globalStorageDir } } }), + }, + } as unknown as Task +} + +/** A checkpoint-service double with a recording restoreFile and a baseline. */ +function serviceWith(baseHash: string | undefined) { + const restoreFile = vi.fn().mockResolvedValue(undefined) + return { baseHash, restoreFile } +} + +async function seedJournal(entries: ChangeJournalEntry[]): Promise { + for (const entry of entries) { + await appendChange(globalStorageDir, "task-rollback", entry) + } +} + +let globalStorageDir: string + +beforeEach(async () => { + globalStorageDir = await fs.mkdtemp(path.join(os.tmpdir(), "b3c-rollback-")) + mockedGetCheckpointService.mockReset() +}) + +afterEach(async () => { + await fs.rm(globalStorageDir, { recursive: true, force: true }) +}) + +describe("rollbackFile (B3c: undo the step's write to the file)", () => { + it("restores the file to the PREVIOUS step's checkpoint when an earlier entry exists", async () => { + await seedJournal([ + { path: "src/a.ts", operation: "create", checkpointId: "sha-1" }, + { path: "src/a.ts", operation: "update", checkpointId: "sha-2" }, + ]) + const service = serviceWith("base-0") + mockedGetCheckpointService.mockResolvedValue(service) + + const outcome = await rollbackFile(makeTask(), "sha-2", "src/a.ts") + + expect(outcome).toEqual({ filePath: "src/a.ts", success: true }) + // The pre-step state is the previous step's post-write checkpoint — not + // the step's own (post-write) checkpoint. + expect(service.restoreFile).toHaveBeenCalledTimes(1) + expect(service.restoreFile).toHaveBeenCalledWith("sha-1", "src/a.ts") + }) + + it("restores from the task-start baseline when the file has no earlier entry", async () => { + await seedJournal([{ path: "src/a.ts", operation: "create", checkpointId: "sha-1" }]) + const service = serviceWith("base-0") + mockedGetCheckpointService.mockResolvedValue(service) + + const outcome = await rollbackFile(makeTask(), "sha-1", "src/a.ts") + + expect(outcome).toEqual({ filePath: "src/a.ts", success: true }) + expect(service.restoreFile).toHaveBeenCalledWith("base-0", "src/a.ts") + }) + + it("resolves through the first entry of a multi-write step", async () => { + // One patch writes the same file twice: two entries share the step + // checkpoint. The pre-step state is still the entry before the first + // one of the step. + await seedJournal([ + { path: "src/a.ts", operation: "create", checkpointId: "sha-1" }, + { path: "src/a.ts", operation: "update", checkpointId: "sha-2" }, + { path: "src/a.ts", operation: "update", checkpointId: "sha-2" }, + ]) + const service = serviceWith("base-0") + mockedGetCheckpointService.mockResolvedValue(service) + + const outcome = await rollbackFile(makeTask(), "sha-2", "src/a.ts") + + expect(outcome.success).toBe(true) + expect(service.restoreFile).toHaveBeenCalledWith("sha-1", "src/a.ts") + }) + + it("rejects rolling back a step that is not the file's latest change", async () => { + // The file was written again by a later step (sha-2): rolling back the + // older step (sha-1) would overwrite the newer state, so it is + // rejected instead of silently destroying it. + await seedJournal([ + { path: "src/a.ts", operation: "create", checkpointId: "sha-1" }, + { path: "src/a.ts", operation: "update", checkpointId: "sha-2" }, + ]) + const service = serviceWith("base-0") + mockedGetCheckpointService.mockResolvedValue(service) + + const outcome = await rollbackFile(makeTask(), "sha-1", "src/a.ts") + + expect(outcome).toEqual({ + filePath: "src/a.ts", + success: false, + error: "File was modified in a later step; roll back the latest change card first", + }) + expect(service.restoreFile).not.toHaveBeenCalled() + }) + + it("fails when the journal location is unavailable (no global storage)", async () => { + // No context on the provider double → the journal cannot even be + // located: a clear failure, not a silent miss on the file lookup. + const task = { + taskId: "task-rollback", + providerRef: { deref: vi.fn().mockReturnValue(undefined) }, + } as unknown as Task + + const service = serviceWith("base-0") + mockedGetCheckpointService.mockResolvedValue(service) + + const outcome = await rollbackFile(task, "sha-1", "src/a.ts") + + expect(outcome).toEqual({ + filePath: "src/a.ts", + success: false, + error: "Change journal is unavailable for this task", + }) + expect(service.restoreFile).not.toHaveBeenCalled() + }) + + it("fails cleanly when the file is not part of the given step checkpoint", async () => { + await seedJournal([{ path: "src/a.ts", operation: "create", checkpointId: "sha-1" }]) + const service = serviceWith("base-0") + mockedGetCheckpointService.mockResolvedValue(service) + + const outcome = await rollbackFile(makeTask(), "sha-2", "src/a.ts") + + expect(outcome).toEqual({ + filePath: "src/a.ts", + success: false, + error: "File is not part of this step's checkpoint", + }) + expect(service.restoreFile).not.toHaveBeenCalled() + }) + + it("fails cleanly when no earlier checkpoint exists and the baseline is unavailable", async () => { + await seedJournal([{ path: "src/a.ts", operation: "create", checkpointId: "sha-1" }]) + const service = serviceWith(undefined) + mockedGetCheckpointService.mockResolvedValue(service) + + const outcome = await rollbackFile(makeTask(), "sha-1", "src/a.ts") + + expect(outcome).toEqual({ + filePath: "src/a.ts", + success: false, + error: "No checkpoint available to restore", + }) + expect(service.restoreFile).not.toHaveBeenCalled() + }) + + it("fails cleanly when checkpoints are not enabled", async () => { + mockedGetCheckpointService.mockResolvedValue(undefined) + + const outcome = await rollbackFile(makeTask(), "sha-1", "src/a.ts") + + expect(outcome).toEqual({ + filePath: "src/a.ts", + success: false, + error: "Checkpoints are not enabled for this task", + }) + }) + + it("reports the service error without throwing", async () => { + await seedJournal([{ path: "src/a.ts", operation: "create", checkpointId: "sha-1" }]) + const service = serviceWith("base-0") + service.restoreFile.mockRejectedValue(new Error("pathspec did not match")) + mockedGetCheckpointService.mockResolvedValue(service) + + const outcome = await rollbackFile(makeTask(), "sha-1", "src/a.ts") + + expect(outcome.success).toBe(false) + expect(outcome.error).toContain("pathspec did not match") + }) + + it("stringifies non-Error rejections into the outcome", async () => { + await seedJournal([{ path: "src/a.ts", operation: "create", checkpointId: "sha-1" }]) + const service = serviceWith("base-0") + service.restoreFile.mockRejectedValue("raw failure") + mockedGetCheckpointService.mockResolvedValue(service) + + const outcome = await rollbackFile(makeTask(), "sha-1", "src/a.ts") + + expect(outcome).toEqual({ filePath: "src/a.ts", success: false, error: "raw failure" }) + }) +}) + +describe("rollbackStep (B3c: undo every file of the step)", () => { + it("restores every step file to its pre-step state", async () => { + await seedJournal([ + { path: "src/a.ts", operation: "create", checkpointId: "sha-1" }, + { path: "src/a.ts", operation: "update", checkpointId: "sha-2" }, + { path: "src/b.ts", operation: "update", checkpointId: "sha-2" }, + ]) + const service = serviceWith("base-0") + mockedGetCheckpointService.mockResolvedValue(service) + + const outcome = await rollbackStep(makeTask(), ["src/a.ts", "src/b.ts"], "sha-2") + + expect(outcome.checkpointId).toBe("sha-2") + expect(outcome.files).toEqual([ + { filePath: "src/a.ts", success: true }, + { filePath: "src/b.ts", success: true }, + ]) + expect(service.restoreFile).toHaveBeenCalledTimes(2) + expect(service.restoreFile).toHaveBeenNthCalledWith(1, "sha-1", "src/a.ts") + // src/b.ts has no earlier entry: its pre-step state is the baseline. + expect(service.restoreFile).toHaveBeenNthCalledWith(2, "base-0", "src/b.ts") + }) + + it("keeps per-file failures isolated from the other step files", async () => { + await seedJournal([ + { path: "src/a.ts", operation: "create", checkpointId: "sha-1" }, + { path: "src/a.ts", operation: "update", checkpointId: "sha-2" }, + { path: "src/b.ts", operation: "create", checkpointId: "sha-2" }, + ]) + const service = serviceWith("base-0") + service.restoreFile.mockRejectedValueOnce(new Error("pathspec did not match")) + mockedGetCheckpointService.mockResolvedValue(service) + + const outcome = await rollbackStep(makeTask(), ["src/a.ts", "src/b.ts"], "sha-2") + + expect(outcome.files[0].success).toBe(false) + expect(outcome.files[0].error).toContain("pathspec did not match") + expect(outcome.files[1]).toEqual({ filePath: "src/b.ts", success: true }) + }) + + it("fails the file cleanly when no earlier checkpoint exists and the baseline is unavailable", async () => { + await seedJournal([{ path: "src/a.ts", operation: "create", checkpointId: "sha-2" }]) + const service = serviceWith(undefined) + mockedGetCheckpointService.mockResolvedValue(service) + + const outcome = await rollbackStep(makeTask(), ["src/a.ts"], "sha-2") + + expect(outcome.checkpointId).toBe("sha-2") + expect(outcome.files).toEqual([ + { filePath: "src/a.ts", success: false, error: "No checkpoint available to restore" }, + ]) + expect(service.restoreFile).not.toHaveBeenCalled() + }) + + it("rejects a file that is not part of the given step checkpoint", async () => { + await seedJournal([{ path: "src/a.ts", operation: "create", checkpointId: "sha-2" }]) + const service = serviceWith("base-0") + mockedGetCheckpointService.mockResolvedValue(service) + + const outcome = await rollbackStep(makeTask(), ["src/a.ts", "src/other.ts"], "sha-2") + + expect(outcome.files[0]).toEqual({ filePath: "src/a.ts", success: true }) + expect(outcome.files[1].success).toBe(false) + expect(outcome.files[1].error).toBe("File is not part of this step's checkpoint") + expect(service.restoreFile).toHaveBeenCalledTimes(1) + }) + + it("rejects the stale file of a step while restoring the others", async () => { + // src/a.ts was written again after this step's checkpoint, so its + // sha-2 entry is no longer the file's latest: only src/b.ts (whose + // latest entry IS sha-2) is restored. + await seedJournal([ + { path: "src/a.ts", operation: "create", checkpointId: "sha-1" }, + { path: "src/a.ts", operation: "update", checkpointId: "sha-2" }, + { path: "src/b.ts", operation: "create", checkpointId: "sha-2" }, + { path: "src/a.ts", operation: "update", checkpointId: "sha-3" }, + ]) + const service = serviceWith("base-0") + mockedGetCheckpointService.mockResolvedValue(service) + + const outcome = await rollbackStep(makeTask(), ["src/a.ts", "src/b.ts"], "sha-2") + + expect(outcome.files[0].success).toBe(false) + expect(outcome.files[0].error).toBe("File was modified in a later step; roll back the latest change card first") + expect(outcome.files[1]).toEqual({ filePath: "src/b.ts", success: true }) + expect(service.restoreFile).toHaveBeenCalledTimes(1) + expect(service.restoreFile).toHaveBeenCalledWith("base-0", "src/b.ts") + }) + + it("falls back to the latest journal entry per file without a step checkpoint id", async () => { + await seedJournal([ + { path: "src/a.ts", operation: "create", checkpointId: "sha-1" }, + { path: "src/a.ts", operation: "update", checkpointId: "sha-2" }, + ]) + const service = serviceWith("base-0") + mockedGetCheckpointService.mockResolvedValue(service) + + const outcome = await rollbackStep(makeTask(), ["src/a.ts"]) + + expect(outcome.checkpointId).toBeUndefined() + expect(outcome.files).toEqual([{ filePath: "src/a.ts", success: true }]) + expect(service.restoreFile).toHaveBeenCalledWith("sha-2", "src/a.ts") + }) + + it("fails listed files without journal entries", async () => { + await seedJournal([{ path: "src/a.ts", operation: "create", checkpointId: "sha-1" }]) + const service = serviceWith("base-0") + mockedGetCheckpointService.mockResolvedValue(service) + + const outcome = await rollbackStep(makeTask(), ["src/a.ts", "src/missing.ts"]) + + expect(outcome.files[0]).toEqual({ filePath: "src/a.ts", success: true }) + expect(outcome.files[1].success).toBe(false) + expect(outcome.files[1].error).toBe("No change journal entry for this file") + }) + + it("fails per file when the journal location is unavailable (no global storage)", async () => { + // No context on the provider double → the journal cannot even be + // located. That is a failure, not an empty journal: reporting the + // step as merely "not part of this checkpoint" would be misleading. + const task = { + taskId: "task-rollback", + providerRef: { deref: vi.fn().mockReturnValue(undefined) }, + } as unknown as Task + + const service = serviceWith("base-0") + mockedGetCheckpointService.mockResolvedValue(service) + + const outcome = await rollbackStep(task, ["src/a.ts"], "sha-2") + + expect(outcome.checkpointId).toBe("sha-2") + expect(outcome.files[0].success).toBe(false) + expect(outcome.files[0].error).toBe("Change journal is unavailable for this task") + expect(service.restoreFile).not.toHaveBeenCalled() + }) + + it("fails every file when checkpoints are not enabled", async () => { + mockedGetCheckpointService.mockResolvedValue(undefined) + + const outcome = await rollbackStep(makeTask(), ["src/a.ts"], "sha-2") + + expect(outcome.checkpointId).toBe("sha-2") + expect(outcome.files).toEqual([ + { filePath: "src/a.ts", success: false, error: "Checkpoints are not enabled for this task" }, + ]) + }) +}) + +describe("restoreLatestFile (B3c: forward direction)", () => { + it("restores the file to its most recent recorded write checkpoint", async () => { + await seedJournal([ + { path: "src/a.ts", operation: "create", checkpointId: "sha-1" }, + { path: "src/b.ts", operation: "update", checkpointId: "sha-1" }, + { path: "src/a.ts", operation: "update", checkpointId: "sha-2" }, + ]) + const service = serviceWith("base-0") + mockedGetCheckpointService.mockResolvedValue(service) + + const outcome = await restoreLatestFile(makeTask(), "src/a.ts") + + expect(outcome).toEqual({ filePath: "src/a.ts", success: true }) + expect(service.restoreFile).toHaveBeenCalledWith("sha-2", "src/a.ts") + }) + + it("is a successful no-op for a file the task never wrote", async () => { + await seedJournal([{ path: "src/b.ts", operation: "update", checkpointId: "sha-1" }]) + const service = serviceWith("base-0") + mockedGetCheckpointService.mockResolvedValue(service) + + const outcome = await restoreLatestFile(makeTask(), "src/a.ts") + + expect(outcome).toEqual({ filePath: "src/a.ts", success: true, noOp: true }) + expect(service.restoreFile).not.toHaveBeenCalled() + }) + + it("fails when the journal location is unavailable (no global storage)", async () => { + // An unavailable journal is not "the task wrote nothing": a no-op + // success would claim a restore that never happened. + const task = { + taskId: "task-rollback", + providerRef: { deref: vi.fn().mockReturnValue(undefined) }, + } as unknown as Task + + const service = serviceWith("base-0") + mockedGetCheckpointService.mockResolvedValue(service) + + const outcome = await restoreLatestFile(task, "src/a.ts") + + expect(outcome).toEqual({ + filePath: "src/a.ts", + success: false, + error: "Change journal is unavailable for this task", + }) + expect(service.restoreFile).not.toHaveBeenCalled() + }) + + it("fails when the journal cannot be read (an I/O error is not an empty journal)", async () => { + // A directory at the journal path makes readFile fail with EISDIR — + // a stand-in for any permission or I/O failure (EACCES etc.). + await fs.mkdir(journalPath(globalStorageDir, "task-rollback"), { recursive: true }) + const service = serviceWith("base-0") + mockedGetCheckpointService.mockResolvedValue(service) + + const outcome = await restoreLatestFile(makeTask(), "src/a.ts") + + expect(outcome.success).toBe(false) + expect(outcome.error).toContain("Change journal could not be read") + expect(service.restoreFile).not.toHaveBeenCalled() + }) + + it("stringifies a non-Error journal read failure into the outcome", async () => { + vi.spyOn(changeJournal, "loadChanges").mockRejectedValueOnce("raw journal failure") + const service = serviceWith("base-0") + mockedGetCheckpointService.mockResolvedValue(service) + + const outcome = await restoreLatestFile(makeTask(), "src/a.ts") + + expect(outcome.success).toBe(false) + expect(outcome.error).toBe("Change journal could not be read: raw journal failure") + expect(service.restoreFile).not.toHaveBeenCalled() + }) + + it("fails cleanly when checkpoints are not enabled", async () => { + mockedGetCheckpointService.mockResolvedValue(undefined) + + const outcome = await restoreLatestFile(makeTask(), "src/a.ts") + + expect(outcome).toEqual({ + filePath: "src/a.ts", + success: false, + error: "Checkpoints are not enabled for this task", + }) + }) + + it("reports the service error without throwing", async () => { + await seedJournal([{ path: "src/a.ts", operation: "create", checkpointId: "sha-1" }]) + const service = serviceWith("base-0") + service.restoreFile.mockRejectedValue(new Error("index.lock held")) + mockedGetCheckpointService.mockResolvedValue(service) + + const outcome = await restoreLatestFile(makeTask(), "src/a.ts") + + expect(outcome.success).toBe(false) + expect(outcome.error).toContain("index.lock") + }) +}) diff --git a/src/core/checkpoints/changeCard.ts b/src/core/checkpoints/changeCard.ts new file mode 100644 index 0000000000..b13f03320d --- /dev/null +++ b/src/core/checkpoints/changeCard.ts @@ -0,0 +1,91 @@ +/** + * Per-step change card builder (B3a). + * + * The card is emitted from `checkpointSave` (index.ts) once the per-write + * checkpoint commit exists, so the payload can key the card by the real + * checkpoint ID and reuse the approval diff + stats the tool already + * computed (threaded through {@link CheckpointWriteInfo}). The card is + * informational and always emitted for write steps, including auto-approved + * ones — which always get the compact ("summary") form regardless of the + * `changeCardDetail` setting. + */ +import { DEFAULT_CHANGE_CARD_DETAIL, type ChangeCardData, type ChangeCardDetail } from "@roo-code/types" + +/** + * The write data a change card is built from. Structurally compatible with + * `CheckpointWriteInfo` (src/core/checkpoints/index.ts), minus the + * `operation` field the card does not need. + */ +export interface ChangeCardWrite { + /** The file path as the tool knows it (relative to the task cwd). */ + path: string + /** { additions, deletions } from the approval diff, when computable. */ + diffStats?: { additions: number; deletions: number } + /** The unified approval diff for this file (reused, not recomputed). */ + diff?: string + /** Whether the tool step was auto-approved (no human interaction). */ + autoApproved?: boolean +} + +/** + * Whether every write of the step was auto-approved. Empty steps are not + * auto-approved (there is nothing for the user to have skipped). + */ +export function isAutoApprovedStep(writes: readonly ChangeCardWrite[]): boolean { + return writes.length > 0 && writes.every((write) => write.autoApproved === true) +} + +/** + * Resolve the card detail level for a step: + * - auto-approved steps always get the compact "summary" card, regardless of + * the user setting (cards for steps the user never saw approving are + * informational only); + * - otherwise the `changeCardDetail` setting applies, defaulting to + * "summary" when unset. + */ +export function resolveChangeCardDetail( + writes: readonly ChangeCardWrite[], + setting: ChangeCardDetail | undefined, +): ChangeCardDetail { + if (isAutoApprovedStep(writes)) { + return "summary" + } + return setting ?? DEFAULT_CHANGE_CARD_DETAIL +} + +/** + * Build the typed change-card payload for one step (one per-write checkpoint). + * + * With `detail: "full"` each file carries its unified diff inline; with + * `detail: "summary"` the diff is omitted and the UI fetches it lazily + * (B3b). + */ +export function buildChangeCard( + checkpointId: string, + writes: readonly ChangeCardWrite[], + detail: ChangeCardDetail, +): ChangeCardData { + return { + checkpointIds: [checkpointId], + files: writes.map((write) => ({ + path: write.path, + additions: write.diffStats?.additions ?? 0, + deletions: write.diffStats?.deletions ?? 0, + ...(detail === "full" && write.diff ? { diff: write.diff } : {}), + })), + totalFiles: writes.length, + detail, + } +} + +/** + * Convenience wrapper: resolve the detail level from the step + setting, then + * build the payload. This is what `checkpointSave` calls. + */ +export function buildChangeCardPayload( + checkpointId: string, + writes: readonly ChangeCardWrite[], + setting: ChangeCardDetail | undefined, +): ChangeCardData { + return buildChangeCard(checkpointId, writes, resolveChangeCardDetail(writes, setting)) +} diff --git a/src/core/checkpoints/changeJournal.ts b/src/core/checkpoints/changeJournal.ts new file mode 100644 index 0000000000..850d2b7f1d --- /dev/null +++ b/src/core/checkpoints/changeJournal.ts @@ -0,0 +1,99 @@ +import fs from "fs/promises" +import * as path from "path" + +/** + * A single entry in the per-task change journal (changes.jsonl). + * + * One line is appended for every successful file write that goes through a + * B1 per-write checkpoint hook. WriteToFileTool and EditFileTool emit one + * entry per write; ApplyPatchTool emits one entry per file change of a fully + * successful patch — those entries all reference the single B1 checkpoint + * that the patch's post-loop hook saves for the whole patch. The task-start + * baseline never produces an entry (it is not a file write). + */ +export interface ChangeJournalEntry { + /** The file path as the tool knows it (relative to task cwd). */ + path: string + /** "create" | "update" | "delete" — derived from what the tool did. */ + operation: "create" | "update" | "delete" + /** The B1 checkpoint commit SHA for this write (from checkpointSave result). */ + checkpointId: string + /** { additions, deletions } from the approval diff; null/omit when not computable. */ + diffStats?: { additions: number; deletions: number } +} + +const JOURNAL_FILENAME = "changes.jsonl" + +/** Derive the per-task checkpoint directory from globalStorageDir and taskId. */ +function taskCheckpointDir(globalStorageDir: string, taskId: string): string { + return path.join(globalStorageDir, "tasks", taskId, "checkpoints") +} + +/** Journal file path for a given task. */ +export function journalPath(globalStorageDir: string, taskId: string): string { + return path.join(taskCheckpointDir(globalStorageDir, taskId), JOURNAL_FILENAME) +} + +/** + * Append one change-journal entry to the per-task changes.jsonl file. + * + * Uses appendFile so each write is a single syscall — minimal torn-write risk. + * Creates parent directories if they don't exist yet (e.g. first checkpoint). + */ +export async function appendChange(globalStorageDir: string, taskId: string, entry: ChangeJournalEntry): Promise { + const filePath = journalPath(globalStorageDir, taskId) + await fs.mkdir(path.dirname(filePath), { recursive: true }) + const line = JSON.stringify(entry) + "\n" + await fs.appendFile(filePath, line) +} + +/** + * Load all change-journal entries for a task. + * + * Torn-tail repair: if the final line is truncated (JSON.parse fails), it is + * silently discarded. The rest of the file is returned in order. An absent + * or empty journal returns [] — but only an ABSENT file. Any other read + * failure (permissions, I/O) is rethrown: a journal that cannot be read must + * not be indistinguishable from one that is legitimately empty. + */ +export async function loadChanges(globalStorageDir: string, taskId: string): Promise { + const filePath = journalPath(globalStorageDir, taskId) + + let content: string + try { + content = await fs.readFile(filePath, "utf8") + } catch (error) { + // A missing journal is a legitimate empty history; any other read + // failure (permissions, I/O) must propagate. Swallowing it would let + // a rollback report a no-op success without reading the history. + if ((error as NodeJS.ErrnoException)?.code === "ENOENT") { + return [] + } + throw error + } + + if (!content.trim()) { + return [] + } + + const lines = content.split("\n") + // Remove trailing empty line from a file that ends with \n. + if (lines[lines.length - 1] === "") { + lines.pop() + } + + const entries: ChangeJournalEntry[] = [] + for (let i = 0; i < lines.length; i++) { + try { + entries.push(JSON.parse(lines[i]) as ChangeJournalEntry) + } catch { + // A corrupt line before the final line (e.g. a partially flushed + // append) must not hide the valid entries after it. The final line + // is still discarded as a torn tail — `continue` at the last index + // ends the loop either way. + continue + } + } + + return entries +} diff --git a/src/core/checkpoints/index.ts b/src/core/checkpoints/index.ts index 26a137b939..d206767ad4 100644 --- a/src/core/checkpoints/index.ts +++ b/src/core/checkpoints/index.ts @@ -16,6 +16,9 @@ import { DIFF_VIEW_URI_SCHEME } from "../../integrations/editor/DiffViewProvider import { CheckpointServiceOptions, RepoPerTaskCheckpointService } from "../../services/checkpoints" +import { appendChange, ChangeJournalEntry } from "./changeJournal" +import { buildChangeCardPayload } from "./changeCard" + const WARNING_THRESHOLD_MS = 5000 function sendCheckpointInitWarn(task: Task, type?: "WAIT_TIMEOUT" | "INIT_TIMEOUT", timeout?: number) { @@ -209,7 +212,38 @@ async function checkGitInstallation( } } -export async function checkpointSave(task: Task, force = false, suppressMessage = false) { +/** + * Write metadata for the per-task change journal (B2). + * + * `path` is the file path as the tool knows it (relative to the task cwd), + * consistent with what the B1 per-write checkpoint hooks see. `diffStats` is + * the { additions, deletions } pair from the approval diff when it was + * computable; omitted otherwise. ApplyPatchTool passes one entry per file + * change of a fully successful patch (an array), all sharing the single + * checkpoint the patch's post-loop hook saves. + */ +export type CheckpointWriteInfo = { + path: string + operation: "create" | "update" | "delete" + diffStats?: { additions: number; deletions: number } + /** + * The unified approval diff for this write, reused verbatim by the B3a + * change card (never recomputed). + */ + diff?: string + /** + * Whether the tool step was auto-approved (no human interaction). Auto- + * approved steps always get the compact ("summary") change card. + */ + autoApproved?: boolean +} + +export async function checkpointSave( + task: Task, + force = false, + suppressMessage = false, + write?: CheckpointWriteInfo | CheckpointWriteInfo[], +) { const service = await getCheckpointService(task) if (!service) { @@ -221,6 +255,48 @@ export async function checkpointSave(task: Task, force = false, suppressMessage // Start the checkpoint process in the background. return service .saveCheckpoint(`Task: ${task.taskId}, Time: ${Date.now()}`, { allowEmpty: force, suppressMessage }) + .then(async (result) => { + // B2: record successful file writes in the per-task change journal. + // Only a real commit produces an entry (an empty or failed save + // resolves to undefined / rejects), and non-write checkpoint calls + // (e.g. the task-start baseline) pass no `write` value at all. + if (result?.commit && write) { + const writes = Array.isArray(write) ? write : [write] + const globalStorageDir = task.providerRef.deref()?.context.globalStorageUri.fsPath + if (globalStorageDir) { + // Append sequentially so journal lines preserve write order. A + // journal failure is logged here and never propagates to the + // checkpoint error handler (checkpoints stay enabled). + try { + for (const w of writes) { + await appendChange(globalStorageDir, task.taskId, { + path: w.path, + operation: w.operation, + checkpointId: result.commit, + ...(w.diffStats ? { diffStats: w.diffStats } : {}), + }) + } + } catch (err) { + console.error("[Task#checkpointSave] failed to append change journal entry", err) + } + } + + // B3a: emit the per-step change card now that the checkpoint commit + // exists. The card reuses the approval diff/stats the tool already + // computed and is always emitted (auto-approved steps included); + // a card failure is logged and never disables checkpoints. + try { + const state = await task.providerRef.deref()?.getState() + const card = buildChangeCardPayload(result.commit, writes, state?.changeCardDetail) + await task.say("change_card", JSON.stringify(card), undefined, undefined, undefined, undefined, { + isNonInteractive: true, + }) + } catch (err) { + console.error("[Task#checkpointSave] failed to emit change card", err) + } + } + return result + }) .catch((err) => { console.error("[Task#checkpointSave] caught unexpected error, disabling checkpoints", err) task.enableCheckpoints = false diff --git a/src/core/checkpoints/rollback.ts b/src/core/checkpoints/rollback.ts new file mode 100644 index 0000000000..508789c957 --- /dev/null +++ b/src/core/checkpoints/rollback.ts @@ -0,0 +1,277 @@ +/** + * Per-file / per-step checkpoint rollback (B3c). + * + * "Rollback" means UNDOING the change-card step: every file the step touched + * is restored to the state it had BEFORE the step ran. The restore target is + * resolved from the B2 change journal (`changes.jsonl`), whose entries record + * — in write order — the checkpoint commit each successful write produced: + * + * - a file written by an earlier step resolves to that earlier entry's + * checkpoint (the file's post-write state after the previous step, i.e. its + * pre-step state); + * - a file no earlier step wrote resolves to the task-start baseline + * (`service.baseHash`): undoing the step that created a file removes it, + * and undoing the step that deleted one brings it back; + * - `restoreLatestFile` is the forward direction: it brings a file back to + * the content of its most recent recorded write (a successful no-op when + * the task never wrote the file). + * + * A file is only rolled back from the card of its most recent step: undoing + * an older step for a file that a later step wrote again would overwrite the + * newer state, so such a rollback is rejected (a full checkpoint restore + * still reaches any older state). A journal that cannot be located or read + * fails the restore instead of masquerading as "the task wrote nothing". + * + * Restores reuse the existing shadow-git service (`getCheckpointService` → + * `RepoPerTaskCheckpointService.restoreFile`, the same instance whose + * `restoreCheckpoint` the checkpoints UI uses) — nothing is forked. Only the + * named file's working-tree content is replaced; the shadow repo's HEAD and + * the checkpoint list are untouched (unlike a full `restoreCheckpoint`). + */ +import type { Task } from "../task/Task" + +import { getCheckpointService } from "./index" +import { loadChanges, type ChangeJournalEntry } from "./changeJournal" + +export interface RollbackFileOutcome { + filePath: string + success: boolean + error?: string + /** True when `restoreLatestFile` found no recorded write: the working tree was left as-is. */ + noOp?: boolean +} + +export interface RollbackStepOutcome { + /** The step checkpoint the files were resolved against, when provided. */ + checkpointId?: string + files: RollbackFileOutcome[] +} + +const NOT_ENABLED_ERROR = "Checkpoints are not enabled for this task" +const NO_TARGET_ERROR = "No checkpoint available to restore" +const NOT_IN_STEP_ERROR = "File is not part of this step's checkpoint" +const NO_ENTRY_ERROR = "No change journal entry for this file" +const NO_JOURNAL_ERROR = "Change journal is unavailable for this task" +const NOT_LATEST_ERROR = "File was modified in a later step; roll back the latest change card first" + +type CheckpointService = NonNullable>> + +/** A readable journal (possibly legitimately empty) or the reason it could not be loaded. */ +type LoadedJournal = { entries: ChangeJournalEntry[] } | { error: string } + +// `undefined` = the journal cannot be located (provider reference gone); read failures (permissions, I/O) propagate. +async function loadTaskEntries(task: Task): Promise { + const globalStorageDir = task.providerRef.deref()?.context.globalStorageUri.fsPath + + return globalStorageDir ? loadChanges(globalStorageDir, task.taskId) : undefined +} + +// Single discriminated result for callers: a readable journal (possibly empty) or a failure. +async function loadTaskJournal(task: Task): Promise { + try { + const entries = await loadTaskEntries(task) + + if (entries === undefined) { + return { error: NO_JOURNAL_ERROR } + } + + return { entries } + } catch (error) { + const message = error instanceof Error ? error.message : String(error) + return { error: `Change journal could not be read: ${message}` } + } +} + +/** The most recent journal entry recorded for `filePath`, if any. */ +function latestEntry(entries: ChangeJournalEntry[], filePath: string): ChangeJournalEntry | undefined { + for (let i = entries.length - 1; i >= 0; i--) { + if (entries[i].path === filePath) { + return entries[i] + } + } + return undefined +} + +/** + * Resolve the checkpoint a file must be restored from in order to undo the + * step identified by `stepCheckpointId` (see the module docstring for the + * resolution rules). + */ +function preStepRestoreTarget( + entries: ChangeJournalEntry[], + filePath: string, + stepCheckpointId: string, +): { target?: string; baseline?: boolean; error?: string } { + const fileEntries = entries.filter((entry) => entry.path === filePath) + const stepIndex = fileEntries.findIndex((entry) => entry.checkpointId === stepCheckpointId) + + if (stepIndex === -1) { + return { error: NOT_IN_STEP_ERROR } + } + + // Only the file's most recent step may be rolled back: restoring an older + // state would overwrite the file's newer writes. A multi-write step shares + // one checkpoint id, so compare on the latest entry's checkpoint id. + const latest = fileEntries[fileEntries.length - 1] + + if (latest.checkpointId !== stepCheckpointId) { + return { error: NOT_LATEST_ERROR } + } + + if (stepIndex === 0) { + return { baseline: true } + } + return { target: fileEntries[stepIndex - 1].checkpointId } +} + +/** + * Run one `restoreFile` and shape the outcome. A failed restore never throws + * out of the rollback API; it is reported on the per-file outcome instead. + */ +async function performRestore( + service: CheckpointService, + target: string, + filePath: string, +): Promise { + try { + await service.restoreFile(target, filePath) + return { filePath, success: true } + } catch (error) { + const message = error instanceof Error ? error.message : String(error) + console.error(`[checkpointRollback] failed to restore ${filePath} from checkpoint ${target}: ${message}`) + return { filePath, success: false, error: message } + } +} + +/** + * Restore a single file to the state it had BEFORE the change-card step + * identified by `stepCheckpointId` — undoing that step's write to the file. + */ +export async function rollbackFile( + task: Task, + stepCheckpointId: string, + filePath: string, +): Promise { + const service = await getCheckpointService(task) + + if (!service) { + return { filePath, success: false, error: NOT_ENABLED_ERROR } + } + + const journal = await loadTaskJournal(task) + + if ("error" in journal) { + return { filePath, success: false, error: journal.error } + } + + const resolved = preStepRestoreTarget(journal.entries, filePath, stepCheckpointId) + + if (resolved.error) { + return { filePath, success: false, error: resolved.error } + } + + const target = resolved.baseline ? service.baseHash : resolved.target + + if (!target) { + return { filePath, success: false, error: NO_TARGET_ERROR } + } + + return performRestore(service, target, filePath) +} + +/** + * Restore every file of a step to the state it had before the step ran. + * + * `stepFiles` comes from the change-card payload (the B2 journal entries for + * the step's checkpoint id). Each file is resolved to its pre-step checkpoint + * through the journal (see the module docstring); without a step checkpoint id + * the latest journal entry per file is used instead (restoring to the file's + * last recorded state, the same direction as `restoreLatestFile`). + */ +export async function rollbackStep( + task: Task, + stepFiles: string[], + stepCheckpointId?: string, +): Promise { + const service = await getCheckpointService(task) + + if (!service) { + return { + checkpointId: stepCheckpointId, + files: stepFiles.map((filePath) => ({ filePath, success: false, error: NOT_ENABLED_ERROR })), + } + } + + const journal = await loadTaskJournal(task) + const files: RollbackFileOutcome[] = [] + + if ("error" in journal) { + const journalError = journal.error + return { + checkpointId: stepCheckpointId, + files: stepFiles.map((filePath) => ({ filePath, success: false, error: journalError })), + } + } + + const entries = journal.entries + + for (const filePath of stepFiles) { + if (stepCheckpointId) { + const resolved = preStepRestoreTarget(entries, filePath, stepCheckpointId) + + if (resolved.error) { + files.push({ filePath, success: false, error: resolved.error }) + continue + } + + const target = resolved.baseline ? service.baseHash : resolved.target + + if (!target) { + files.push({ filePath, success: false, error: NO_TARGET_ERROR }) + continue + } + + files.push(await performRestore(service, target, filePath)) + continue + } + + const latest = latestEntry(entries, filePath) + + if (!latest) { + files.push({ filePath, success: false, error: NO_ENTRY_ERROR }) + continue + } + + files.push(await performRestore(service, latest.checkpointId, filePath)) + } + + return { checkpointId: stepCheckpointId, files } +} + +/** + * Restore one file to the latest recorded version: the content of its most + * recent successful write checkpoint (the forward direction to a rollback). + * A file the task never wrote has no recorded version — the working tree is + * left as-is and the outcome is a successful no-op. + */ +export async function restoreLatestFile(task: Task, filePath: string): Promise { + const service = await getCheckpointService(task) + + if (!service) { + return { filePath, success: false, error: NOT_ENABLED_ERROR } + } + + const journal = await loadTaskJournal(task) + + if ("error" in journal) { + return { filePath, success: false, error: journal.error } + } + + const latest = latestEntry(journal.entries, filePath) + + if (!latest) { + return { filePath, success: true, noOp: true } + } + + return performRestore(service, latest.checkpointId, filePath) +} diff --git a/src/core/task/Task.ts b/src/core/task/Task.ts index 349d9c51d3..531fbff9ab 100644 --- a/src/core/task/Task.ts +++ b/src/core/task/Task.ts @@ -103,6 +103,7 @@ import { ToolRepetitionDetector } from "../tools/ToolRepetitionDetector" import { restoreTodoListForTask } from "../tools/UpdateTodoListTool" import { FileContextTracker } from "../context-tracking/FileContextTracker" import { RooIgnoreController } from "../ignore/RooIgnoreController" +import { ObservationRegistry } from "./observationRegistry" import { RooProtectedController } from "../protect/RooProtectedController" import { type AssistantMessageContent, presentAssistantMessage } from "../assistant-message" import { NativeToolCallParser } from "../assistant-message/NativeToolCallParser" @@ -181,6 +182,7 @@ export class Task extends EventEmitter implements TaskLike { readonly parentTask: Task | undefined = undefined readonly taskNumber: number readonly workspacePath: string + readonly observationRegistry = new ObservationRegistry() /** * The mode associated with this task. Persisted across sessions @@ -314,6 +316,9 @@ export class Task extends EventEmitter implements TaskLike { public lastMessageTs?: number private autoApprovalTimeoutRef?: NodeJS.Timeout + // B1: task-start baseline, recorded at most once (initiateTaskLoop also runs on resume). + private taskStartBaselineDone = false + // Tool Use consecutiveMistakeCount: number = 0 consecutiveMistakeLimit: number @@ -1130,6 +1135,15 @@ export class Task extends EventEmitter implements TaskLike { // - Final state is emitted when updates stop (trailing: true) this.debouncedEmitTokenUsage(tokenUsage, this.toolUsage) + // Guard: don't update the history item for abandoned tasks. Fire-and-forget + // saveClineMessages() calls can reach updateTaskHistory() after + // abandonSubtask's atomicUpdatePair() has already cleared + // parentTaskId/rootTaskId; writing this live Task's stale values would + // silently reattach the severed parent-child link. + if (this.abandoned) { + return false + } + const provider = this.providerRef.deref() const existingStatus = provider?.taskHistoryStore.get(this.taskId)?.status await provider?.updateTaskHistory(existingStatus ? { ...historyItem, status: existingStatus } : historyItem) @@ -2492,6 +2506,17 @@ export class Task extends EventEmitter implements TaskLike { // arm needed. void getCheckpointService(this) + // B1 task-start baseline: a suppressed pre-task root commit (default-on). + if (!this.taskStartBaselineDone) { + this.taskStartBaselineDone = true + const baselineEnabled = (await this.providerRef.deref()?.getState())?.perWriteCheckpoints + if (baselineEnabled !== false) { + // allowEmpty=true so a clean workspace still produces the baseline + // commit; awaited so the first per-write checkpoint cannot interleave. + await this.checkpointSave(true, true) + } + } + let nextUserContent = userContent let includeFileDetails = true diff --git a/src/core/task/__tests__/Task.spec.ts b/src/core/task/__tests__/Task.spec.ts index 37e228f887..142899c478 100644 --- a/src/core/task/__tests__/Task.spec.ts +++ b/src/core/task/__tests__/Task.spec.ts @@ -13,6 +13,7 @@ import { type GlobalState, type ProviderSettings, type ModelInfo, + type HistoryItem, type TaskLike, } from "@roo-code/types" import { TelemetryService } from "@roo-code/telemetry" @@ -27,6 +28,9 @@ import { ContextProxy } from "../../config/ContextProxy" import { processUserContentMentions } from "../../mentions/processUserContentMentions" import { MultiSearchReplaceDiffStrategy } from "../../diff/strategies/multi-search-replace" import type { ApiMessage } from "../../task-persistence" +import * as taskMetadataModule from "../../task-persistence/taskMetadata" +import * as taskMessagesModule from "../../task-persistence/taskMessages" +import { getApiMetrics } from "../../../shared/getApiMetrics" type TaskTestAccess = { getSystemPrompt: () => Promise @@ -3289,6 +3293,98 @@ describe("Cline", () => { }) }) + describe("task-start baseline (B1 perWriteCheckpoints)", () => { + it("records one suppressed baseline checkpoint per Task instance at loop start", async () => { + const task = new Task({ + provider: mockProvider, + apiConfiguration: mockApiConfig, + task: "baseline task", + startTask: false, + }) + const taskAccess = getTaskTestAccess(task) + const saveSpy = vi.spyOn(task, "checkpointSave").mockResolvedValue(undefined) + const state = await mockProvider.getState() + vi.spyOn(mockProvider, "getState").mockResolvedValue(state) + + task.abort = true + + await taskAccess.initiateTaskLoop([]) + await taskAccess.initiateTaskLoop([]) + + expect(saveSpy).toHaveBeenCalledOnce() + expect(saveSpy).toHaveBeenCalledWith(true, true) + }) + + it("records the baseline checkpoint when the setting is unset (default-on)", async () => { + const task = new Task({ + provider: mockProvider, + apiConfiguration: mockApiConfig, + task: "baseline unset task", + startTask: false, + }) + const taskAccess = getTaskTestAccess(task) + const saveSpy = vi.spyOn(task, "checkpointSave").mockResolvedValue(undefined) + const state = await mockProvider.getState() + // Unset: the property is absent from the state, so default-on applies. + const unsetState = { ...state } + Reflect.deleteProperty(unsetState, "perWriteCheckpoints") + vi.spyOn(mockProvider, "getState").mockResolvedValue(unsetState as typeof state) + + task.abort = true + + await taskAccess.initiateTaskLoop([]) + + expect(saveSpy).toHaveBeenCalledOnce() + expect(saveSpy).toHaveBeenCalledWith(true, true) + }) + + it("does not record a baseline checkpoint when perWriteCheckpoints is disabled", async () => { + const task = new Task({ + provider: mockProvider, + apiConfiguration: mockApiConfig, + task: "baseline disabled task", + startTask: false, + }) + const taskAccess = getTaskTestAccess(task) + const saveSpy = vi.spyOn(task, "checkpointSave").mockResolvedValue(undefined) + const state = await mockProvider.getState() + vi.spyOn(mockProvider, "getState").mockResolvedValue({ ...state, perWriteCheckpoints: false }) + + task.abort = true + + await taskAccess.initiateTaskLoop([]) + + expect(saveSpy).not.toHaveBeenCalled() + }) + + it("awaits the baseline checkpoint before entering the request loop", async () => { + const task = new Task({ + provider: mockProvider, + apiConfiguration: mockApiConfig, + task: "baseline await task", + startTask: false, + }) + const taskAccess = getTaskTestAccess(task) + type SaveResult = Awaited> + let resolveSave: (value: SaveResult | PromiseLike) => void = () => {} + const saveSpy = vi + .spyOn(task, "checkpointSave") + .mockImplementation(() => new Promise((resolve) => (resolveSave = resolve))) + const requestSpy = vi.spyOn(task, "recursivelyMakeClineRequests").mockResolvedValue(true) + vi.spyOn(mockProvider, "getState").mockResolvedValue({ ...(await mockProvider.getState()) }) + const loopPromise = taskAccess.initiateTaskLoop([]) + + // The loop must not enter while the baseline checkpoint is still in flight. + await new Promise((resolve) => setTimeout(resolve, 0)) + expect(saveSpy).toHaveBeenCalledOnce() + expect(requestSpy).not.toHaveBeenCalled() + + resolveSave() + await loopPromise + expect(requestSpy).toHaveBeenCalled() + }) + }) + describe("start()", () => { it("should be a no-op if the task was already started in the constructor", () => { const task = new Task({ @@ -4297,3 +4393,76 @@ describe("pushToolResultToUserContent", () => { expect(task.userMessageContent[2]).toEqual(toolResult) }) }) + +describe("saveClineMessages abandoned guard (#1021)", () => { + beforeEach(() => { + if (!TelemetryService.hasInstance()) { + TelemetryService.createInstance([]) + } + }) + + it("persists messages but does not update task history when the task was abandoned", async () => { + // The history item carries the stale link: a fire-and-forget save that + // reaches updateTaskHistory() after abandonSubtask's atomicUpdatePair() + // cleared parentTaskId/rootTaskId would silently reattach the severed + // parent-child link. + const staleHistoryItem: HistoryItem = { + id: "orphan-subtask", + number: 7, + ts: Date.now(), + task: "orphan subtask", + tokensIn: 0, + tokensOut: 0, + totalCost: 0, + rootTaskId: "stale-root", + parentTaskId: "stale-parent", + } + const saveSpy = vi.spyOn(taskMessagesModule, "saveTaskMessages").mockResolvedValue(undefined) + const metaSpy = vi + .spyOn(taskMetadataModule, "taskMetadata") + .mockResolvedValue({ historyItem: staleHistoryItem, tokenUsage: getApiMetrics([]) }) + + try { + const mockProvider = { + context: { + globalStorageUri: { fsPath: "/test/storage" }, + globalState: { + get: vi.fn().mockImplementation(() => undefined), + update: vi.fn().mockResolvedValue(undefined), + keys: vi.fn().mockReturnValue([]), + }, + }, + getState: vi.fn().mockResolvedValue({ + apiConfiguration: { apiProvider: providerIdentifiers.anthropic, apiKey: "test-key" }, + mcpEnabled: false, + }), + getMcpHub: vi.fn().mockReturnValue(undefined), + postMessageToWebview: vi.fn().mockResolvedValue(undefined), + updateTaskHistory: vi.fn().mockResolvedValue(undefined), + // Task receives a full ClineProvider at runtime; this focused unit test only + // exercises these methods, so the partial double is cast (same pattern as the + // "Subtask Rate Limiting" block above). + } as unknown as MockedClineProvider + + const task = new Task({ + provider: mockProvider, + apiConfiguration: { apiProvider: providerIdentifiers.anthropic, apiKey: "test-key" }, + task: "orphan subtask", + startTask: false, + }) + + // abandonSubtask severs the link, then aborts the subtask with + // isAbandoned=true; an in-flight fire-and-forget save lands here. + task.abandoned = true + + const saved = await getTaskTestAccess(task).saveClineMessages() + + expect(saved).toBe(false) + expect(saveSpy).toHaveBeenCalledTimes(1) // messages are still persisted + expect(mockProvider.updateTaskHistory).not.toHaveBeenCalled() // history link is not reattached + } finally { + saveSpy.mockRestore() + metaSpy.mockRestore() + } + }) +}) diff --git a/src/core/task/__tests__/observationRegistry.spec.ts b/src/core/task/__tests__/observationRegistry.spec.ts new file mode 100644 index 0000000000..51b73aabde --- /dev/null +++ b/src/core/task/__tests__/observationRegistry.spec.ts @@ -0,0 +1,72 @@ +import { describe, it, expect, vi } from "vitest" + +import { ObservationRegistry } from "../observationRegistry" + +describe("ObservationRegistry", () => { + it("observe → get returns the recorded version and observedAt", () => { + const reg = new ObservationRegistry() + reg.observe("/a/b/c.ts", "1:2:300:4000000000:5000000000") + + const obs = reg.get("/a/b/c.ts") + expect(obs).toBeDefined() + expect(obs!.version).toBe("1:2:300:4000000000:5000000000") + expect(typeof obs!.observedAt).toBe("number") + }) + + it("re-observe replaces the entry with a fresh observedAt", () => { + vi.useFakeTimers() + const reg = new ObservationRegistry() + reg.observe("/a/b/c.ts", "v1") + const first = reg.get("/a/b/c.ts")! + expect(first.version).toBe("v1") + + vi.advanceTimersByTime(50) + reg.observe("/a/b/c.ts", "v2") + const second = reg.get("/a/b/c.ts")! + expect(second.version).toBe("v2") + expect(second.observedAt).toBeGreaterThan(first.observedAt) + + vi.useRealTimers() + }) + + it("has returns true for observed paths, false otherwise", () => { + const reg = new ObservationRegistry() + reg.observe("/x.ts", "t1") + expect(reg.has("/x.ts")).toBe(true) + expect(reg.has("/y.ts")).toBe(false) + }) + + it("size reflects the number of observed entries", () => { + const reg = new ObservationRegistry() + expect(reg.size).toBe(0) + reg.observe("/a.ts", "t1") + reg.observe("/b.ts", "t2") + expect(reg.size).toBe(2) + }) + + it("clear removes all entries and resets size to 0", () => { + const reg = new ObservationRegistry() + reg.observe("/a.ts", "t1") + reg.observe("/b.ts", "t2") + reg.clear() + expect(reg.size).toBe(0) + expect(reg.get("/a.ts")).toBeUndefined() + expect(reg.has("/b.ts")).toBe(false) + }) + + it("get on empty registry returns undefined", () => { + const reg = new ObservationRegistry() + expect(reg.get("/any.ts")).toBeUndefined() + }) + + it("separate instances are independent — observing in one does not appear in the other", () => { + const regA = new ObservationRegistry() + const regB = new ObservationRegistry() + regA.observe("/shared.ts", "v1") + expect(regA.get("/shared.ts")).toBeDefined() + expect(regB.get("/shared.ts")).toBeUndefined() + regB.observe("/shared.ts", "v2") + expect(regA.get("/shared.ts")!.version).toBe("v1") + expect(regB.get("/shared.ts")!.version).toBe("v2") + }) +}) diff --git a/src/core/task/observationRegistry.ts b/src/core/task/observationRegistry.ts new file mode 100644 index 0000000000..871f80225b --- /dev/null +++ b/src/core/task/observationRegistry.ts @@ -0,0 +1,47 @@ +/** + * Per-task file observation registry (upstream epic #1375, phase A2). + * + * Each Task owns its own instance so parent and subtask observations are + * independent. The S4 guarded-write will compare these versions against the + * token recomputed pre-write to detect stale reads or file replacement. + * + * Pure in-memory — zero I/O, no dependencies. No behavior change in this PR: + * observations are recorded but not consulted. + */ + +export interface FileObservation { + /** Version token derived from on-disk fs.stat (bigint mode). */ + version: string + /** Millisecond timestamp when the observation was recorded. */ + observedAt: number +} + +export class ObservationRegistry { + private readonly entries = new Map() + + /** + * Record an observation for a file at its absolute path. + * + * Re-observing replaces the entry with a fresh observedAt timestamp and + * the new version token. + */ + observe(absolutePath: string, version: string): void { + this.entries.set(absolutePath, { version, observedAt: Date.now() }) + } + + get(absolutePath: string): FileObservation | undefined { + return this.entries.get(absolutePath) + } + + has(absolutePath: string): boolean { + return this.entries.has(absolutePath) + } + + clear(): void { + this.entries.clear() + } + + get size(): number { + return this.entries.size + } +} diff --git a/src/core/tools/ApplyDiffTool.ts b/src/core/tools/ApplyDiffTool.ts index 3b664b3bd2..9773c3b922 100644 --- a/src/core/tools/ApplyDiffTool.ts +++ b/src/core/tools/ApplyDiffTool.ts @@ -13,6 +13,9 @@ import { unescapeHtmlEntities } from "../../utils/text-normalization" import { EXPERIMENT_IDS, experiments } from "../../shared/experiments" import { computeDiffStats, sanitizeUnifiedDiff } from "../diff/stats" import type { ToolUse } from "../../shared/tools" +import { versionTokenOfStat } from "../../utils/versionToken" +import { checkAutoApproval } from "../auto-approval" +import { checkpointSave } from "../checkpoints" import { BaseTool, ToolCallbacks } from "./BaseTool" @@ -68,7 +71,21 @@ export class ApplyDiffTool extends BaseTool<"apply_diff"> { return } + // S4b (trial addendum): apply_diff reads the file itself to apply the patch. + // Record that read as an S2 observation under the ReadFileTool contract + // (observed only when the pre- and post-read tokens agree) so the guarded + // edit publish in the prevent-focus-disruption branch can compare-and-swap + // against the token this tool observed, instead of being rejected as an + // unobserved edit. + const preReadStats = await fs.stat(absolutePath, { bigint: true }).catch(() => undefined) const originalContent: string = await fs.readFile(absolutePath, "utf-8") + const postReadStats = await fs.stat(absolutePath, { bigint: true }).catch(() => undefined) + if (preReadStats && postReadStats) { + const preReadToken = versionTokenOfStat(preReadStats) + if (preReadToken === versionTokenOfStat(postReadStats)) { + task.observationRegistry.observe(absolutePath, preReadToken) + } + } // Apply the diff to the original content const diffResult = (await task.diffStrategy?.applyDiff( @@ -144,9 +161,13 @@ export class ApplyDiffTool extends BaseTool<"apply_diff"> { diff: diffContent, } + // Hoisted: both save branches build the same approval message, and the + // B3a per-write checkpoint (below) needs it for auto-approval parity. + let completeMessage = "" + if (isPreventFocusDisruptionEnabled) { // Direct file write without diff view - const completeMessage = JSON.stringify({ + completeMessage = JSON.stringify({ ...sharedMessageProps, diff: diffContent, content: unifiedPatch, @@ -173,7 +194,8 @@ export class ApplyDiffTool extends BaseTool<"apply_diff"> { return } - // Save directly without showing diff view or opening the file + // Save directly without showing diff view or opening the file. The diff is + // applied to an existing file, so edit-guard semantics require a prior read. task.diffViewProvider.editType = "modify" task.diffViewProvider.originalContent = originalContent await task.diffViewProvider.saveDirectly( @@ -182,6 +204,7 @@ export class ApplyDiffTool extends BaseTool<"apply_diff"> { false, diagnosticsEnabled, writeDelayMs, + "edit", ) } else { // Original behavior with diff view @@ -191,7 +214,7 @@ export class ApplyDiffTool extends BaseTool<"apply_diff"> { await task.diffViewProvider.update(diffResult.content, true) task.diffViewProvider.scrollToFirstDiff() - const completeMessage = JSON.stringify({ + completeMessage = JSON.stringify({ ...sharedMessageProps, diff: diffContent, content: unifiedPatch, @@ -224,6 +247,36 @@ export class ApplyDiffTool extends BaseTool<"apply_diff"> { await task.diffViewProvider.saveChanges(diagnosticsEnabled, writeDelayMs) } + // B3a: per-write checkpoint + change card for the apply_diff write, with + // the same parity as write_to_file / edit_file / apply_patch (this tool + // previously wrote with no checkpoint and no card, so its edits had no + // per-file rollback surface in chat). The applied unified diff and its + // stats were already computed above for the tool message and are reused + // verbatim; auto-approved steps always get the compact card. Live setting + // with default-on semantics: skip only when explicitly false. apply_diff + // only ever modifies an existing file (a missing file errors out before + // the diff is applied), so the operation is always "update". + const perWriteCheckpoints = state?.perWriteCheckpoints + if (perWriteCheckpoints !== false) { + const autoApproved = + ( + await checkAutoApproval({ + state, + cwd: task.cwd, + ask: "tool", + text: completeMessage, + isProtected: isWriteProtected, + }) + ).decision === "approve" + await checkpointSave(task, false, true, { + path: relPath, + operation: "update", + ...(diffStats ? { diffStats: { additions: diffStats.added, deletions: diffStats.removed } } : {}), + ...(unifiedPatch ? { diff: unifiedPatch } : {}), + ...(autoApproved ? { autoApproved: true } : {}), + }).catch(() => {}) + } + // Track file edit operation if (relPath) { await task.fileContextTracker.trackFileContext(relPath, "roo_edited" as RecordSource) diff --git a/src/core/tools/ApplyPatchTool.ts b/src/core/tools/ApplyPatchTool.ts index 56b2bf8909..bf6413de3c 100644 --- a/src/core/tools/ApplyPatchTool.ts +++ b/src/core/tools/ApplyPatchTool.ts @@ -6,11 +6,14 @@ import { type ClineSayTool, DEFAULT_WRITE_DELAY_MS } from "@roo-code/types" import { getReadablePath } from "../../utils/path" import { isPathOutsideWorkspace } from "../../utils/pathUtils" import { Task } from "../task/Task" +import { checkpointSave } from "../checkpoints" +import { checkAutoApproval } from "../auto-approval" import { formatResponse } from "../prompts/responses" import { RecordSource } from "../context-tracking/FileContextTrackerTypes" import { fileExistsAtPath } from "../../utils/fs" import { EXPERIMENT_IDS, experiments } from "../../shared/experiments" import { sanitizeUnifiedDiff, computeDiffStats } from "../diff/stats" +import { versionTokenOfStat } from "../../utils/versionToken" import { BaseTool, ToolCallbacks } from "./BaseTool" import type { ToolUse } from "../../shared/tools" import { parsePatch, ParseError, processAllHunks } from "./apply-patch" @@ -20,6 +23,18 @@ interface ApplyPatchParams { patch: string } +/** + * B2: result of a single file operation within a patch. `succeeded` controls + * the whole-patch success state (and therefore the per-patch checkpoint), + * while `wrote` records whether the operation actually wrote a file — a no-op + * update must not produce a change-journal entry for a file that was never + * written. + */ +interface ApplyPatchFileOpResult { + succeeded: boolean + wrote: boolean +} + export class ApplyPatchTool extends BaseTool<"apply_patch"> { readonly name = "apply_patch" as const @@ -85,10 +100,25 @@ export class ApplyPatchTool extends BaseTool<"apply_patch"> { return } - // Process each hunk + // Process each hunk. The read doubles as the S2 observation for the + // guarded publish (ReadFileTool contract: stat before and after the + // read, observe only when the on-disk version is unchanged between the + // two stats). Without it, the in-place modify publish is an unobserved + // write and the composed chat-diff default rejects it ("File already + // exists ... and was not read before this write") even though this tool + // just read the exact content the patch was applied to. const readFile = async (filePath: string): Promise => { const absolutePath = path.resolve(task.cwd, filePath) - return await fs.readFile(absolutePath, "utf8") + const preReadStats = await fs.stat(absolutePath, { bigint: true }).catch(() => undefined) + const content: string = await fs.readFile(absolutePath, "utf8") + const postReadStats = await fs.stat(absolutePath, { bigint: true }).catch(() => undefined) + if (preReadStats && postReadStats) { + const preReadToken = versionTokenOfStat(preReadStats) + if (preReadToken === versionTokenOfStat(postReadStats)) { + task.observationRegistry.observe(absolutePath, preReadToken) + } + } + return content } let changes: ApplyPatchFileChange[] @@ -102,7 +132,13 @@ export class ApplyPatchTool extends BaseTool<"apply_patch"> { return } - // Process each file change + // Process each file change. The handlers report whether their file + // operation succeeded (which controls the whole-patch checkpoint) and + // whether it actually wrote a file (which controls the change journal + // — a no-op update must not be journaled). A rejected approval or a + // failed local write never gets checkpointed as a success. + let patchSucceeded = true + const successfulChanges: ApplyPatchFileChange[] = [] for (const change of changes) { const relPath = change.path const absolutePath = path.resolve(task.cwd, relPath) @@ -112,7 +148,12 @@ export class ApplyPatchTool extends BaseTool<"apply_patch"> { if (!accessAllowed) { await task.say("rooignore_error", relPath) pushToolResult(formatResponse.rooIgnoreError(relPath)) - return + // B2 partial flush: break, not return - an earlier hunk may have + // already written a file, and those writes must still receive the + // checkpoint, journal entry, and change card. Failing the patch + // also keeps the consecutive-mistake counter from resetting. + patchSucceeded = false + break } // Check if file is write-protected @@ -120,17 +161,100 @@ export class ApplyPatchTool extends BaseTool<"apply_patch"> { if (change.type === "add") { // Create new file - await this.handleAddFile(change, absolutePath, relPath, task, callbacks, isWriteProtected) + const addResult = await this.handleAddFile( + change, + absolutePath, + relPath, + task, + callbacks, + isWriteProtected, + ) + patchSucceeded = addResult.succeeded && patchSucceeded + if (addResult.wrote) { + successfulChanges.push(change) + } } else if (change.type === "delete") { // Delete file - await this.handleDeleteFile(absolutePath, relPath, task, callbacks, isWriteProtected) + const deleteResult = await this.handleDeleteFile( + change, + absolutePath, + relPath, + task, + callbacks, + isWriteProtected, + ) + patchSucceeded = deleteResult.succeeded && patchSucceeded + if (deleteResult.wrote) { + successfulChanges.push(change) + } } else if (change.type === "update") { - // Update file - await this.handleUpdateFile(change, absolutePath, relPath, task, callbacks, isWriteProtected) + // Update file (a no-op update succeeds without writing) + const updateResult = await this.handleUpdateFile( + change, + absolutePath, + relPath, + task, + callbacks, + isWriteProtected, + ) + patchSucceeded = updateResult.succeeded && patchSucceeded + if (updateResult.wrote) { + successfulChanges.push(change) + } } } - task.consecutiveMistakeCount = 0 + // Reset the consecutive-mistake counter only after a fully successful + // patch: a failed operation (missing file, rejected move, ...) increments + // the counter, and the count must survive a partially written patch so + // the auto-approval safety net still engages across consecutive failed + // patches. + if (patchSucceeded) { + task.consecutiveMistakeCount = 0 + } + + // B1: one checkpoint for the whole patch (not per file). Live + // setting with default-on semantics: skip only when explicitly false. + // B3a partial flush: the checkpoint and journal are also taken when at + // least one file operation wrote, even if a later hunk of the same + // patch failed - the journal then documents exactly the subset that + // was written, and the failed operation was already reported through + // pushToolResult. A fully failed patch (nothing written) leaves no + // checkpoint behind. + if (patchSucceeded || successfulChanges.length > 0) { + const perWriteCheckpoints = (await task.providerRef?.deref()?.getState())?.perWriteCheckpoints + if (perWriteCheckpoints !== false) { + // B2: one journal entry per file that was actually written by + // the patch (the simplest correct design for multi-file patches), + // all referencing the single checkpoint above. A no-op update + // contributes no entry because nothing was written. `movePath`, + // when present, is the file's final location. B3a: the per-file + // approval diff/stats and auto-approval state, retained by the + // handlers, feed the per-step change card. + // Awaited: a later write must not interleave with this patch's + // staging/commit/journal/change-card work. checkpointSave never + // rejects (service call wrapped in try/catch upstream). + await checkpointSave( + task, + false, + true, + successfulChanges.map((change) => ({ + path: change.movePath ?? change.path, + operation: change.type === "add" ? "create" : change.type, + ...(change.diffStats + ? { + diffStats: { + additions: change.diffStats.added, + deletions: change.diffStats.removed, + }, + } + : {}), + ...(change.diff ? { diff: change.diff } : {}), + ...(change.autoApproved ? { autoApproved: true } : {}), + })), + ) + } + } } catch (error) { await handleError("apply patch", error as Error) await task.diffViewProvider.reset() @@ -144,7 +268,7 @@ export class ApplyPatchTool extends BaseTool<"apply_patch"> { task: Task, callbacks: ToolCallbacks, isWriteProtected: boolean, - ): Promise { + ): Promise { const { askApproval, pushToolResult } = callbacks // Check if file already exists @@ -155,7 +279,7 @@ export class ApplyPatchTool extends BaseTool<"apply_patch"> { const errorMessage = `File already exists: ${relPath}. Use Update File instead.` await task.say("error", errorMessage) pushToolResult(formatResponse.toolError(errorMessage)) - return + return { succeeded: false, wrote: false } } const newContent = change.newContent || "" @@ -194,6 +318,21 @@ export class ApplyPatchTool extends BaseTool<"apply_patch"> { diffStats, } satisfies ClineSayTool) + // B3a: retain the approval diff/stats and auto-approval state so the + // post-loop checkpoint hook can build the per-step change card. + change.diff = sanitizedDiff + change.diffStats = diffStats + change.autoApproved = + ( + await checkAutoApproval({ + state, + cwd: task.cwd, + ask: "tool", + text: completeMessage, + isProtected: isWriteProtected, + }) + ).decision === "approve" + // Show diff view if focus disruption prevention is disabled if (!isPreventFocusDisruptionEnabled) { await task.diffViewProvider.open(relPath) @@ -209,12 +348,21 @@ export class ApplyPatchTool extends BaseTool<"apply_patch"> { } pushToolResult("Changes were rejected by the user.") await task.diffViewProvider.reset() - return + return { succeeded: false, wrote: false } } // Save the changes if (isPreventFocusDisruptionEnabled) { - await task.diffViewProvider.saveDirectly(relPath, newContent, true, diagnosticsEnabled, writeDelayMs) + // Guarded publish: the patch supplies the complete new content, so create-guard + // semantics apply (an unobserved existing target is rejected, not overwritten). + await task.diffViewProvider.saveDirectly( + relPath, + newContent, + true, + diagnosticsEnabled, + writeDelayMs, + "create", + ) } else { await task.diffViewProvider.saveChanges(diagnosticsEnabled, writeDelayMs) } @@ -227,15 +375,17 @@ export class ApplyPatchTool extends BaseTool<"apply_patch"> { pushToolResult(message) await task.diffViewProvider.reset() task.processQueuedMessages() + return { succeeded: true, wrote: true } } private async handleDeleteFile( + change: ApplyPatchFileChange, absolutePath: string, relPath: string, task: Task, callbacks: ToolCallbacks, isWriteProtected: boolean, - ): Promise { + ): Promise { const { askApproval, pushToolResult } = callbacks // Check if file exists @@ -246,7 +396,7 @@ export class ApplyPatchTool extends BaseTool<"apply_patch"> { const errorMessage = `File not found: ${relPath}. Cannot delete a non-existent file.` await task.say("error", errorMessage) pushToolResult(formatResponse.toolError(errorMessage)) - return + return { succeeded: false, wrote: false } } const isOutsideWorkspace = isPathOutsideWorkspace(absolutePath) @@ -264,11 +414,24 @@ export class ApplyPatchTool extends BaseTool<"apply_patch"> { isProtected: isWriteProtected, } satisfies ClineSayTool) + // B3a: auto-approval state feeds the per-step change card (deletes have + // no diff to thread). + change.autoApproved = + ( + await checkAutoApproval({ + state: await task.providerRef.deref()?.getState(), + cwd: task.cwd, + ask: "tool", + text: completeMessage, + isProtected: isWriteProtected, + }) + ).decision === "approve" + const didApprove = await askApproval("tool", completeMessage, undefined, isWriteProtected) if (!didApprove) { pushToolResult("Delete operation was rejected by the user.") - return + return { succeeded: false, wrote: false } } // Delete the file @@ -278,12 +441,13 @@ export class ApplyPatchTool extends BaseTool<"apply_patch"> { const errorMessage = `Failed to delete file '${relPath}': ${error instanceof Error ? error.message : String(error)}` await task.say("error", errorMessage) pushToolResult(formatResponse.toolError(errorMessage)) - return + return { succeeded: false, wrote: false } } task.didEditFile = true pushToolResult(`Successfully deleted ${relPath}`) task.processQueuedMessages() + return { succeeded: true, wrote: true } } private async handleUpdateFile( @@ -293,9 +457,13 @@ export class ApplyPatchTool extends BaseTool<"apply_patch"> { task: Task, callbacks: ToolCallbacks, isWriteProtected: boolean, - ): Promise { + ): Promise { const { askApproval, pushToolResult } = callbacks + // A move reports failure when the original file cannot be deleted + // after the copy (both paths would remain on disk). + let moveSucceeded = true + // Check if file exists const fileExists = await fileExistsAtPath(absolutePath) if (!fileExists) { @@ -304,7 +472,7 @@ export class ApplyPatchTool extends BaseTool<"apply_patch"> { const errorMessage = `File not found: ${relPath}. Cannot update a non-existent file.` await task.say("error", errorMessage) pushToolResult(formatResponse.toolError(errorMessage)) - return + return { succeeded: false, wrote: false } } const originalContent = change.originalContent || "" @@ -318,9 +486,13 @@ export class ApplyPatchTool extends BaseTool<"apply_patch"> { // Generate and validate diff const diff = formatResponse.createPrettyPatch(relPath, originalContent, newContent) if (!diff) { + // A no-op change is not a failure: the patch processed cleanly and + // nothing was written, so the whole-patch success state is kept — + // but `wrote` stays false so the change journal does not document a + // write that never happened. pushToolResult(`No changes needed for '${relPath}'`) await task.diffViewProvider.reset() - return + return { succeeded: true, wrote: false } } // Check experiment settings @@ -351,6 +523,21 @@ export class ApplyPatchTool extends BaseTool<"apply_patch"> { diffStats, } satisfies ClineSayTool) + // B3a: retain the approval diff/stats and auto-approval state so the + // post-loop checkpoint hook can build the per-step change card. + change.diff = sanitizedDiff + change.diffStats = diffStats + change.autoApproved = + ( + await checkAutoApproval({ + state, + cwd: task.cwd, + ask: "tool", + text: completeMessage, + isProtected: isWriteProtected, + }) + ).decision === "approve" + // Show diff view if focus disruption prevention is disabled if (!isPreventFocusDisruptionEnabled) { await task.diffViewProvider.open(relPath) @@ -366,7 +553,7 @@ export class ApplyPatchTool extends BaseTool<"apply_patch"> { } pushToolResult("Changes were rejected by the user.") await task.diffViewProvider.reset() - return + return { succeeded: false, wrote: false } } // Handle file move if specified @@ -379,7 +566,7 @@ export class ApplyPatchTool extends BaseTool<"apply_patch"> { await task.say("rooignore_error", change.movePath) pushToolResult(formatResponse.rooIgnoreError(change.movePath)) await task.diffViewProvider.reset() - return + return { succeeded: false, wrote: false } } // Check if destination path is write-protected @@ -391,7 +578,7 @@ export class ApplyPatchTool extends BaseTool<"apply_patch"> { await task.say("error", errorMessage) pushToolResult(formatResponse.toolError(errorMessage)) await task.diffViewProvider.reset() - return + return { succeeded: false, wrote: false } } // Check if destination path is outside workspace @@ -403,17 +590,19 @@ export class ApplyPatchTool extends BaseTool<"apply_patch"> { await task.say("error", errorMessage) pushToolResult(formatResponse.toolError(errorMessage)) await task.diffViewProvider.reset() - return + return { succeeded: false, wrote: false } } // Save new content to the new path if (isPreventFocusDisruptionEnabled) { + // The move destination is published with the complete new content. await task.diffViewProvider.saveDirectly( change.movePath, newContent, false, diagnosticsEnabled, writeDelayMs, + "create", ) } else { // Write to new path and delete old file @@ -422,18 +611,35 @@ export class ApplyPatchTool extends BaseTool<"apply_patch"> { await fs.writeFile(moveAbsolutePath, newContent, "utf8") } - // Delete the original file + // Delete the original file. A failed deletion leaves both paths on + // disk, so the move must be reported as a failure rather than + // checkpointed and journaled as a completed move. try { await fs.unlink(absolutePath) } catch (error) { + moveSucceeded = false console.error(`Failed to delete original file after move: ${error}`) + task.consecutiveMistakeCount++ + task.recordToolError("apply_patch") + const errorMessage = `Move of '${relPath}' to '${change.movePath}' failed: could not delete the original file.` + await task.say("error", errorMessage) + pushToolResult(formatResponse.toolError(errorMessage)) } await task.fileContextTracker.trackFileContext(change.movePath, "roo_edited" as RecordSource) } else { // Save changes to the same file if (isPreventFocusDisruptionEnabled) { - await task.diffViewProvider.saveDirectly(relPath, newContent, false, diagnosticsEnabled, writeDelayMs) + // Guarded publish: the patched file content is complete, so create-guard + // semantics apply (stale observed versions are rejected with a re-read hint). + await task.diffViewProvider.saveDirectly( + relPath, + newContent, + false, + diagnosticsEnabled, + writeDelayMs, + "create", + ) } else { await task.diffViewProvider.saveChanges(diagnosticsEnabled, writeDelayMs) } @@ -447,6 +653,13 @@ export class ApplyPatchTool extends BaseTool<"apply_patch"> { pushToolResult(message) await task.diffViewProvider.reset() task.processQueuedMessages() + if (!moveSucceeded) { + // The destination file was written on disk before the source + // deletion failed, so the write must still be checkpointed and + // journaled; the move itself is reported as failed. + return { succeeded: false, wrote: true } + } + return { succeeded: true, wrote: true } } override async handlePartial(task: Task, block: ToolUse<"apply_patch">): Promise { diff --git a/src/core/tools/EditFileTool.ts b/src/core/tools/EditFileTool.ts index a7301e2ac9..0e34308e16 100644 --- a/src/core/tools/EditFileTool.ts +++ b/src/core/tools/EditFileTool.ts @@ -11,6 +11,8 @@ import { RecordSource } from "../context-tracking/FileContextTrackerTypes" import { fileExistsAtPath } from "../../utils/fs" import { EXPERIMENT_IDS, experiments } from "../../shared/experiments" import { sanitizeUnifiedDiff, computeDiffStats } from "../diff/stats" +import { checkpointSave } from "../../core/checkpoints" +import { checkAutoApproval } from "../auto-approval" import type { ToolUse } from "../../shared/tools" import { BaseTool, ToolCallbacks } from "./BaseTool" @@ -392,6 +394,7 @@ export class EditFileTool extends BaseTool<"edit_file"> { const state = await provider?.getState() const diagnosticsEnabled = state?.diagnosticsEnabled ?? true const writeDelayMs = state?.writeDelayMs ?? DEFAULT_WRITE_DELAY_MS + const perWriteCheckpoints = state?.perWriteCheckpoints ?? true const isPreventFocusDisruptionEnabled = experiments.isEnabled( state?.experiments ?? {}, EXPERIMENT_IDS.PREVENT_FOCUS_DISRUPTION, @@ -436,13 +439,16 @@ export class EditFileTool extends BaseTool<"edit_file"> { // Save the changes if (isPreventFocusDisruptionEnabled) { - // Direct file write without diff view or opening the file + // Direct file write without diff view or opening the file. In-place edits + // use edit-guard semantics (a prior read is required); new-file creation + // keeps create-guard semantics. await task.diffViewProvider.saveDirectly( relPath, newContent, isNewFile, diagnosticsEnabled, writeDelayMs, + isNewFile ? "create" : "edit", ) } else { // Call saveChanges to update the DiffViewProvider properties @@ -463,6 +469,34 @@ export class EditFileTool extends BaseTool<"edit_file"> { pushToolResult(message + replacementInfo) + if (perWriteCheckpoints) { + // B2: the change-journal entry for this edit is appended inside + // checkpointSave (the hook stays a single call site), keyed by the + // checkpoint commit that call produces. B3a threads the approval + // diff (for the change card) and whether the step was auto- + // approved (auto-approved steps always get the compact card). + const autoApproved = + ( + await checkAutoApproval({ + state, + cwd: task.cwd, + ask: "tool", + text: completeMessage, + isProtected: isWriteProtected, + }) + ).decision === "approve" + // Awaited: a later tool block must not interleave with this edit's + // staging/commit/journal/change-card work. checkpointSave never + // rejects (service call wrapped in try/catch upstream). + await checkpointSave(task, false, true, { + path: relPath, + operation: isNewFile ? "create" : "update", + diffStats: diffStats ? { additions: diffStats.added, deletions: diffStats.removed } : undefined, + ...(sanitizedDiff ? { diff: sanitizedDiff } : {}), + ...(autoApproved ? { autoApproved: true } : {}), + }) + } + await task.diffViewProvider.reset() this.resetPartialState() diff --git a/src/core/tools/EditTool.ts b/src/core/tools/EditTool.ts index 2ae8bf4ed0..561225c79a 100644 --- a/src/core/tools/EditTool.ts +++ b/src/core/tools/EditTool.ts @@ -11,6 +11,8 @@ import { RecordSource } from "../context-tracking/FileContextTrackerTypes" import { fileExistsAtPath } from "../../utils/fs" import { EXPERIMENT_IDS, experiments } from "../../shared/experiments" import { sanitizeUnifiedDiff, computeDiffStats } from "../diff/stats" +import { checkpointSave } from "../../core/checkpoints" +import { checkAutoApproval } from "../auto-approval" import type { ToolUse } from "../../shared/tools" import { BaseTool, ToolCallbacks } from "./BaseTool" @@ -167,6 +169,7 @@ export class EditTool extends BaseTool<"edit"> { const state = await provider?.getState() const diagnosticsEnabled = state?.diagnosticsEnabled ?? true const writeDelayMs = state?.writeDelayMs ?? DEFAULT_WRITE_DELAY_MS + const perWriteCheckpoints = state?.perWriteCheckpoints ?? true const isPreventFocusDisruptionEnabled = experiments.isEnabled( state?.experiments ?? {}, EXPERIMENT_IDS.PREVENT_FOCUS_DISRUPTION, @@ -211,8 +214,16 @@ export class EditTool extends BaseTool<"edit"> { // Save the changes if (isPreventFocusDisruptionEnabled) { - // Direct file write without diff view or opening the file - await task.diffViewProvider.saveDirectly(relPath, newContent, false, diagnosticsEnabled, writeDelayMs) + // Direct file write without diff view or opening the file. This tool only + // edits existing files, so edit-guard semantics require a prior read. + await task.diffViewProvider.saveDirectly( + relPath, + newContent, + false, + diagnosticsEnabled, + writeDelayMs, + "edit", + ) } else { // Call saveChanges to update the DiffViewProvider properties await task.diffViewProvider.saveChanges(diagnosticsEnabled, writeDelayMs) @@ -229,6 +240,34 @@ export class EditTool extends BaseTool<"edit"> { const message = await task.diffViewProvider.pushToolWriteResult(task, task.cwd, false) pushToolResult(message) + if (perWriteCheckpoints) { + // B2: the change-journal entry for this edit is appended inside + // checkpointSave (the hook stays a single call site), keyed by the + // checkpoint commit that call produces. B3a threads the approval + // diff (for the change card) and whether the step was auto- + // approved (auto-approved steps always get the compact card). + const autoApproved = + ( + await checkAutoApproval({ + state, + cwd: task.cwd, + ask: "tool", + text: completeMessage, + isProtected: isWriteProtected, + }) + ).decision === "approve" + // Awaited: a later tool block must not interleave with this edit's + // staging/commit/journal/change-card work. checkpointSave never + // rejects (service call wrapped in try/catch upstream). + await checkpointSave(task, false, true, { + path: relPath, + operation: "update", + diffStats: diffStats ? { additions: diffStats.added, deletions: diffStats.removed } : undefined, + ...(sanitizedDiff ? { diff: sanitizedDiff } : {}), + ...(autoApproved ? { autoApproved: true } : {}), + }) + } + await task.diffViewProvider.reset() this.resetPartialState() diff --git a/src/core/tools/ReadFileTool.ts b/src/core/tools/ReadFileTool.ts index 2107cfe21b..3647c631ee 100644 --- a/src/core/tools/ReadFileTool.ts +++ b/src/core/tools/ReadFileTool.ts @@ -16,6 +16,7 @@ import type { ReadFileParams, ReadFileMode, ReadFileToolParams, FileEntry, LineR import { isLegacyReadFileParams, type ClineSayTool } from "@roo-code/types" import { Task } from "../task/Task" +import { versionTokenOfStat } from "../../utils/versionToken" import { formatResponse } from "../prompts/responses" import { RecordSource } from "../context-tracking/FileContextTrackerTypes" import { isPathOutsideWorkspace } from "../../utils/pathUtils" @@ -214,12 +215,29 @@ export class ReadFileTool extends BaseTool<"read_file"> { // Read text file content with lossy UTF-8 conversion // Reading as Buffer first allows graceful handling of non-UTF8 bytes // (they become U+FFFD replacement characters instead of throwing) + // A2 (epic #1375): capture the on-disk token before the read so a mutation + // landing mid-read is detected by the post-read stat below. + const preReadStats = await fs.stat(fullPath, { bigint: true }).catch(() => undefined) const buffer = await fs.readFile(fullPath) const fileContent = buffer.toString("utf-8") const result = this.processTextFile(fileContent, entry) await task.fileContextTracker.trackFileContext(relPath, "read_tool" as RecordSource) + // A2 (plan #33 / epic #1375): record the observed on-disk version for the future write guard. + // The token is captured before AND after the read; the target is observed only + // when both match — a mutation between the two stats means the content the model + // received is not the on-disk state, and observing it would let a later write + // match a token the model never saw. A stat failure leaves the target + // unobserved and never fails the read. + const postReadStats = await fs.stat(fullPath, { bigint: true }).catch(() => undefined) + if (preReadStats && postReadStats) { + const preReadToken = versionTokenOfStat(preReadStats) + if (preReadToken === versionTokenOfStat(postReadStats)) { + task.observationRegistry.observe(fullPath, preReadToken) + } + } + updateFileResult(relPath, { nativeContent: `File: ${relPath}\n${result}`, }) @@ -768,6 +786,9 @@ export class ReadFileTool extends BaseTool<"read_file"> { } // Read text file + // A2 (epic #1375): capture the on-disk token before the read so a mutation + // landing mid-read is detected by the post-read stat below. + const preReadStats = await fs.stat(fullPath, { bigint: true }).catch(() => undefined) const rawContent = await fs.readFile(fullPath, "utf8") // Handle line ranges if specified @@ -799,6 +820,19 @@ export class ReadFileTool extends BaseTool<"read_file"> { // Track file in context await task.fileContextTracker.trackFileContext(relPath, "read_tool") + + // A2 (plan #33 / epic #1375): mirror the native path — record the observed + // on-disk version so legacy-format reads also feed the future write guard. + // Observe only when the pre-read and post-read tokens match (a mutation between + // them means the returned content is not the on-disk state). A stat failure + // leaves the target unobserved and never fails the read. + const postReadStats = await fs.stat(fullPath, { bigint: true }).catch(() => undefined) + if (preReadStats && postReadStats) { + const preReadToken = versionTokenOfStat(preReadStats) + if (preReadToken === versionTokenOfStat(postReadStats)) { + task.observationRegistry.observe(fullPath, preReadToken) + } + } } catch (error) { const errorMsg = error instanceof Error ? error.message : String(error) results.push(`File: ${relPath}\nError: ${errorMsg}`) diff --git a/src/core/tools/SearchReplaceTool.ts b/src/core/tools/SearchReplaceTool.ts index e29b124010..b421543606 100644 --- a/src/core/tools/SearchReplaceTool.ts +++ b/src/core/tools/SearchReplaceTool.ts @@ -11,6 +11,8 @@ import { RecordSource } from "../context-tracking/FileContextTrackerTypes" import { fileExistsAtPath } from "../../utils/fs" import { EXPERIMENT_IDS, experiments } from "../../shared/experiments" import { sanitizeUnifiedDiff, computeDiffStats } from "../diff/stats" +import { checkpointSave } from "../../core/checkpoints" +import { checkAutoApproval } from "../auto-approval" import type { ToolUse } from "../../shared/tools" import { BaseTool, ToolCallbacks } from "./BaseTool" @@ -163,6 +165,7 @@ export class SearchReplaceTool extends BaseTool<"search_replace"> { const state = await provider?.getState() const diagnosticsEnabled = state?.diagnosticsEnabled ?? true const writeDelayMs = state?.writeDelayMs ?? DEFAULT_WRITE_DELAY_MS + const perWriteCheckpoints = state?.perWriteCheckpoints ?? true const isPreventFocusDisruptionEnabled = experiments.isEnabled( state?.experiments ?? {}, EXPERIMENT_IDS.PREVENT_FOCUS_DISRUPTION, @@ -207,8 +210,16 @@ export class SearchReplaceTool extends BaseTool<"search_replace"> { // Save the changes if (isPreventFocusDisruptionEnabled) { - // Direct file write without diff view or opening the file - await task.diffViewProvider.saveDirectly(relPath, newContent, false, diagnosticsEnabled, writeDelayMs) + // Direct file write without diff view or opening the file. This tool only + // edits existing files, so edit-guard semantics require a prior read. + await task.diffViewProvider.saveDirectly( + relPath, + newContent, + false, + diagnosticsEnabled, + writeDelayMs, + "edit", + ) } else { // Call saveChanges to update the DiffViewProvider properties await task.diffViewProvider.saveChanges(diagnosticsEnabled, writeDelayMs) @@ -225,6 +236,35 @@ export class SearchReplaceTool extends BaseTool<"search_replace"> { const message = await task.diffViewProvider.pushToolWriteResult(task, task.cwd, false) pushToolResult(message) + if (perWriteCheckpoints) { + // B2: the change-journal entry for this search-and-replace is appended + // inside checkpointSave (the hook stays a single call site), keyed + // by the checkpoint commit that call produces. B3a threads the + // approval diff (for the change card) and whether the step was auto- + // approved (auto-approved steps always get the compact card). + const autoApproved = + ( + await checkAutoApproval({ + state, + cwd: task.cwd, + ask: "tool", + text: completeMessage, + isProtected: isWriteProtected, + }) + ).decision === "approve" + // Awaited: a later tool block must not interleave with this + // search-and-replace's staging/commit/journal/change-card work. + // checkpointSave never rejects (service call wrapped in try/catch + // upstream). + await checkpointSave(task, false, true, { + path: relPath, + operation: "update", + diffStats: diffStats ? { additions: diffStats.added, deletions: diffStats.removed } : undefined, + ...(sanitizedDiff ? { diff: sanitizedDiff } : {}), + ...(autoApproved ? { autoApproved: true } : {}), + }) + } + await task.diffViewProvider.reset() this.resetPartialState() diff --git a/src/core/tools/WriteToFileTool.ts b/src/core/tools/WriteToFileTool.ts index ae026b4b86..b1e4a46345 100644 --- a/src/core/tools/WriteToFileTool.ts +++ b/src/core/tools/WriteToFileTool.ts @@ -1,5 +1,4 @@ import path from "path" -import delay from "delay" import fs from "fs/promises" import { type ClineSayTool, DEFAULT_WRITE_DELAY_MS } from "@roo-code/types" @@ -13,7 +12,9 @@ import { getReadablePath } from "../../utils/path" import { isPathOutsideWorkspace } from "../../utils/pathUtils" import { unescapeHtmlEntities } from "../../utils/text-normalization" import { EXPERIMENT_IDS, experiments } from "../../shared/experiments" -import { convertNewFileToUnifiedDiff, computeDiffStats, sanitizeUnifiedDiff } from "../diff/stats" +import { convertNewFileToUnifiedDiff, computeDiffStats, sanitizeUnifiedDiff, type DiffStats } from "../diff/stats" +import { checkpointSave } from "../checkpoints" +import { checkAutoApproval } from "../auto-approval" import type { ToolUse } from "../../shared/tools" import { BaseTool, ToolCallbacks } from "./BaseTool" @@ -103,11 +104,20 @@ export class WriteToFileTool extends BaseTool<"write_to_file"> { const state = await provider?.getState() const diagnosticsEnabled = state?.diagnosticsEnabled ?? true const writeDelayMs = state?.writeDelayMs ?? DEFAULT_WRITE_DELAY_MS + const perWriteCheckpoints = state?.perWriteCheckpoints ?? true const isPreventFocusDisruptionEnabled = experiments.isEnabled( state?.experiments ?? {}, EXPERIMENT_IDS.PREVENT_FOCUS_DISRUPTION, ) + // B2: the approval-diff stats for the write, shared by both the + // approval message and the change-journal entry below. B3a also + // reuses the sanitized unified diff itself for the per-step change + // card (never recomputed). + let approvalDiffStats: DiffStats | null = null + let approvalDiff = "" + let completeMessage = "" + if (isPreventFocusDisruptionEnabled) { task.diffViewProvider.editType = fileExists ? "modify" : "create" if (fileExists) { @@ -121,10 +131,12 @@ export class WriteToFileTool extends BaseTool<"write_to_file"> { ? formatResponse.createPrettyPatch(relPath, task.diffViewProvider.originalContent, newContent) : convertNewFileToUnifiedDiff(newContent, relPath) unified = sanitizeUnifiedDiff(unified) - const completeMessage = JSON.stringify({ + approvalDiffStats = computeDiffStats(unified) + approvalDiff = unified + completeMessage = JSON.stringify({ ...sharedMessageProps, content: unified, - diffStats: computeDiffStats(unified) || undefined, + diffStats: approvalDiffStats || undefined, } satisfies ClineSayTool) const didApprove = await askApproval("tool", completeMessage, undefined, isWriteProtected) @@ -133,7 +145,16 @@ export class WriteToFileTool extends BaseTool<"write_to_file"> { return } - await task.diffViewProvider.saveDirectly(relPath, newContent, false, diagnosticsEnabled, writeDelayMs) + // Guarded publish: this write carries the complete file content, so it uses + // create-guard semantics (unobserved targets may only be created when absent). + await task.diffViewProvider.saveDirectly( + relPath, + newContent, + false, + diagnosticsEnabled, + writeDelayMs, + "create", + ) } else { if (!task.diffViewProvider.isEditing) { const partialMessage = JSON.stringify(sharedMessageProps) @@ -146,17 +167,18 @@ export class WriteToFileTool extends BaseTool<"write_to_file"> { true, ) - await delay(300) task.diffViewProvider.scrollToFirstDiff() let unified = fileExists ? formatResponse.createPrettyPatch(relPath, task.diffViewProvider.originalContent, newContent) : convertNewFileToUnifiedDiff(newContent, relPath) unified = sanitizeUnifiedDiff(unified) - const completeMessage = JSON.stringify({ + approvalDiffStats = computeDiffStats(unified) + approvalDiff = unified + completeMessage = JSON.stringify({ ...sharedMessageProps, content: unified, - diffStats: computeDiffStats(unified) || undefined, + diffStats: approvalDiffStats || undefined, } satisfies ClineSayTool) const didApprove = await askApproval("tool", completeMessage, undefined, isWriteProtected) @@ -179,6 +201,35 @@ export class WriteToFileTool extends BaseTool<"write_to_file"> { pushToolResult(message) + if (perWriteCheckpoints) { + // B2: the change-journal entry for this write is appended inside + // checkpointSave (the hook stays a single call site), keyed by the + // checkpoint commit that call produces. Await so the checkpoint + // (staging + commit) finishes before the next queued write starts; + // otherwise two writes can collapse into one commit. B3a threads + // the approval diff (for the change card) and whether the step was + // auto-approved (auto-approved steps always get the compact card). + const autoApproved = + ( + await checkAutoApproval({ + state, + cwd: task.cwd, + ask: "tool", + text: completeMessage, + isProtected: isWriteProtected, + }) + ).decision === "approve" + await checkpointSave(task, false, true, { + path: relPath, + operation: fileExists ? "update" : "create", + diffStats: approvalDiffStats + ? { additions: approvalDiffStats.added, deletions: approvalDiffStats.removed } + : undefined, + ...(approvalDiff ? { diff: approvalDiff } : {}), + ...(autoApproved ? { autoApproved: true } : {}), + }).catch(() => {}) + } + await task.diffViewProvider.reset() this.resetPartialState() diff --git a/src/core/tools/__tests__/applyDiffTool.changeCard.spec.ts b/src/core/tools/__tests__/applyDiffTool.changeCard.spec.ts new file mode 100644 index 0000000000..de8914ce00 --- /dev/null +++ b/src/core/tools/__tests__/applyDiffTool.changeCard.spec.ts @@ -0,0 +1,205 @@ +// npx vitest run core/tools/__tests__/applyDiffTool.changeCard.spec.ts + +import type { MockedFunction } from "vitest" + +import { fileExistsAtPath } from "../../../utils/fs" +import { checkAutoApproval } from "../../auto-approval" +import { checkpointSave } from "../../checkpoints" +import type { Task } from "../../task/Task" +import { ApplyDiffTool } from "../ApplyDiffTool" + +vi.mock("fs/promises", () => ({ + default: { + readFile: vi.fn().mockResolvedValue("original file content\n"), + // The trial aggregate adds the S4b stat-pair self-observation around this + // read (ApplyDiffTool records its own read under the ReadFileTool contract); + // its `.catch(() => undefined)` fallback keeps the observation no-op when + // stat rejects, so this spec stays green on both the component branch and + // the trial aggregate. + stat: vi.fn().mockRejectedValue(new Error("stat unavailable in this spec")), + }, +})) + +vi.mock("../../../utils/fs", () => ({ + fileExistsAtPath: vi.fn().mockResolvedValue(true), +})) + +vi.mock("../../prompts/responses", () => ({ + formatResponse: { + toolError: vi.fn((msg: string) => `Error: ${msg}`), + rooIgnoreError: vi.fn((filePath: string) => `Access denied: ${filePath}`), + createPrettyPatch: vi.fn(() => "mock-diff"), + }, +})) + +vi.mock("../../diff/stats", () => ({ + // The real DiffStats shape is { added, removed } (the tool maps it to the + // change-card { additions, deletions } pair). + sanitizeUnifiedDiff: vi.fn((diff: string) => diff), + computeDiffStats: vi.fn(() => ({ added: 1, removed: 1 })), +})) + +vi.mock("../../checkpoints", () => ({ + checkpointSave: vi.fn().mockResolvedValue(undefined), +})) + +vi.mock("../../auto-approval", () => ({ + checkAutoApproval: vi.fn().mockResolvedValue({ decision: "ask" }), +})) + +describe("ApplyDiffTool.execute - per-write checkpoint and change card (B3a, epic #1375)", () => { + const mockedFileExistsAtPath = fileExistsAtPath as MockedFunction + const mockCheckpointSave = checkpointSave as MockedFunction + const mockCheckAutoApproval = checkAutoApproval as MockedFunction + + let tool: ApplyDiffTool + let mockTask: Pick< + Task, + | "cwd" + | "consecutiveMistakeCount" + | "consecutiveMistakeCountForApplyDiff" + | "recordToolError" + | "rooIgnoreController" + | "rooProtectedController" + | "processQueuedMessages" + | "didEditFile" + | "api" + | "diffStrategy" + | "diffViewProvider" + | "providerRef" + | "fileContextTracker" + > + let mockSaveDirectly: MockedFunction<(...args: unknown[]) => Promise> + let mockGetState: MockedFunction<() => Promise>> + let mockAskApproval: MockedFunction<(...args: unknown[]) => Promise> + let mockHandleError: MockedFunction<(...args: unknown[]) => Promise> + let mockPushToolResult: MockedFunction<(...args: unknown[]) => void> + + beforeEach(() => { + vi.clearAllMocks() + + mockedFileExistsAtPath.mockResolvedValue(true) + + // Structural stubs for the prevent-focus-disruption save path: the real + // DiffViewProvider is out of scope here, so vi.fn() doubles stand in for + // the members the tool touches. + mockSaveDirectly = vi.fn().mockResolvedValue({ + newProblemsMessage: "", + userEdits: undefined, + finalContent: "new content", + }) + const diffViewProviderStub = { + editType: undefined as "create" | "modify" | undefined, + originalContent: undefined as string | undefined, + saveDirectly: mockSaveDirectly, + pushToolWriteResult: vi.fn().mockResolvedValue("Saved file"), + reset: vi.fn().mockResolvedValue(undefined), + } + + mockGetState = vi.fn().mockResolvedValue({ + diagnosticsEnabled: true, + writeDelayMs: 1000, + // Exercise the focus-disruption (saveDirectly) save path. + experiments: { preventFocusDisruption: true }, + }) + + mockTask = { + cwd: "/workspace/project", + consecutiveMistakeCount: 0, + consecutiveMistakeCountForApplyDiff: new Map(), + recordToolError: vi.fn(), + rooIgnoreController: { + validateAccess: vi.fn().mockReturnValue(true), + } as unknown as Task["rooIgnoreController"], + rooProtectedController: { + isWriteProtected: vi.fn().mockReturnValue(false), + } as unknown as Task["rooProtectedController"], + processQueuedMessages: vi.fn(), + didEditFile: false, + api: { + getModel: () => ({ id: "claude-sonnet-4-5" }), + } as unknown as Task["api"], + diffStrategy: { + applyDiff: vi.fn().mockResolvedValue({ success: true, content: "modified file content\n" }), + } as unknown as Task["diffStrategy"], + diffViewProvider: diffViewProviderStub as unknown as Task["diffViewProvider"], + providerRef: { + deref: vi.fn().mockReturnValue({ + getState: mockGetState, + }), + } as unknown as Task["providerRef"], + fileContextTracker: { + trackFileContext: vi.fn().mockResolvedValue(undefined), + } as unknown as Task["fileContextTracker"], + } + + mockAskApproval = vi.fn().mockResolvedValue(true) + mockHandleError = vi.fn().mockResolvedValue(undefined) + mockPushToolResult = vi.fn() + + tool = new ApplyDiffTool() + }) + + it("records a per-write checkpoint with the applied diff after a successful write", async () => { + await tool.execute({ path: "src/thing.ts", diff: "unified diff" }, mockTask as Task, { + askApproval: mockAskApproval, + handleError: mockHandleError, + pushToolResult: mockPushToolResult, + }) + + expect(mockCheckpointSave).toHaveBeenCalledTimes(1) + expect(mockCheckpointSave).toHaveBeenCalledWith(mockTask, false, true, { + path: "src/thing.ts", + operation: "update", + diffStats: { additions: 1, deletions: 1 }, + diff: "mock-diff", + }) + }) + + it("marks auto-approved steps so the card renders compact", async () => { + mockCheckAutoApproval.mockResolvedValueOnce({ decision: "approve" }) + + await tool.execute({ path: "src/thing.ts", diff: "unified diff" }, mockTask as Task, { + askApproval: mockAskApproval, + handleError: mockHandleError, + pushToolResult: mockPushToolResult, + }) + + expect(mockCheckpointSave).toHaveBeenCalledTimes(1) + expect(mockCheckpointSave.mock.calls[0]?.[3]).toEqual(expect.objectContaining({ autoApproved: true })) + }) + + it("skips the checkpoint when perWriteCheckpoints is explicitly disabled", async () => { + mockGetState.mockResolvedValueOnce({ + diagnosticsEnabled: true, + writeDelayMs: 1000, + experiments: { preventFocusDisruption: true }, + perWriteCheckpoints: false, + }) + + await tool.execute({ path: "src/thing.ts", diff: "unified diff" }, mockTask as Task, { + askApproval: mockAskApproval, + handleError: mockHandleError, + pushToolResult: mockPushToolResult, + }) + + expect(mockCheckpointSave).not.toHaveBeenCalled() + // The write itself still happens (the setting gates the checkpoint only), + // and auto-approval is never consulted when there is no card to build. + expect(mockSaveDirectly).toHaveBeenCalled() + expect(mockCheckAutoApproval).not.toHaveBeenCalled() + }) + + it("records nothing when the approval is declined", async () => { + mockAskApproval.mockResolvedValue(false) + + await tool.execute({ path: "src/thing.ts", diff: "unified diff" }, mockTask as Task, { + askApproval: mockAskApproval, + handleError: mockHandleError, + pushToolResult: mockPushToolResult, + }) + + expect(mockCheckpointSave).not.toHaveBeenCalled() + expect(mockSaveDirectly).not.toHaveBeenCalled() + }) +}) diff --git a/src/core/tools/__tests__/applyDiffTool.guardedWrite.spec.ts b/src/core/tools/__tests__/applyDiffTool.guardedWrite.spec.ts new file mode 100644 index 0000000000..7f8f6721d6 --- /dev/null +++ b/src/core/tools/__tests__/applyDiffTool.guardedWrite.spec.ts @@ -0,0 +1,211 @@ +// npx vitest run core/tools/__tests__/applyDiffTool.guardedWrite.spec.ts + +import type { MockedFunction } from "vitest" + +import * as fsPromises from "fs/promises" +import path from "path" + +import { fileExistsAtPath } from "../../../utils/fs" +import type { Task } from "../../task/Task" +import { ApplyDiffTool } from "../ApplyDiffTool" + +vi.mock("fs/promises", () => { + // BigInt stats so the S4b observation (ReadFileTool contract) can build a + // deterministic version token. Stable values = pre/post tokens agree. The + // named `stat` export is the same double as `default.stat` (the tool imports + // the default namespace; the spec asserts through the named one). + const statMock = vi.fn().mockResolvedValue({ + dev: 7n, + ino: 4242n, + size: 1234n, + mtimeNs: 1_700_000_000_123_456_789n, + ctimeNs: 1_700_000_000_789_999_999n, + }) + return { + default: { + readFile: vi.fn().mockResolvedValue("original file content\n"), + stat: statMock, + }, + stat: statMock, + } +}) + +vi.mock("../../../utils/fs", () => ({ + fileExistsAtPath: vi.fn().mockResolvedValue(true), +})) + +vi.mock("../../prompts/responses", () => ({ + formatResponse: { + toolError: vi.fn((msg: string) => `Error: ${msg}`), + rooIgnoreError: vi.fn((filePath: string) => `Access denied: ${filePath}`), + createPrettyPatch: vi.fn(() => "mock-diff"), + }, +})) + +vi.mock("../../diff/stats", () => ({ + sanitizeUnifiedDiff: vi.fn((diff: string) => diff), + computeDiffStats: vi.fn(() => ({ additions: 1, deletions: 1 })), +})) + +describe("ApplyDiffTool.execute - guarded write (S4b, epic #1375)", () => { + const mockedFileExistsAtPath = fileExistsAtPath as MockedFunction + + let tool: ApplyDiffTool + let mockTask: Pick< + Task, + | "cwd" + | "consecutiveMistakeCount" + | "consecutiveMistakeCountForApplyDiff" + | "recordToolError" + | "rooIgnoreController" + | "rooProtectedController" + | "say" + | "processQueuedMessages" + | "didEditFile" + | "api" + | "diffStrategy" + | "diffViewProvider" + | "providerRef" + | "fileContextTracker" + | "observationRegistry" + > + let mockSaveDirectly: MockedFunction<(...args: unknown[]) => Promise> + let mockObserve: MockedFunction<(...args: unknown[]) => void> + let mockAskApproval: MockedFunction<(...args: unknown[]) => Promise> + let mockHandleError: MockedFunction<(...args: unknown[]) => Promise> + let mockPushToolResult: MockedFunction<(...args: unknown[]) => void> + + beforeEach(() => { + vi.clearAllMocks() + + mockedFileExistsAtPath.mockResolvedValue(true) + + mockSaveDirectly = vi.fn().mockResolvedValue({ + newProblemsMessage: "", + userEdits: undefined, + finalContent: "new content", + }) + + // S2 observation registry double: the tool records its own read here before + // the guarded edit publish (S4b trial addendum). + mockObserve = vi.fn() + + // Structural stubs for the guarded-write path: the real DiffViewProvider is + // out of scope here, so vi.fn() doubles stand in for the members the tool + // touches (the saveDirectly double also records the writeKind plumbing). + const diffViewProviderStub = { + editType: undefined as "create" | "modify" | undefined, + originalContent: undefined as string | undefined, + saveDirectly: mockSaveDirectly, + pushToolWriteResult: vi.fn().mockResolvedValue("Saved file"), + reset: vi.fn().mockResolvedValue(undefined), + } + mockTask = { + cwd: "/workspace/project", + consecutiveMistakeCount: 0, + consecutiveMistakeCountForApplyDiff: new Map(), + recordToolError: vi.fn(), + rooIgnoreController: { + validateAccess: vi.fn().mockReturnValue(true), + } as unknown as Task["rooIgnoreController"], + rooProtectedController: { + isWriteProtected: vi.fn().mockReturnValue(false), + } as unknown as Task["rooProtectedController"], + say: vi.fn().mockResolvedValue(undefined), + processQueuedMessages: vi.fn(), + didEditFile: false, + api: { + getModel: () => ({ id: "claude-sonnet-4-5" }), + } as unknown as Task["api"], + diffStrategy: { + applyDiff: vi.fn().mockResolvedValue({ success: true, content: "modified file content\n" }), + } as unknown as Task["diffStrategy"], + diffViewProvider: diffViewProviderStub as unknown as Task["diffViewProvider"], + providerRef: { + deref: vi.fn().mockReturnValue({ + getState: vi.fn().mockResolvedValue({ + diagnosticsEnabled: true, + writeDelayMs: 1000, + // Exercise the focus-disruption (saveDirectly) save path. + experiments: { preventFocusDisruption: true }, + }), + }), + } as unknown as Task["providerRef"], + fileContextTracker: { + trackFileContext: vi.fn().mockResolvedValue(undefined), + } as unknown as Task["fileContextTracker"], + observationRegistry: { + observe: mockObserve, + } as unknown as Task["observationRegistry"], + } + + mockAskApproval = vi.fn().mockResolvedValue(true) + mockHandleError = vi.fn().mockResolvedValue(undefined) + mockPushToolResult = vi.fn() + + tool = new ApplyDiffTool() + }) + + it("publishes through the guarded saveDirectly with edit kind", async () => { + await tool.execute({ path: "src/thing.ts", diff: "unified diff" }, mockTask as Task, { + askApproval: mockAskApproval, + handleError: mockHandleError, + pushToolResult: mockPushToolResult, + }) + + expect(mockSaveDirectly).toHaveBeenCalledWith( + "src/thing.ts", + "modified file content\n", + false, + true, + 1000, + "edit", + ) + // S4b (trial addendum): the tool's own read is recorded as an observation so + // the guarded edit publish can CAS against the observed token. + expect(mockObserve).toHaveBeenCalledTimes(1) + expect(mockObserve.mock.calls[0]?.[0]).toBe(path.resolve(mockTask.cwd, "src/thing.ts")) + expect(mockObserve.mock.calls[0]?.[1]).toBe("7:4242:1234:1700000000123456789:1700000000789999999") + expect(mockPushToolResult).toHaveBeenCalledWith("Saved file") + expect(mockTask.didEditFile).toBe(true) + expect(mockHandleError).not.toHaveBeenCalled() + }) + + it("does not observe when the pre- and post-read tokens disagree", async () => { + // A mutation between the two stats means the content the diff was applied + // to is not the on-disk state: leave the target unobserved (ReadFileTool + // contract) so the guarded write gets a stale rejection, not a false match. + const preStats = { dev: 7n, ino: 4242n, size: 1234n, mtimeNs: 1n, ctimeNs: 2n } + const postStats = { dev: 7n, ino: 4242n, size: 9999n, mtimeNs: 3n, ctimeNs: 4n } + const mockedStat = vi.mocked(fsPromises.stat) + mockedStat.mockResolvedValueOnce(preStats as unknown as Awaited>) + mockedStat.mockResolvedValueOnce(postStats as unknown as Awaited>) + + await tool.execute({ path: "src/thing.ts", diff: "unified diff" }, mockTask as Task, { + askApproval: mockAskApproval, + handleError: mockHandleError, + pushToolResult: mockPushToolResult, + }) + + expect(mockObserve).not.toHaveBeenCalled() + // The publish still runs (the saveDirectly double stands in for the guard); + // the registry simply carries no observation for this path. + expect(mockSaveDirectly).toHaveBeenCalled() + }) + + it("surfaces the unobserved edit remediation as a tool error", async () => { + const guardError = new Error("File not read yet -- read the file, then retry.") + mockSaveDirectly.mockRejectedValue(guardError) + + await tool.execute({ path: "src/thing.ts", diff: "unified diff" }, mockTask as Task, { + askApproval: mockAskApproval, + handleError: mockHandleError, + pushToolResult: mockPushToolResult, + }) + + expect(mockHandleError).toHaveBeenCalledWith("applying diff", guardError) + expect(vi.mocked(mockTask.diffViewProvider.reset)).toHaveBeenCalled() + expect(mockTask.didEditFile).toBe(false) + expect(mockPushToolResult).not.toHaveBeenCalledWith("Saved file") + }) +}) diff --git a/src/core/tools/__tests__/applyPatchTool.execute.spec.ts b/src/core/tools/__tests__/applyPatchTool.execute.spec.ts index 72ffb112bc..3c1af124fd 100644 --- a/src/core/tools/__tests__/applyPatchTool.execute.spec.ts +++ b/src/core/tools/__tests__/applyPatchTool.execute.spec.ts @@ -4,13 +4,41 @@ import type { MockedFunction } from "vitest" import { fileExistsAtPath } from "../../../utils/fs" import { isPathOutsideWorkspace } from "../../../utils/pathUtils" +import path from "path" +import * as fsPromises from "fs/promises" import type { Task } from "../../task/Task" +import { checkpointSave } from "../../checkpoints" +import { ObservationRegistry } from "../../task/observationRegistry" import { ApplyPatchTool } from "../ApplyPatchTool" +// The vi.mock factory exposes the fs/promises functions under a `default` +// property (matching the SUT's default import), which the static module type +// does not declare; cast once at this boundary rather than at each call site. +const mockedFsPromises = vi.mocked( + fsPromises as unknown as { + default: { + stat: ReturnType + unlink: MockedFunction + writeFile: MockedFunction + } + }, +) + vi.mock("fs/promises", () => ({ default: { readFile: vi.fn().mockResolvedValue("original file content\n"), + // Stable on-disk version for the S2 self-read observation (the hunk + // read now stats before and after; equal tokens record the observe). + stat: vi.fn().mockResolvedValue({ + dev: 7n, + ino: 4242n, + size: 1234n, + mtimeNs: 1_700_000_000_123_456_789n, + ctimeNs: 1_700_000_000_789_999_999n, + }), unlink: vi.fn().mockResolvedValue(undefined), + mkdir: vi.fn().mockResolvedValue(undefined), + writeFile: vi.fn().mockResolvedValue(undefined), }, })) @@ -22,6 +50,13 @@ vi.mock("../../../utils/pathUtils", () => ({ isPathOutsideWorkspace: vi.fn().mockReturnValue(false), })) +vi.mock("../../checkpoints", () => ({ + getCheckpointService: vi.fn(), + checkpointSave: vi.fn().mockResolvedValue(undefined), + checkpointRestore: vi.fn(), + checkpointDiff: vi.fn(), +})) + describe("ApplyPatchTool.execute - delete file success path", () => { const mockedFileExistsAtPath = fileExistsAtPath as MockedFunction const mockedIsPathOutsideWorkspace = isPathOutsideWorkspace as MockedFunction @@ -38,6 +73,10 @@ describe("ApplyPatchTool.execute - delete file success path", () => { | "say" | "processQueuedMessages" | "didEditFile" + | "providerRef" + | "diffViewProvider" + | "fileContextTracker" + | "observationRegistry" > let mockAskApproval: MockedFunction<(...args: unknown[]) => Promise> let mockHandleError: MockedFunction<(...args: unknown[]) => Promise> @@ -52,6 +91,15 @@ describe("ApplyPatchTool.execute - delete file success path", () => { mockTask = { cwd: "/workspace/project", consecutiveMistakeCount: 0, + observationRegistry: new ObservationRegistry(), + providerRef: { + deref: vi.fn().mockReturnValue({ + // Trial addendum: pin the legacy diff-editor path; the L2 default flipped + // preventFocusDisruption to true (chat-diff is the default approval path). + // The PFD-branch test below opts in explicitly. + getState: vi.fn().mockResolvedValue({ experiments: { preventFocusDisruption: false } }), + }), + } as unknown as Task["providerRef"], recordToolUsage: vi.fn(), recordToolError: vi.fn(), rooIgnoreController: { @@ -63,6 +111,21 @@ describe("ApplyPatchTool.execute - delete file success path", () => { say: vi.fn().mockResolvedValue(undefined), processQueuedMessages: vi.fn(), didEditFile: false, + diffViewProvider: { + editType: "modify", + originalContent: undefined, + open: vi.fn().mockResolvedValue(undefined), + update: vi.fn().mockResolvedValue(undefined), + scrollToFirstDiff: vi.fn(), + revertChanges: vi.fn().mockResolvedValue(undefined), + reset: vi.fn().mockResolvedValue(undefined), + saveDirectly: vi.fn().mockResolvedValue({ finalContent: "saved" }), + saveChanges: vi.fn().mockResolvedValue(undefined), + pushToolWriteResult: vi.fn().mockResolvedValue("File saved successfully"), + } as unknown as Task["diffViewProvider"], + fileContextTracker: { + trackFileContext: vi.fn().mockResolvedValue(undefined), + } as unknown as Task["fileContextTracker"], } mockAskApproval = vi.fn().mockResolvedValue(true) @@ -93,4 +156,812 @@ describe("ApplyPatchTool.execute - delete file success path", () => { expect(mockTask.recordToolUsage).not.toHaveBeenCalled() expect(mockTask.recordToolError).not.toHaveBeenCalled() }) + + describe("per-write checkpoints (B1)", () => { + const deletePatch = `*** Begin Patch +*** Delete File: src/obsolete.ts +*** End Patch` + const mockedCheckpointSave = checkpointSave as MockedFunction + + it("records one suppressed checkpoint for the whole patch (default-on)", async () => { + await tool.execute({ patch: deletePatch }, mockTask as Task, { + askApproval: mockAskApproval, + handleError: mockHandleError, + pushToolResult: mockPushToolResult, + }) + + expect(mockPushToolResult).toHaveBeenCalledWith(expect.stringContaining("Successfully deleted")) + expect(mockedCheckpointSave).toHaveBeenCalledOnce() + // B2: the delete patch produces one journal write, referencing the + // single checkpoint saved for the whole patch. + expect(mockedCheckpointSave).toHaveBeenCalledWith(mockTask as Task, false, true, [ + { path: "src/obsolete.ts", operation: "delete" }, + ]) + }) + + it("does not record a checkpoint when perWriteCheckpoints is disabled", async () => { + // Structural cast for the test double (matches the mock style used for the controllers above). + const ref = (mockTask["providerRef"] as unknown as { deref: MockedFunction<() => unknown> }).deref + ref.mockReturnValue({ + getState: vi.fn().mockResolvedValue({ perWriteCheckpoints: false }), + }) + + await tool.execute({ patch: deletePatch }, mockTask as Task, { + askApproval: mockAskApproval, + handleError: mockHandleError, + pushToolResult: mockPushToolResult, + }) + + expect(mockPushToolResult).toHaveBeenCalledWith(expect.stringContaining("Successfully deleted")) + expect(mockedCheckpointSave).not.toHaveBeenCalled() + }) + + it("does not record a checkpoint when patch processing fails", async () => { + // A malformed patch fails at parse time, before the change loop and + // the post-loop checkpoint hook. + const badPatch = `*** Begin Patch +*** This is not a valid hunk +*** End Patch` + + await tool.execute({ patch: badPatch }, mockTask as Task, { + askApproval: mockAskApproval, + handleError: mockHandleError, + pushToolResult: mockPushToolResult, + }) + + expect(mockTask.recordToolError).toHaveBeenCalledWith("apply_patch") + expect(mockedCheckpointSave).not.toHaveBeenCalled() + }) + }) + + describe("checkpoint only for fully successful patches (B1)", () => { + const mockedCheckpointSave = checkpointSave as MockedFunction + const deletePatch = `*** Begin Patch +*** Delete File: src/obsolete.ts +*** End Patch` + const addPatch = `*** Begin Patch +*** Add File: src/new.ts ++hello ++world +*** End Patch` + const updatePatch = `*** Begin Patch +*** Update File: src/test.ts +@@ +-original file content ++modified content +*** End Patch` + const updateNoDiffPatch = `*** Begin Patch +*** Update File: src/test.ts +@@ +-original file content ++original file content +*** End Patch` + const movePatch = `*** Begin Patch +*** Update File: src/test.ts +*** Move to: src/moved.ts +@@ +-original file content ++modified content +*** End Patch` + + it("does not record a checkpoint when the user rejects the patch", async () => { + // Rejected approval: the handler early-returns without recording a + // tool error, so the success flag must come from the handler itself. + mockAskApproval.mockResolvedValue(false) + + await tool.execute({ patch: deletePatch }, mockTask as Task, { + askApproval: mockAskApproval, + handleError: mockHandleError, + pushToolResult: mockPushToolResult, + }) + + expect(mockPushToolResult).toHaveBeenCalledWith("Delete operation was rejected by the user.") + expect(mockedCheckpointSave).not.toHaveBeenCalled() + }) + + it("does not record a checkpoint when the file to delete does not exist", async () => { + mockedFileExistsAtPath.mockResolvedValue(false) + + await tool.execute({ patch: deletePatch }, mockTask as Task, { + askApproval: mockAskApproval, + handleError: mockHandleError, + pushToolResult: mockPushToolResult, + }) + + expect(mockPushToolResult).toHaveBeenCalledWith(expect.stringContaining("File not found")) + expect(mockedCheckpointSave).not.toHaveBeenCalled() + }) + + it("does not record a checkpoint when the delete write fails", async () => { + mockedFsPromises.default.unlink.mockRejectedValueOnce(new Error("EBUSY")) + + await tool.execute({ patch: deletePatch }, mockTask as Task, { + askApproval: mockAskApproval, + handleError: mockHandleError, + pushToolResult: mockPushToolResult, + }) + + expect(mockPushToolResult).toHaveBeenCalledWith(expect.stringContaining("Failed to delete file")) + expect(mockedCheckpointSave).not.toHaveBeenCalled() + }) + + it("does not record a checkpoint when the added file already exists", async () => { + // fileExistsAtPath resolves true by default in beforeEach. + + await tool.execute({ patch: addPatch }, mockTask as Task, { + askApproval: mockAskApproval, + handleError: mockHandleError, + pushToolResult: mockPushToolResult, + }) + + expect(mockPushToolResult).toHaveBeenCalledWith(expect.stringContaining("File already exists")) + expect(mockedCheckpointSave).not.toHaveBeenCalled() + }) + + it("does not record a checkpoint when the user rejects the add", async () => { + mockedFileExistsAtPath.mockResolvedValue(false) + mockAskApproval.mockResolvedValue(false) + + await tool.execute({ patch: addPatch }, mockTask as Task, { + askApproval: mockAskApproval, + handleError: mockHandleError, + pushToolResult: mockPushToolResult, + }) + + expect(mockPushToolResult).toHaveBeenCalledWith("Changes were rejected by the user.") + expect(mockedCheckpointSave).not.toHaveBeenCalled() + }) + + it("records a checkpoint when the add succeeds", async () => { + mockedFileExistsAtPath.mockResolvedValue(false) + + await tool.execute({ patch: addPatch }, mockTask as Task, { + askApproval: mockAskApproval, + handleError: mockHandleError, + pushToolResult: mockPushToolResult, + }) + + expect(mockPushToolResult).toHaveBeenCalledWith("File saved successfully") + expect(mockedCheckpointSave).toHaveBeenCalledOnce() + }) + + it("does not record a checkpoint when the file to update does not exist", async () => { + mockedFileExistsAtPath.mockResolvedValue(false) + + await tool.execute({ patch: updatePatch }, mockTask as Task, { + askApproval: mockAskApproval, + handleError: mockHandleError, + pushToolResult: mockPushToolResult, + }) + + expect(mockPushToolResult).toHaveBeenCalledWith(expect.stringContaining("File not found")) + expect(mockedCheckpointSave).not.toHaveBeenCalled() + }) + + it("records a checkpoint when the update is a no-op (no changes needed)", async () => { + // A no-op change is not a failure, so the whole-patch checkpoint still runs. + + await tool.execute({ patch: updateNoDiffPatch }, mockTask as Task, { + askApproval: mockAskApproval, + handleError: mockHandleError, + pushToolResult: mockPushToolResult, + }) + + expect(mockPushToolResult).toHaveBeenCalledWith(expect.stringContaining("No changes needed")) + expect(mockedCheckpointSave).toHaveBeenCalledOnce() + }) + + it("does not record a checkpoint when the user rejects the update", async () => { + mockAskApproval.mockResolvedValue(false) + + await tool.execute({ patch: updatePatch }, mockTask as Task, { + askApproval: mockAskApproval, + handleError: mockHandleError, + pushToolResult: mockPushToolResult, + }) + + expect(mockPushToolResult).toHaveBeenCalledWith("Changes were rejected by the user.") + expect(mockedCheckpointSave).not.toHaveBeenCalled() + }) + + it("does not record a checkpoint when the move destination is not allowed", async () => { + // First validateAccess call (source path, in the execute loop) passes; + // the move destination check inside the handler fails. + const validateAccess = ( + mockTask["rooIgnoreController"] as unknown as { validateAccess: MockedFunction<() => boolean> } + ).validateAccess + validateAccess.mockReturnValueOnce(true).mockReturnValue(false) + + await tool.execute({ patch: movePatch }, mockTask as Task, { + askApproval: mockAskApproval, + handleError: mockHandleError, + pushToolResult: mockPushToolResult, + }) + + expect(mockTask.say).toHaveBeenCalledWith("rooignore_error", "src/moved.ts") + expect(mockedCheckpointSave).not.toHaveBeenCalled() + }) + + it("checkpoints the written subset when a later hunk is access-denied", async () => { + // Hunk 1 (src/first.ts) writes; hunk 2 (src/denied.ts) is rejected by + // validateAccess. The access-denied branch must not bypass the partial + // flush: the earlier write still receives the checkpoint/journal/card. + // Hunk 2's context matches the mocked file content so the patch + // passes pre-processing; the denial happens at the per-file access check. + const partialDenyPatch = `*** Begin Patch +*** Add File: src/first.ts ++hello +*** Update File: src/denied.ts +@@ +-original file content ++new content +*** End Patch` + const validateAccess = ( + mockTask["rooIgnoreController"] as unknown as { validateAccess: MockedFunction<() => boolean> } + ).validateAccess + validateAccess.mockReturnValueOnce(true).mockReturnValueOnce(false) + // The add target does not exist, so hunk 1 writes; fileExistsAtPath + // defaults to true and would otherwise reject the add. + mockedFileExistsAtPath.mockResolvedValueOnce(false) + + await tool.execute({ patch: partialDenyPatch }, mockTask as Task, { + askApproval: mockAskApproval, + handleError: mockHandleError, + pushToolResult: mockPushToolResult, + }) + + expect(mockTask.say).toHaveBeenCalledWith("rooignore_error", "src/denied.ts") + // Only the first (written) hunk is checkpointed. + expect(mockedCheckpointSave).toHaveBeenCalledOnce() + expect(mockedCheckpointSave).toHaveBeenCalledWith(mockTask, false, true, [ + expect.objectContaining({ path: "src/first.ts", operation: "create" }), + ]) + }) + + it("does not record a checkpoint when the move destination is write-protected", async () => { + // Source path check (execute loop) passes; the move destination fails. + const isWriteProtected = ( + mockTask["rooProtectedController"] as unknown as { + isWriteProtected: MockedFunction<(p: string) => boolean> + } + ).isWriteProtected + isWriteProtected.mockReturnValueOnce(false).mockReturnValue(true) + + await tool.execute({ patch: movePatch }, mockTask as Task, { + askApproval: mockAskApproval, + handleError: mockHandleError, + pushToolResult: mockPushToolResult, + }) + + expect(mockPushToolResult).toHaveBeenCalledWith( + expect.stringContaining("Cannot move file to write-protected path"), + ) + expect(mockedCheckpointSave).not.toHaveBeenCalled() + }) + + it("does not record a checkpoint when the move destination is outside the workspace", async () => { + // Source path (first call) is inside; the move destination (second) + // call is outside the workspace. + mockedIsPathOutsideWorkspace.mockReturnValueOnce(false).mockReturnValue(true) + + await tool.execute({ patch: movePatch }, mockTask as Task, { + askApproval: mockAskApproval, + handleError: mockHandleError, + pushToolResult: mockPushToolResult, + }) + + expect(mockPushToolResult).toHaveBeenCalledWith( + expect.stringContaining("Cannot move file to path outside workspace"), + ) + expect(mockedCheckpointSave).not.toHaveBeenCalled() + }) + + it("keeps the mistake count when a patch operation fails", async () => { + // Source path check (execute loop) passes; the move destination fails. + const isWriteProtected = ( + mockTask["rooProtectedController"] as unknown as { + isWriteProtected: MockedFunction<(p: string) => boolean> + } + ).isWriteProtected + isWriteProtected.mockReturnValueOnce(false).mockReturnValue(true) + + await tool.execute({ patch: movePatch }, mockTask as Task, { + askApproval: mockAskApproval, + handleError: mockHandleError, + pushToolResult: mockPushToolResult, + }) + + // The failed operation incremented the counter; the end-of-loop reset + // must only run for a fully successful patch, so the count survives. + expect(mockTask.consecutiveMistakeCount).toBe(1) + }) + + it("clears the mistake count after a fully successful patch", async () => { + mockTask.consecutiveMistakeCount = 2 + + await tool.execute({ patch: movePatch }, mockTask as Task, { + askApproval: mockAskApproval, + handleError: mockHandleError, + pushToolResult: mockPushToolResult, + }) + + expect(mockTask.consecutiveMistakeCount).toBe(0) + }) + + it("records a checkpoint when the move succeeds", async () => { + await tool.execute({ patch: movePatch }, mockTask as Task, { + askApproval: mockAskApproval, + handleError: mockHandleError, + pushToolResult: mockPushToolResult, + }) + + // path is platform-dependent (Windows resolves cwd to a drive path); + // assert on the written content instead. + expect(mockedFsPromises.default.writeFile).toHaveBeenCalledWith( + expect.any(String), + "modified content\n", + "utf8", + ) + expect(mockedCheckpointSave).toHaveBeenCalledOnce() + }) + + it("awaits the patch checkpoint before execute settles", async () => { + type SaveResult = Awaited> + let resolveSave: (value: SaveResult | PromiseLike) => void = () => {} + const saveDeferred = new Promise((resolve) => (resolveSave = resolve)) + mockedCheckpointSave.mockImplementationOnce(() => saveDeferred) + + const executePromise = tool.execute({ patch: movePatch }, mockTask as Task, { + askApproval: mockAskApproval, + handleError: mockHandleError, + pushToolResult: mockPushToolResult, + }) + + // execute must not settle while the checkpoint is still in flight: + // a later write would otherwise interleave with this patch's staged work. + let settled = false + void executePromise.finally(() => (settled = true)) + await new Promise((resolve) => setTimeout(resolve, 0)) + expect(mockedCheckpointSave).toHaveBeenCalledOnce() + expect(settled).toBe(false) + + resolveSave() + await executePromise + expect(settled).toBe(true) + }) + + it("records a checkpoint when the in-place update succeeds", async () => { + await tool.execute({ patch: updatePatch }, mockTask as Task, { + askApproval: mockAskApproval, + handleError: mockHandleError, + pushToolResult: mockPushToolResult, + }) + + expect(mockPushToolResult).toHaveBeenCalledWith("File saved successfully") + expect(mockedCheckpointSave).toHaveBeenCalledOnce() + }) + + it("records one journal write per file change for a multi-file patch", async () => { + // src/a.ts does not exist (add); src/b.ts does (update). + mockedFileExistsAtPath.mockImplementation((filePath: string) => + Promise.resolve(!String(filePath).toLowerCase().endsWith("a.ts")), + ) + const multiPatch = [ + "*** Begin Patch", + "*** Add File: src/a.ts", + "+alpha", + "*** Update File: src/b.ts", + "@@", + "-original file content", + "+second content", + "*** End Patch", + ].join(String.fromCharCode(10)) + + await tool.execute({ patch: multiPatch }, mockTask as Task, { + askApproval: mockAskApproval, + handleError: mockHandleError, + pushToolResult: mockPushToolResult, + }) + + expect(mockedCheckpointSave).toHaveBeenCalledOnce() + // B3a: each write threads the approval diff and stats computed by its + // handler so the per-step change card can reuse them verbatim. + expect(mockedCheckpointSave).toHaveBeenCalledWith(mockTask as Task, false, true, [ + { + path: "src/a.ts", + operation: "create", + diffStats: { additions: 1, deletions: 0 }, + diff: expect.stringContaining("+alpha"), + }, + { + path: "src/b.ts", + operation: "update", + diffStats: { additions: 1, deletions: 1 }, + diff: expect.stringContaining("+second content"), + }, + ]) + }) + }) + + describe("change-card threading (B3a)", () => { + const mockedCheckpointSave = checkpointSave as MockedFunction + const deletePatch = `*** Begin Patch + *** Delete File: src/obsolete.ts + *** End Patch` + + it("threads autoApproved into the checkpoint writes for auto-approved steps", async () => { + // B3a: auto-approved steps carry autoApproved on every write so + // checkpointSave can force the compact (summary) change card. + const ref = (mockTask["providerRef"] as unknown as { deref: MockedFunction<() => unknown> }).deref + ref.mockReturnValue({ + getState: vi.fn().mockResolvedValue({ autoApprovalEnabled: true, alwaysAllowWrite: true }), + }) + + await tool.execute({ patch: deletePatch }, mockTask as Task, { + askApproval: mockAskApproval, + handleError: mockHandleError, + pushToolResult: mockPushToolResult, + }) + + expect(mockedCheckpointSave).toHaveBeenCalledWith(mockTask as Task, false, true, [ + { path: "src/obsolete.ts", operation: "delete", autoApproved: true }, + ]) + }) + + it("omits autoApproved when the step is not auto-approved", async () => { + // The default provider state ({}) disables auto-approval, so no write + // carries the autoApproved flag and the card follows the user setting. + await tool.execute({ patch: deletePatch }, mockTask as Task, { + askApproval: mockAskApproval, + handleError: mockHandleError, + pushToolResult: mockPushToolResult, + }) + + expect(mockedCheckpointSave).toHaveBeenCalledWith(mockTask as Task, false, true, [ + { path: "src/obsolete.ts", operation: "delete" }, + ]) + }) + + it("journals only the files actually written for a mixed no-op and write patch", async () => { + // src/same.ts exists and the hunk rewrites identical content (a + // no-op update); src/new.ts does not exist (a real write). + mockedFileExistsAtPath.mockImplementation((filePath: string) => + Promise.resolve(!String(filePath).toLowerCase().endsWith("new.ts")), + ) + const mixedPatch = [ + "*** Begin Patch", + "*** Update File: src/same.ts", + "@@", + "-original file content", + "+original file content", + "*** Add File: src/new.ts", + "+fresh content", + "*** End Patch", + ].join(String.fromCharCode(10)) + + await tool.execute({ patch: mixedPatch }, mockTask as Task, { + askApproval: mockAskApproval, + handleError: mockHandleError, + pushToolResult: mockPushToolResult, + }) + + // The no-op update is reported to the model... + expect(mockPushToolResult).toHaveBeenCalledWith("No changes needed for 'src/same.ts'") + // ...but the journal documents only the file that was actually + // written, even though the whole patch succeeded. + expect(mockedCheckpointSave).toHaveBeenCalledOnce() + expect(mockedCheckpointSave).toHaveBeenCalledWith(mockTask as Task, false, true, [ + { + path: "src/new.ts", + operation: "create", + diffStats: { additions: 1, deletions: 0 }, + diff: expect.stringContaining("+fresh content"), + }, + ]) + }) + + it("still checkpoints the successful subset when a later hunk fails", async () => { + // src/first.ts already exists (the add fails); src/second.ts does not + // (the add writes). The whole patch fails, but the written file is + // still documented by the checkpoint and journal. + mockedFileExistsAtPath.mockImplementation((filePath: string) => + Promise.resolve(String(filePath).toLowerCase().endsWith("first.ts")), + ) + const partialPatch = [ + "*** Begin Patch", + "*** Add File: src/first.ts", + "+boom", + "*** Add File: src/second.ts", + "+fresh", + "*** End Patch", + ].join(String.fromCharCode(10)) + + await tool.execute({ patch: partialPatch }, mockTask as Task, { + askApproval: mockAskApproval, + handleError: mockHandleError, + pushToolResult: mockPushToolResult, + }) + + // The failed operation is reported to the model... + expect(mockPushToolResult).toHaveBeenCalledWith(expect.stringContaining("File already exists")) + // ...and the successful subset is checkpointed and journaled. + expect(mockedCheckpointSave).toHaveBeenCalledOnce() + expect(mockedCheckpointSave).toHaveBeenCalledWith(mockTask as Task, false, true, [ + { + path: "src/second.ts", + operation: "create", + diffStats: { additions: 1, deletions: 0 }, + diff: expect.stringContaining("+fresh"), + }, + ]) + }) + + it("reports a failed move when the original file cannot be deleted", async () => { + mockedFsPromises.default.unlink.mockRejectedValueOnce(new Error("EBUSY: resource busy")) + const movePatch = [ + "*** Begin Patch", + "*** Update File: src/old.ts", + "*** Move to: src/new-location.ts", + "@@", + "-original file content", + "+new content", + "*** End Patch", + ].join(String.fromCharCode(10)) + + await tool.execute({ patch: movePatch }, mockTask as Task, { + askApproval: mockAskApproval, + handleError: mockHandleError, + pushToolResult: mockPushToolResult, + }) + + // The copy succeeded but the source still exists, so the move is + // reported as a failed tool error - but the destination write was + // made on disk and must still be covered by the checkpoint/journal. + expect(mockedCheckpointSave).toHaveBeenCalledOnce() + expect(mockedCheckpointSave).toHaveBeenCalledWith(mockTask as Task, false, true, [ + expect.objectContaining({ path: "src/new-location.ts", operation: "update" }), + ]) + expect(mockTask.recordToolError).toHaveBeenCalledWith("apply_patch") + expect(mockPushToolResult).toHaveBeenCalledWith( + expect.stringContaining("could not delete the original file"), + ) + }) + }) +}) + +describe("ApplyPatchTool.execute - guarded write (S4b, epic #1375)", () => { + const mockedFileExistsAtPath = fileExistsAtPath as MockedFunction + + let tool: ApplyPatchTool + let mockTask: Pick< + Task, + | "cwd" + | "consecutiveMistakeCount" + | "recordToolError" + | "rooIgnoreController" + | "rooProtectedController" + | "say" + | "processQueuedMessages" + | "didEditFile" + | "diffViewProvider" + | "providerRef" + | "fileContextTracker" + | "observationRegistry" + > + let mockSaveDirectly: MockedFunction<(...args: unknown[]) => Promise> + let mockAskApproval: MockedFunction<(...args: unknown[]) => Promise> + let mockHandleError: MockedFunction<(...args: unknown[]) => Promise> + let mockPushToolResult: MockedFunction<(...args: unknown[]) => void> + + const updatePatch = `*** Begin Patch +*** Update File: src/thing.ts +@@ +-original file content ++modified file content +*** End Patch` + + const addPatch = `*** Begin Patch +*** Add File: src/new.ts ++new line one ++new line two +*** End Patch` + + const movePatch = `*** Begin Patch +*** Update File: src/old.ts +*** Move to: src/new.ts +@@ +-original file content ++modified file content +*** End Patch` + + beforeEach(() => { + vi.clearAllMocks() + + mockedFileExistsAtPath.mockResolvedValue(true) + + mockSaveDirectly = vi.fn().mockResolvedValue({ + newProblemsMessage: "", + userEdits: undefined, + finalContent: "new content", + }) + + // Structural stubs for the guarded-write path: the real DiffViewProvider is + // out of scope here, so vi.fn() doubles stand in for the members the tool + // touches (the saveDirectly double also records the writeKind plumbing). + const diffViewProviderStub = { + editType: undefined as "create" | "modify" | undefined, + originalContent: undefined as string | undefined, + saveDirectly: mockSaveDirectly, + pushToolWriteResult: vi.fn().mockResolvedValue("Saved file"), + reset: vi.fn().mockResolvedValue(undefined), + } + mockTask = { + cwd: "/workspace/project", + consecutiveMistakeCount: 0, + observationRegistry: new ObservationRegistry(), + recordToolError: vi.fn(), + rooIgnoreController: { + validateAccess: vi.fn().mockReturnValue(true), + } as unknown as Task["rooIgnoreController"], + rooProtectedController: { + isWriteProtected: vi.fn().mockReturnValue(false), + } as unknown as Task["rooProtectedController"], + say: vi.fn().mockResolvedValue(undefined), + processQueuedMessages: vi.fn(), + didEditFile: false, + diffViewProvider: diffViewProviderStub as unknown as Task["diffViewProvider"], + providerRef: { + deref: vi.fn().mockReturnValue({ + getState: vi.fn().mockResolvedValue({ + diagnosticsEnabled: true, + writeDelayMs: 1000, + // Exercise the focus-disruption (saveDirectly) save path. + experiments: { preventFocusDisruption: true }, + }), + }), + } as unknown as Task["providerRef"], + fileContextTracker: { + trackFileContext: vi.fn().mockResolvedValue(undefined), + } as unknown as Task["fileContextTracker"], + } + + mockAskApproval = vi.fn().mockResolvedValue(true) + mockHandleError = vi.fn().mockResolvedValue(undefined) + mockPushToolResult = vi.fn() + + tool = new ApplyPatchTool() + }) + + it("update: publishes through the guarded saveDirectly with create kind", async () => { + await tool.execute({ patch: updatePatch }, mockTask as Task, { + askApproval: mockAskApproval, + handleError: mockHandleError, + pushToolResult: mockPushToolResult, + }) + + expect(mockSaveDirectly).toHaveBeenCalledWith( + "src/thing.ts", + "modified file content\n", + false, + true, + 1000, + "create", + ) + expect(mockPushToolResult).toHaveBeenCalledWith("Saved file") + expect(mockTask.didEditFile).toBe(true) + expect(mockHandleError).not.toHaveBeenCalled() + }) + + it("update: observes the hunk read so the guarded publish is not unobserved", async () => { + await tool.execute({ patch: updatePatch }, mockTask as Task, { + askApproval: mockAskApproval, + handleError: mockHandleError, + pushToolResult: mockPushToolResult, + }) + + // The hunk read doubles as the S2 observation (ReadFileTool contract): + // stable pre/post stats record the version token, so the in-place modify + // publish is not rejected as an unobserved write. + const observed = mockTask.observationRegistry.get(path.resolve("/workspace/project", "src/thing.ts")) + expect(observed?.version).toBe("7:4242:1234:1700000000123456789:1700000000789999999") + }) + + it("update: does not observe when the pre- and post-read tokens disagree", async () => { + // The file changed mid-read: pre/post stats differ, so no observation is + // recorded and the guarded publish surfaces the unobserved-existing + // remediation instead of publishing against a stale version. + const statMock = mockedFsPromises.default.stat + statMock.mockResolvedValueOnce({ dev: 7n, ino: 4242n, size: 1234n, mtimeNs: 1n, ctimeNs: 2n }) + statMock.mockResolvedValueOnce({ dev: 7n, ino: 4242n, size: 9999n, mtimeNs: 3n, ctimeNs: 4n }) + + const guardError = new Error( + "File already exists at /workspace/project/src/thing.ts and was not read before this write -- read the file first, then retry.", + ) + mockSaveDirectly.mockRejectedValue(guardError) + + await tool.execute({ patch: updatePatch }, mockTask as Task, { + askApproval: mockAskApproval, + handleError: mockHandleError, + pushToolResult: mockPushToolResult, + }) + + expect(mockTask.observationRegistry.has(path.resolve("/workspace/project", "src/thing.ts"))).toBe(false) + expect(mockHandleError).toHaveBeenCalled() + }) + + it("add: publishes the new file through the guarded saveDirectly with create kind", async () => { + mockedFileExistsAtPath.mockResolvedValueOnce(false) + + await tool.execute({ patch: addPatch }, mockTask as Task, { + askApproval: mockAskApproval, + handleError: mockHandleError, + pushToolResult: mockPushToolResult, + }) + + expect(mockSaveDirectly).toHaveBeenCalledWith( + "src/new.ts", + "new line one\nnew line two\n", + true, + true, + 1000, + "create", + ) + expect(mockPushToolResult).toHaveBeenCalledWith("Saved file") + expect(mockTask.didEditFile).toBe(true) + expect(mockHandleError).not.toHaveBeenCalled() + }) + + it("move: publishes the destination through the guarded saveDirectly with create kind", async () => { + await tool.execute({ patch: movePatch }, mockTask as Task, { + askApproval: mockAskApproval, + handleError: mockHandleError, + pushToolResult: mockPushToolResult, + }) + + expect(mockSaveDirectly).toHaveBeenCalledWith( + "src/new.ts", + "modified file content\n", + false, + true, + 1000, + "create", + ) + expect(mockTask.didEditFile).toBe(true) + expect(mockHandleError).not.toHaveBeenCalled() + }) + + it("update: surfaces the unobserved-existing remediation as a tool error", async () => { + const guardError = new Error( + "File already exists at /workspace/project/src/thing.ts and was not read before this write -- read the file first, then retry.", + ) + mockSaveDirectly.mockRejectedValue(guardError) + + await tool.execute({ patch: updatePatch }, mockTask as Task, { + askApproval: mockAskApproval, + handleError: mockHandleError, + pushToolResult: mockPushToolResult, + }) + + expect(mockHandleError).toHaveBeenCalledWith("apply patch", guardError) + expect(vi.mocked(mockTask.diffViewProvider.reset)).toHaveBeenCalled() + expect(mockTask.didEditFile).toBe(false) + expect(mockPushToolResult).not.toHaveBeenCalledWith("Saved file") + }) + + it("update: surfaces the stale-version remediation as a tool error", async () => { + const guardError = new Error( + "Stale version -- the file changed since you read it (expected v1, current v2); re-read the file, then retry.", + ) + mockSaveDirectly.mockRejectedValue(guardError) + + await tool.execute({ patch: updatePatch }, mockTask as Task, { + askApproval: mockAskApproval, + handleError: mockHandleError, + pushToolResult: mockPushToolResult, + }) + + expect(mockHandleError).toHaveBeenCalledWith("apply patch", guardError) + expect(vi.mocked(mockTask.diffViewProvider.reset)).toHaveBeenCalled() + expect(mockTask.didEditFile).toBe(false) + }) }) diff --git a/src/core/tools/__tests__/editFileTool.spec.ts b/src/core/tools/__tests__/editFileTool.spec.ts index 1ff8d52a8d..b4430c410f 100644 --- a/src/core/tools/__tests__/editFileTool.spec.ts +++ b/src/core/tools/__tests__/editFileTool.spec.ts @@ -7,6 +7,8 @@ import { fileExistsAtPath } from "../../../utils/fs" import { isPathOutsideWorkspace } from "../../../utils/pathUtils" import { getReadablePath } from "../../../utils/path" import { ToolUse, ToolResponse, AskApproval, HandleError, PushToolResult } from "../../../shared/tools" +import { checkpointSave } from "../../checkpoints" +import { computeDiffStats } from "../../diff/stats" import { editFileTool } from "../EditFileTool" vi.mock("fs/promises", () => ({ @@ -56,7 +58,16 @@ vi.mock("../../../utils/path", () => ({ vi.mock("../../diff/stats", () => ({ sanitizeUnifiedDiff: vi.fn((diff) => diff), - computeDiffStats: vi.fn(() => ({ additions: 1, deletions: 1 })), + // The real computeDiffStats returns { added, removed } (DiffStats) — + // keep the mock faithful to the production shape. + computeDiffStats: vi.fn(() => ({ added: 1, removed: 1 })), +})) + +vi.mock("../../checkpoints", () => ({ + getCheckpointService: vi.fn(), + checkpointSave: vi.fn().mockResolvedValue(undefined), + checkpointRestore: vi.fn(), + checkpointDiff: vi.fn(), })) vi.mock("vscode", () => ({ @@ -115,7 +126,9 @@ describe("editFileTool", () => { getState: vi.fn().mockResolvedValue({ diagnosticsEnabled: true, writeDelayMs: 1000, - experiments: {}, + // Pin the legacy diff-editor path explicitly: the PREVENT_FOCUS_DISRUPTION default + // flipped to true (L2, plan #33), so these tests must opt out. + experiments: { preventFocusDisruption: false } as Record, }), }), } @@ -168,6 +181,11 @@ describe("editFileTool", () => { fileContent?: string isPartial?: boolean accessAllowed?: boolean + experiments?: Record + // Trial addendum: extra extension-state fields merged into the pinned state + // (perWriteCheckpoints, auto-approval settings, ...) so the helper's + // providerRef pin does not clobber per-test state. + state?: Record } = {}, ): Promise { const fileExists = options.fileExists ?? true @@ -178,6 +196,17 @@ describe("editFileTool", () => { mockedFileExistsAtPath.mockResolvedValue(fileExists) mockedFsReadFile.mockResolvedValue(fileContent) mockTask.rooIgnoreController.validateAccess.mockReturnValue(accessAllowed) + mockTask.providerRef.deref.mockReturnValue({ + getState: vi.fn().mockResolvedValue({ + diagnosticsEnabled: true, + writeDelayMs: 1000, + // Trial addendum: pin the legacy diff-editor path for the pre-existing suites; + // the L2 default flipped preventFocusDisruption to true. Tests exercising + // the chat-diff branch opt in via the experiments option. + experiments: options.experiments ?? { preventFocusDisruption: false }, + ...options.state, + }), + }) const nativeArgs: Record = { file_path: testFilePath, @@ -576,6 +605,34 @@ describe("editFileTool", () => { }) }) + describe("focus disruption default (L2: chat-diff is the default approval path)", () => { + it("saves via saveDirectly without opening the diff editor when no experiment value is stored", async () => { + mockAskApproval.mockResolvedValue(true) + // No stored experiment value: the default (flipped to true in L2) resolves + // to the chat-diff path. Trial addendum: the empty experiments object is + // threaded through the helper (its providerRef pin would otherwise clobber + // this test's stored-value scenario). + await executeEditFileTool({}, { experiments: {} }) + + expect(mockTask.diffViewProvider.saveDirectly).toHaveBeenCalled() + expect(mockTask.diffViewProvider.open).not.toHaveBeenCalled() + expect(mockTask.diffViewProvider.saveChanges).not.toHaveBeenCalled() + expect(mockTask.didEditFile).toBe(true) + }) + + it("saves via saveDirectly when the user has explicitly enabled the experiment", async () => { + mockAskApproval.mockResolvedValue(true) + // Explicit stored true: same chat-diff routing as the default. + // Trial addendum: threaded through the helper's experiments option. + await executeEditFileTool({}, { experiments: { preventFocusDisruption: true } }) + + expect(mockTask.diffViewProvider.saveDirectly).toHaveBeenCalled() + expect(mockTask.diffViewProvider.open).not.toHaveBeenCalled() + expect(mockTask.diffViewProvider.saveChanges).not.toHaveBeenCalled() + expect(mockTask.didEditFile).toBe(true) + }) + }) + describe("partial block handling", () => { it("handles partial block without errors after path stabilizes", async () => { // Path stabilization requires two consecutive calls with the same path @@ -687,6 +744,93 @@ describe("editFileTool", () => { }) }) + describe("guarded write (S4b, epic #1375)", () => { + const focusDisruption = { preventFocusDisruption: true } + + it("publishes an existing-file edit through saveDirectly with edit kind", async () => { + const result = await executeEditFileTool( + { old_string: "Line 2", new_string: "Modified Line 2" }, + { fileExists: true, fileContent: "Line 1\nLine 2\nLine 3", experiments: focusDisruption }, + ) + + expect(mockTask.diffViewProvider.saveDirectly).toHaveBeenCalledWith( + testFilePath, + "Line 1\nModified Line 2\nLine 3", + false, + true, + 1000, + "edit", + ) + expect(mockTask.diffViewProvider.saveChanges).not.toHaveBeenCalled() + expect(mockTask.didEditFile).toBe(true) + expect(result).toContain("Tool result message") + expect(mockHandleError).not.toHaveBeenCalled() + }) + + it("publishes new-file creation through saveDirectly with create kind", async () => { + await executeEditFileTool( + { old_string: "", new_string: "New file content" }, + { fileExists: false, experiments: focusDisruption }, + ) + + expect(mockTask.diffViewProvider.saveDirectly).toHaveBeenCalledWith( + testFilePath, + "New file content", + true, + true, + 1000, + "create", + ) + expect(mockTask.didEditFile).toBe(true) + expect(mockHandleError).not.toHaveBeenCalled() + }) + + it("surfaces the unobserved edit remediation as a tool error and publishes nothing", async () => { + const guardError = new Error("File not read yet -- read the file, then retry.") + mockTask.diffViewProvider.saveDirectly.mockRejectedValue(guardError) + + const result = await executeEditFileTool( + { old_string: "Line 2", new_string: "Modified Line 2" }, + { fileExists: true, fileContent: "Line 1\nLine 2\nLine 3", experiments: focusDisruption }, + ) + + expect(mockHandleError).toHaveBeenCalledWith("edit_file", guardError) + expect(result).toBeUndefined() + expect(mockTask.diffViewProvider.reset).toHaveBeenCalled() + expect(mockTask.didToolFailInCurrentTurn).toBe(true) + expect(mockTask.didEditFile).toBe(false) + }) + + it("surfaces the stale-version remediation as a tool error and publishes nothing", async () => { + const guardError = new Error( + "Stale version -- the file changed since you read it (expected v1, current v2); re-read the file, then retry.", + ) + mockTask.diffViewProvider.saveDirectly.mockRejectedValue(guardError) + + const result = await executeEditFileTool( + { old_string: "Line 2", new_string: "Modified Line 2" }, + { fileExists: true, fileContent: "Line 1\nLine 2\nLine 3", experiments: focusDisruption }, + ) + + expect(mockHandleError).toHaveBeenCalledWith("edit_file", guardError) + expect(result).toBeUndefined() + expect(mockTask.diffViewProvider.reset).toHaveBeenCalled() + expect(mockTask.didEditFile).toBe(false) + }) + + it("still fails a literal mismatch with the existing message before the guard runs", async () => { + const result = await executeEditFileTool( + { old_string: "NonExistent", new_string: "Whatever" }, + { fileExists: true, fileContent: "Line 1\nLine 2\nLine 3", experiments: focusDisruption }, + ) + + expect(result).toContain("No match found") + expect(result).toContain("") + expect(mockTask.diffViewProvider.saveDirectly).not.toHaveBeenCalled() + expect(mockHandleError).not.toHaveBeenCalled() + }) + }) + describe("CRLF normalization", () => { it("preserves CRLF line endings on output", async () => { const contentWithCRLF = "Line 1\r\nLine 2\r\nLine 3" @@ -774,4 +918,107 @@ describe("editFileTool", () => { expect(mockAskApproval).toHaveBeenCalled() }) }) + + describe("per-write checkpoints (B1)", () => { + const mockedCheckpointSave = checkpointSave as MockedFunction + + it("records one suppressed checkpoint after a successful edit (default-on)", async () => { + await executeEditFileTool({}) + + expect(mockTask.consecutiveMistakeCount).toBe(0) + expect(mockedCheckpointSave).toHaveBeenCalledOnce() + // B2: the write info threads the path, operation, and the approval + // diff stats into the checkpoint hook. B3a: the approval diff itself is + // threaded verbatim for the per-step change card. + expect(mockedCheckpointSave).toHaveBeenCalledWith(mockTask, false, true, { + path: testFilePath, + operation: "update", + diffStats: { additions: 1, deletions: 1 }, + diff: "mock-diff", + }) + }) + + it("does not record a checkpoint when perWriteCheckpoints is disabled", async () => { + // Trial addendum: thread the state through the helper (its providerRef pin + // would otherwise clobber this test's extension state). + await executeEditFileTool({}, { state: { perWriteCheckpoints: false } }) + + expect(mockTask.consecutiveMistakeCount).toBe(0) + expect(mockedCheckpointSave).not.toHaveBeenCalled() + }) + + it("does not record a checkpoint when the edit fails", async () => { + mockTask.diffViewProvider.saveChanges.mockRejectedValue(new Error("save failed")) + + await executeEditFileTool({}) + + expect(mockHandleError).toHaveBeenCalledWith("edit_file", expect.any(Error)) + expect(mockedCheckpointSave).not.toHaveBeenCalled() + }) + + it("records the checkpoint with a create operation for a new file", async () => { + await executeEditFileTool({ old_string: "", new_string: "New file content" }, { fileExists: false }) + + expect(mockTask.consecutiveMistakeCount).toBe(0) + expect(mockedCheckpointSave).toHaveBeenCalledOnce() + expect(mockedCheckpointSave).toHaveBeenCalledWith(mockTask, false, true, { + path: testFilePath, + operation: "create", + diffStats: { additions: 1, deletions: 1 }, + diff: "mock-diff", + }) + }) + + it("omits diff stats from the checkpoint write when the diff has no stats", async () => { + // A null approval diff produces no diffStats on the journal write. + // The diff itself is still threaded for the change card (B3a). + vi.mocked(computeDiffStats).mockReturnValueOnce(null) + + await executeEditFileTool({}) + + expect(mockTask.consecutiveMistakeCount).toBe(0) + expect(mockedCheckpointSave).toHaveBeenCalledOnce() + expect(mockedCheckpointSave).toHaveBeenCalledWith(mockTask, false, true, { + path: testFilePath, + operation: "update", + diff: "mock-diff", + }) + }) + + it("threads autoApproved into the checkpoint write for auto-approved steps", async () => { + // B3a: when the step is auto-approved the checkpoint write carries + // autoApproved so checkpointSave can force the compact change card. + // Trial addendum: thread the auto-approval state through the helper. + await executeEditFileTool({}, { state: { autoApprovalEnabled: true, alwaysAllowWrite: true } }) + + expect(mockedCheckpointSave).toHaveBeenCalledWith(mockTask, false, true, { + path: testFilePath, + operation: "update", + diffStats: { additions: 1, deletions: 1 }, + diff: "mock-diff", + autoApproved: true, + }) + }) + + it("awaits the edit checkpoint before execute settles", async () => { + type SaveResult = Awaited> + let resolveSave: (value: SaveResult | PromiseLike) => void = () => {} + const saveDeferred = new Promise((resolve) => (resolveSave = resolve)) + mockedCheckpointSave.mockImplementationOnce(() => saveDeferred) + + const executePromise = executeEditFileTool({}) + + // execute must not settle while the checkpoint is still in flight: + // a later tool block would otherwise interleave with this edit's staged work. + let settled = false + void executePromise.finally(() => (settled = true)) + await new Promise((resolve) => setTimeout(resolve, 0)) + expect(mockedCheckpointSave).toHaveBeenCalledOnce() + expect(settled).toBe(false) + + resolveSave() + await executePromise + expect(settled).toBe(true) + }) + }) }) diff --git a/src/core/tools/__tests__/editSearchReplaceTool.changeCard.spec.ts b/src/core/tools/__tests__/editSearchReplaceTool.changeCard.spec.ts new file mode 100644 index 0000000000..3397c1450f --- /dev/null +++ b/src/core/tools/__tests__/editSearchReplaceTool.changeCard.spec.ts @@ -0,0 +1,243 @@ +// npx vitest run core/tools/__tests__/editSearchReplaceTool.changeCard.spec.ts + +import type { MockedFunction } from "vitest" + +import { fileExistsAtPath } from "../../../utils/fs" +import { checkAutoApproval } from "../../auto-approval" +import { checkpointSave } from "../../checkpoints" +import type { Task } from "../../task/Task" +import { EditTool } from "../EditTool" +import type { ToolCallbacks } from "../BaseTool" +import { SearchReplaceTool } from "../SearchReplaceTool" + +/** + * The shared surface of the two string-replacement edit tools: both execute a + * single `old_string` -> `new_string` replacement and report through the same + * ToolCallbacks, which is all the change-card tests exercise. + */ +interface EditLikeTool { + execute( + params: { file_path: string; old_string: string; new_string: string }, + task: Task, + callbacks: ToolCallbacks, + ): Promise +} + +vi.mock("fs/promises", () => ({ + default: { + // Contains exactly one occurrence of the old_string below. + readFile: vi.fn().mockResolvedValue("old line\n"), + }, +})) + +vi.mock("../../../utils/fs", () => ({ + fileExistsAtPath: vi.fn().mockResolvedValue(true), +})) + +vi.mock("../../prompts/responses", () => ({ + formatResponse: { + toolError: vi.fn((msg: string) => `Error: ${msg}`), + rooIgnoreError: vi.fn((filePath: string) => `Access denied: ${filePath}`), + createPrettyPatch: vi.fn(() => "mock-diff"), + }, +})) + +vi.mock("../../diff/stats", () => ({ + // The real DiffStats shape is { added, removed } (the tool maps it to the + // change-card { additions, deletions } pair). + sanitizeUnifiedDiff: vi.fn((diff: string) => diff), + computeDiffStats: vi.fn(() => ({ added: 1, removed: 1 })), +})) + +vi.mock("../../checkpoints", () => ({ + checkpointSave: vi.fn().mockResolvedValue(undefined), +})) + +vi.mock("../../auto-approval", () => ({ + checkAutoApproval: vi.fn().mockResolvedValue({ decision: "ask" }), +})) + +interface Stubs { + mockTask: Pick< + Task, + | "cwd" + | "consecutiveMistakeCount" + | "recordToolError" + | "rooIgnoreController" + | "rooProtectedController" + | "processQueuedMessages" + | "didEditFile" + | "diffViewProvider" + | "providerRef" + | "fileContextTracker" + > + mockSaveDirectly: MockedFunction<(...args: unknown[]) => Promise> + mockGetState: MockedFunction<() => Promise>> +} + +/** + * Structural stubs for the prevent-focus-disruption save path: the real + * DiffViewProvider is out of scope here, so vi.fn() doubles stand in for the + * members the edit tools touch. + */ +function buildStubs(): Stubs { + const mockSaveDirectly = vi.fn().mockResolvedValue({ + newProblemsMessage: "", + userEdits: undefined, + finalContent: "new line\n", + }) + const diffViewProviderStub = { + editType: undefined as "create" | "modify" | undefined, + originalContent: undefined as string | undefined, + open: vi.fn().mockResolvedValue(undefined), + update: vi.fn().mockResolvedValue(undefined), + scrollToFirstDiff: vi.fn(), + saveDirectly: mockSaveDirectly, + saveChanges: vi.fn().mockResolvedValue({ + newProblemsMessage: "", + userEdits: undefined, + finalContent: "new line\n", + }), + revertChanges: vi.fn().mockResolvedValue(undefined), + pushToolWriteResult: vi.fn().mockResolvedValue("Saved file"), + reset: vi.fn().mockResolvedValue(undefined), + } + const mockGetState = vi.fn().mockResolvedValue({ + diagnosticsEnabled: true, + writeDelayMs: 1000, + // Exercise the focus-disruption (saveDirectly) save path. + experiments: { preventFocusDisruption: true }, + }) + const mockTask: Stubs["mockTask"] = { + cwd: "/workspace/project", + consecutiveMistakeCount: 0, + recordToolError: vi.fn(), + rooIgnoreController: { + validateAccess: vi.fn().mockReturnValue(true), + } as unknown as Task["rooIgnoreController"], + rooProtectedController: { + isWriteProtected: vi.fn().mockReturnValue(false), + } as unknown as Task["rooProtectedController"], + processQueuedMessages: vi.fn(), + didEditFile: false, + diffViewProvider: diffViewProviderStub as unknown as Task["diffViewProvider"], + providerRef: { + deref: vi.fn().mockReturnValue({ + getState: mockGetState, + }), + } as unknown as Task["providerRef"], + fileContextTracker: { + trackFileContext: vi.fn().mockResolvedValue(undefined), + } as unknown as Task["fileContextTracker"], + } + return { mockTask, mockSaveDirectly, mockGetState } +} + +function cardTests(getTool: () => EditLikeTool) { + const mockedFileExistsAtPath = fileExistsAtPath as MockedFunction + const mockCheckpointSave = checkpointSave as MockedFunction + const mockCheckAutoApproval = checkAutoApproval as MockedFunction + + let tool: EditLikeTool + let stubs: Stubs + let mockAskApproval: MockedFunction<(...args: unknown[]) => Promise> + let mockHandleError: MockedFunction<(...args: unknown[]) => Promise> + let mockPushToolResult: MockedFunction<(...args: unknown[]) => void> + + beforeEach(() => { + vi.clearAllMocks() + mockedFileExistsAtPath.mockResolvedValue(true) + stubs = buildStubs() + tool = getTool() + mockAskApproval = vi.fn().mockResolvedValue(true) + mockHandleError = vi.fn().mockResolvedValue(undefined) + mockPushToolResult = vi.fn() + }) + + it("records a per-write checkpoint with the approval diff after a successful write", async () => { + await tool.execute( + { file_path: "src/thing.ts", old_string: "old line", new_string: "new line" }, + stubs.mockTask as Task, + { + askApproval: mockAskApproval, + handleError: mockHandleError, + pushToolResult: mockPushToolResult, + }, + ) + + expect(mockCheckpointSave).toHaveBeenCalledTimes(1) + expect(mockCheckpointSave).toHaveBeenCalledWith(stubs.mockTask, false, true, { + path: "src/thing.ts", + operation: "update", + diffStats: { additions: 1, deletions: 1 }, + diff: "mock-diff", + }) + }) + + it("marks auto-approved steps so the card renders compact", async () => { + mockCheckAutoApproval.mockResolvedValueOnce({ decision: "approve" }) + + await tool.execute( + { file_path: "src/thing.ts", old_string: "old line", new_string: "new line" }, + stubs.mockTask as Task, + { + askApproval: mockAskApproval, + handleError: mockHandleError, + pushToolResult: mockPushToolResult, + }, + ) + + expect(mockCheckpointSave).toHaveBeenCalledTimes(1) + expect(mockCheckpointSave.mock.calls[0]?.[3]).toEqual(expect.objectContaining({ autoApproved: true })) + }) + + it("skips the checkpoint when perWriteCheckpoints is explicitly disabled", async () => { + stubs.mockGetState.mockResolvedValueOnce({ + diagnosticsEnabled: true, + writeDelayMs: 1000, + experiments: { preventFocusDisruption: true }, + perWriteCheckpoints: false, + }) + + await tool.execute( + { file_path: "src/thing.ts", old_string: "old line", new_string: "new line" }, + stubs.mockTask as Task, + { + askApproval: mockAskApproval, + handleError: mockHandleError, + pushToolResult: mockPushToolResult, + }, + ) + + expect(mockCheckpointSave).not.toHaveBeenCalled() + // The write itself still happens (the setting gates the checkpoint only), + // and auto-approval is never consulted when there is no card to build. + expect(stubs.mockSaveDirectly).toHaveBeenCalled() + expect(mockCheckAutoApproval).not.toHaveBeenCalled() + }) + + it("records nothing when the approval is declined", async () => { + mockAskApproval.mockResolvedValue(false) + + await tool.execute( + { file_path: "src/thing.ts", old_string: "old line", new_string: "new line" }, + stubs.mockTask as Task, + { + askApproval: mockAskApproval, + handleError: mockHandleError, + pushToolResult: mockPushToolResult, + }, + ) + + expect(mockCheckpointSave).not.toHaveBeenCalled() + expect(stubs.mockSaveDirectly).not.toHaveBeenCalled() + }) +} + +describe("EditTool.execute - per-write checkpoint and change card (B3a, epic #1375)", () => { + cardTests(() => new EditTool()) +}) + +describe("SearchReplaceTool.execute - per-write checkpoint and change card (B3a, epic #1375)", () => { + cardTests(() => new SearchReplaceTool()) +}) diff --git a/src/core/tools/__tests__/editTool.spec.ts b/src/core/tools/__tests__/editTool.spec.ts index a5f665b9e5..09014867bf 100644 --- a/src/core/tools/__tests__/editTool.spec.ts +++ b/src/core/tools/__tests__/editTool.spec.ts @@ -59,6 +59,14 @@ vi.mock("../../diff/stats", () => ({ computeDiffStats: vi.fn(() => ({ additions: 1, deletions: 1 })), })) +vi.mock("../../checkpoints", () => ({ + checkpointSave: vi.fn().mockResolvedValue(undefined), +})) + +vi.mock("../../auto-approval", () => ({ + checkAutoApproval: vi.fn().mockResolvedValue({ decision: "ask" }), +})) + vi.mock("vscode", () => ({ window: { showWarningMessage: vi.fn().mockResolvedValue(undefined), @@ -111,7 +119,9 @@ describe("editTool", () => { getState: vi.fn().mockResolvedValue({ diagnosticsEnabled: true, writeDelayMs: 1000, - experiments: {}, + // Pin the legacy diff-editor path explicitly: the PREVENT_FOCUS_DISRUPTION default + // flipped to true (L2, plan #33), so these tests must opt out. + experiments: { preventFocusDisruption: false } as Record, }), }), } @@ -169,6 +179,10 @@ describe("editTool", () => { fileContent?: string isPartial?: boolean accessAllowed?: boolean + experiments?: Record + // Trial addendum: extra extension-state fields merged into the pinned state + // so the helper's providerRef pin does not clobber per-test state. + state?: Record } = {}, ): Promise { const fileExists = options.fileExists ?? true @@ -179,6 +193,17 @@ describe("editTool", () => { mockedFileExistsAtPath.mockResolvedValue(fileExists) mockedFsReadFile.mockResolvedValue(fileContent) mockTask.rooIgnoreController.validateAccess.mockReturnValue(accessAllowed) + mockTask.providerRef.deref.mockReturnValue({ + getState: vi.fn().mockResolvedValue({ + diagnosticsEnabled: true, + writeDelayMs: 1000, + // Trial addendum: pin the legacy diff-editor path for the pre-existing suites; + // the L2 default flipped preventFocusDisruption to true. Tests exercising + // the chat-diff branch opt in via the experiments option. + experiments: options.experiments ?? { preventFocusDisruption: false }, + ...options.state, + }), + }) const defaultParams = { file_path: testFilePath, @@ -361,6 +386,33 @@ describe("editTool", () => { }) }) + describe("focus disruption default (L2: chat-diff is the default approval path)", () => { + it("saves via saveDirectly without opening the diff editor when no experiment value is stored", async () => { + mockAskApproval.mockResolvedValue(true) + // No stored experiment value: the default (flipped to true in L2) resolves + // to the chat-diff path. Trial addendum: the empty experiments object is + // threaded through the helper (its providerRef pin would otherwise clobber + // this test's stored-value scenario). + await executeEditTool({}, { experiments: {} }) + + expect(mockTask.diffViewProvider.saveDirectly).toHaveBeenCalled() + expect(mockTask.diffViewProvider.open).not.toHaveBeenCalled() + expect(mockTask.diffViewProvider.saveChanges).not.toHaveBeenCalled() + expect(mockTask.didEditFile).toBe(true) + }) + + it("still uses the diff-editor path when the user has opted out (stored false)", async () => { + mockAskApproval.mockResolvedValue(true) + // Base mock pins experiments: { preventFocusDisruption: false } (legacy path). + + await executeEditTool() + + expect(mockTask.diffViewProvider.open).toHaveBeenCalled() + expect(mockTask.diffViewProvider.saveChanges).toHaveBeenCalled() + expect(mockTask.diffViewProvider.saveDirectly).not.toHaveBeenCalled() + }) + }) + describe("partial block handling", () => { it("handles partial block without errors after path stabilizes", async () => { // Path stabilization requires two consecutive calls with the same path @@ -424,4 +476,42 @@ describe("editTool", () => { expect(mockTask.fileContextTracker.trackFileContext).toHaveBeenCalledWith(testFilePath, "roo_edited") }) }) + + describe("guarded write (S4b, epic #1375)", () => { + const focusDisruption = { preventFocusDisruption: true } + + it("publishes through saveDirectly with edit kind", async () => { + const result = await executeEditTool( + { old_string: "Line 2", new_string: "Modified Line 2" }, + { fileContent: "Line 1\nLine 2\nLine 3", experiments: focusDisruption }, + ) + + expect(mockTask.diffViewProvider.saveDirectly).toHaveBeenCalledWith( + testFilePath, + "Line 1\nModified Line 2\nLine 3", + false, + true, + 1000, + "edit", + ) + expect(mockTask.didEditFile).toBe(true) + expect(result).toBe("Tool result message") + expect(mockHandleError).not.toHaveBeenCalled() + }) + + it("surfaces the unobserved edit remediation as a tool error and publishes nothing", async () => { + const guardError = new Error("File not read yet -- read the file, then retry.") + mockTask.diffViewProvider.saveDirectly.mockRejectedValue(guardError) + + const result = await executeEditTool( + { old_string: "Line 2", new_string: "Modified Line 2" }, + { fileContent: "Line 1\nLine 2\nLine 3", experiments: focusDisruption }, + ) + + expect(mockHandleError).toHaveBeenCalledWith("edit", guardError) + expect(result).toBeUndefined() + expect(mockTask.diffViewProvider.reset).toHaveBeenCalled() + expect(mockTask.didEditFile).toBe(false) + }) + }) }) diff --git a/src/core/tools/__tests__/guardedWrite.spec.ts b/src/core/tools/__tests__/guardedWrite.spec.ts new file mode 100644 index 0000000000..122c1c6d9b --- /dev/null +++ b/src/core/tools/__tests__/guardedWrite.spec.ts @@ -0,0 +1,454 @@ +/** + * Tests for the guarded-write compare-and-swap core (upstream epic #1375, + * phase A4a). + * + * Covers guard selection through the S2 observation registry, version-token + * CAS, remediation messages, and the per-absolute-path FIFO chain: FIFO + * ordering, exactly-one winner under concurrency, no wedge after a rejected + * link, and independence across paths. + */ + +import * as fs from "fs/promises" +import * as path from "path" + +import { describe, expect, it, beforeEach, vi } from "vitest" + +import { createIfAbsent, guardedWrite, replaceIfVersion, resetChain } from "../guardedWrite" +import { safeWriteText } from "../../../services/file-safety/safeWriteText" +import { computeVersionToken } from "../../../utils/versionToken" +import { ObservationRegistry } from "../../task/observationRegistry" +import type { Task } from "../../task/Task" + +// -- Mocks ------------------------------------------------------------------- + +vi.mock("fs/promises", () => ({ + access: vi.fn(), + stat: vi.fn(), +})) + +vi.mock("../../../utils/versionToken", () => ({ + computeVersionToken: vi.fn(), +})) + +vi.mock("../../../services/file-safety/safeWriteText", () => ({ + safeWriteText: vi.fn(), +})) + +const mockedFsAccess = vi.mocked(fs.access) +const mockedComputeVersionToken = vi.mocked(computeVersionToken) +const mockedSafeWriteText = vi.mocked(safeWriteText) + +// -- Fixtures ---------------------------------------------------------------- + +const WORKSPACE = "/test/workspace" + +/** Resolve a fixture path the same way guardedWrite resolves task.cwd-relative paths. */ +const abs = (relPath: string): string => path.resolve(WORKSPACE, relPath) + +interface MockTaskOptions { + cwd?: string + observationRegistry?: ObservationRegistry +} + +/** + * Minimal structural Task: guardedWrite only reads task.cwd and + * task.observationRegistry. The real Task constructor needs the full provider + * machinery, so a single documented double cast stands in for the class. + */ +function createMockTask(options: MockTaskOptions = {}): Task { + const task = { + cwd: options.cwd ?? WORKSPACE, + observationRegistry: options.observationRegistry ?? new ObservationRegistry(), + } + return task as unknown as Task +} + +// -- Tests ------------------------------------------------------------------- + +describe("guardedWrite (S4a, epic #1375)", () => { + beforeEach(() => { + vi.resetAllMocks() + resetChain() + }) + + describe("unobserved create", () => { + it("succeeds when the file is absent and publishes via safeWriteText", async () => { + mockedFsAccess.mockRejectedValue({ code: "ENOENT" }) + const task = createMockTask() + + await guardedWrite(task, "new-file.txt", "hello", "create") + + expect(mockedSafeWriteText).toHaveBeenCalledTimes(1) + expect(mockedSafeWriteText).toHaveBeenCalledWith(abs("new-file.txt"), "hello") + }) + + it("fails with the read-first remediation when the file exists - nothing published", async () => { + mockedFsAccess.mockResolvedValue(undefined) + const task = createMockTask() + + await expect(guardedWrite(task, "existing.txt", "hello", "create")).rejects.toThrow( + "File already exists at " + + abs("existing.txt") + + " and was not read before this write -- read the file first, then retry.", + ) + expect(mockedSafeWriteText).not.toHaveBeenCalled() + }) + + it("rethrows I/O errors that are not ENOENT verbatim (no guard verdict on access failure)", async () => { + const failures = [{ code: "EACCES" }, null, "volume offline", new Error("EIO-ish failure")] + for (const failure of failures) { + mockedFsAccess.mockRejectedValueOnce(failure) + await expect(createIfAbsent(abs("io-error.txt"), "x")).rejects.toBe(failure) + } + expect(mockedSafeWriteText).not.toHaveBeenCalled() + }) + }) + + describe("deleted-after-read target", () => { + it("normalizes an ENOENT from the version token into the re-read remediation", async () => { + const reg = new ObservationRegistry() + reg.observe(abs("vanished.txt"), "v1") + const task = createMockTask({ observationRegistry: reg }) + + // The file was deleted after the read: the token computation fails + // with a raw ENOENT, which the guard must convert into the standard + // re-read-then-retry contract. + mockedComputeVersionToken.mockRejectedValue({ code: "ENOENT" }) + + await expect(guardedWrite(task, "vanished.txt", "next", "update")).rejects.toThrow( + "File was deleted after it was read", + ) + expect(mockedSafeWriteText).not.toHaveBeenCalled() + }) + + it("rethrows non-ENOENT token failures verbatim from replaceIfVersion", async () => { + const failure = { code: "EACCES" } + mockedComputeVersionToken.mockRejectedValueOnce(failure) + + await expect(replaceIfVersion(abs("locked.txt"), "v1", "next")).rejects.toBe(failure) + expect(mockedSafeWriteText).not.toHaveBeenCalled() + }) + }) + describe("unobserved update", () => { + it("succeeds when the file is absent (same create guard)", async () => { + mockedFsAccess.mockRejectedValue({ code: "ENOENT" }) + const task = createMockTask() + + await guardedWrite(task, "new-file.txt", "hello", "update") + + expect(mockedSafeWriteText).toHaveBeenCalledWith(abs("new-file.txt"), "hello") + }) + + it("fails with the read-first remediation when the file exists - nothing published", async () => { + mockedFsAccess.mockResolvedValue(undefined) + const task = createMockTask() + + await expect(guardedWrite(task, "existing.txt", "hello", "update")).rejects.toThrow( + "File already exists at " + + abs("existing.txt") + + " and was not read before this write -- read the file first, then retry.", + ) + expect(mockedSafeWriteText).not.toHaveBeenCalled() + }) + }) + + describe("observed create", () => { + it("recreates a file that vanished after the read", async () => { + const reg = new ObservationRegistry() + reg.observe(abs("gone.txt"), "v1") + mockedFsAccess.mockRejectedValue({ code: "ENOENT" }) + const task = createMockTask({ observationRegistry: reg }) + + await guardedWrite(task, "gone.txt", "back", "create") + + expect(mockedSafeWriteText).toHaveBeenCalledTimes(1) + expect(mockedSafeWriteText).toHaveBeenCalledWith(abs("gone.txt"), "back") + }) + + it("goes through the version guard when the file still exists", async () => { + const reg = new ObservationRegistry() + reg.observe(abs("kept.txt"), "v1") + mockedFsAccess.mockResolvedValue(undefined) + mockedComputeVersionToken.mockResolvedValue("v1") + const task = createMockTask({ observationRegistry: reg }) + + await guardedWrite(task, "kept.txt", "rewritten", "create") + + expect(mockedSafeWriteText).toHaveBeenCalledWith(abs("kept.txt"), "rewritten") + }) + + it("fails with the stale remediation suffix when the version moved", async () => { + const reg = new ObservationRegistry() + reg.observe(abs("kept.txt"), "v1") + mockedFsAccess.mockResolvedValue(undefined) + mockedComputeVersionToken.mockResolvedValue("v2") + const task = createMockTask({ observationRegistry: reg }) + + await expect(guardedWrite(task, "kept.txt", "rewritten", "create")).rejects.toThrow( + "Stale version -- the file changed since you read it (expected v1, current v2); re-read the file, then retry.", + ) + expect(mockedSafeWriteText).not.toHaveBeenCalled() + }) + + it("defers to the version guard when the access check is denied (not ENOENT)", async () => { + const reg = new ObservationRegistry() + reg.observe(abs("locked.txt"), "v1") + mockedFsAccess.mockRejectedValue({ code: "EACCES" }) + mockedComputeVersionToken.mockResolvedValue("v2") + const task = createMockTask({ observationRegistry: reg }) + + await expect(guardedWrite(task, "locked.txt", "rewritten", "create")).rejects.toThrow( + "Stale version -- the file changed since you read it (expected v1, current v2); re-read the file, then retry.", + ) + expect(mockedSafeWriteText).not.toHaveBeenCalled() + }) + }) + + describe("observed update (version CAS)", () => { + it("publishes when the on-disk version matches the observation", async () => { + const reg = new ObservationRegistry() + reg.observe(abs("doc.txt"), "v1") + mockedComputeVersionToken.mockResolvedValue("v1") + const task = createMockTask({ observationRegistry: reg }) + + await guardedWrite(task, "doc.txt", "new content", "update") + + expect(mockedComputeVersionToken).toHaveBeenCalledWith(abs("doc.txt")) + expect(mockedSafeWriteText).toHaveBeenCalledWith(abs("doc.txt"), "new content") + }) + + it("fails with the stale remediation suffix when the version moved - nothing published", async () => { + const reg = new ObservationRegistry() + reg.observe(abs("doc.txt"), "v1") + mockedComputeVersionToken.mockResolvedValue("v2") + const task = createMockTask({ observationRegistry: reg }) + + await expect(guardedWrite(task, "doc.txt", "new content", "update")).rejects.toThrow( + "Stale version -- the file changed since you read it (expected v1, current v2); re-read the file, then retry.", + ) + expect(mockedSafeWriteText).not.toHaveBeenCalled() + }) + }) + + describe("edit", () => { + it("fails read-first when the file was never observed - nothing published, no I/O", async () => { + const task = createMockTask() + + await expect(guardedWrite(task, "any.txt", "patched", "edit")).rejects.toThrow( + "File not read yet -- read the file, then retry.", + ) + expect(mockedSafeWriteText).not.toHaveBeenCalled() + expect(mockedComputeVersionToken).not.toHaveBeenCalled() + expect(mockedFsAccess).not.toHaveBeenCalled() + }) + + it("publishes when the version matches the observation", async () => { + const reg = new ObservationRegistry() + reg.observe(abs("doc.txt"), "v1") + mockedComputeVersionToken.mockResolvedValue("v1") + const task = createMockTask({ observationRegistry: reg }) + + await guardedWrite(task, "doc.txt", "patched", "edit") + + expect(mockedSafeWriteText).toHaveBeenCalledWith(abs("doc.txt"), "patched") + }) + + it("fails with the stale remediation suffix when the version moved", async () => { + const reg = new ObservationRegistry() + reg.observe(abs("doc.txt"), "v1") + mockedComputeVersionToken.mockResolvedValue("v3") + const task = createMockTask({ observationRegistry: reg }) + + await expect(guardedWrite(task, "doc.txt", "patched", "edit")).rejects.toThrow( + "Stale version -- the file changed since you read it (expected v1, current v3); re-read the file, then retry.", + ) + expect(mockedSafeWriteText).not.toHaveBeenCalled() + }) + }) + + describe("concurrency: per-path FIFO chain", () => { + it("two concurrent updates on one path - exactly one publishes, the other fails stale", async () => { + const reg = new ObservationRegistry() + reg.observe(abs("shared.txt"), "v1") + mockedComputeVersionToken.mockResolvedValue("v1") + const task = createMockTask({ observationRegistry: reg }) + + // The first publish changes the on-disk state (new token). + mockedSafeWriteText.mockImplementation(async () => { + mockedComputeVersionToken.mockResolvedValue("v2") + }) + + const p1 = guardedWrite(task, "shared.txt", "first", "update") + const p2 = guardedWrite(task, "shared.txt", "second", "update") + const [r1, r2] = await Promise.allSettled([p1, p2]) + + if (r1.status !== "fulfilled" || r2.status !== "rejected") { + throw new Error("expected exactly one publish, got " + r1.status + " / " + r2.status) + } + expect(mockedSafeWriteText).toHaveBeenCalledTimes(1) + expect(r2.reason.message).toBe( + "Stale version -- the file changed since you read it (expected v1, current v2); re-read the file, then retry.", + ) + }) + + it("observed-absent then two concurrent creates - the second fails stale", async () => { + const reg = new ObservationRegistry() + reg.observe(abs("absent.txt"), "v1") // read before, file later vanished + mockedFsAccess.mockRejectedValue({ code: "ENOENT" }) + const task = createMockTask({ observationRegistry: reg }) + + let publishes = 0 + mockedSafeWriteText.mockImplementation(async () => { + publishes += 1 + if (publishes === 1) { + // After the first publish the file exists again under a new token. + mockedFsAccess.mockResolvedValue(undefined) + mockedComputeVersionToken.mockResolvedValue("v2") + } + }) + + const p1 = guardedWrite(task, "absent.txt", "first", "create") + const p2 = guardedWrite(task, "absent.txt", "second", "create") + const [r1, r2] = await Promise.allSettled([p1, p2]) + + if (r1.status !== "fulfilled" || r2.status !== "rejected") { + throw new Error("expected exactly one publish, got " + r1.status + " / " + r2.status) + } + expect(publishes).toBe(1) + expect(r2.reason.message).toContain("Stale version") + expect(r2.reason.message).toContain("re-read the file, then retry.") + }) + + it("the chain settles after a rejection - a later matching write still runs", async () => { + const reg = new ObservationRegistry() + reg.observe(abs("settle.txt"), "v1") + mockedComputeVersionToken.mockResolvedValue("v2") // already stale at v1 + const task = createMockTask({ observationRegistry: reg }) + + const p1 = guardedWrite(task, "settle.txt", "first", "update") + await expect(p1).rejects.toThrow("Stale version") + + // No resetChain: the rejected link must not wedge the chain. The + // caller re-reads the file (observation refreshed to v2) and retries. + reg.observe(abs("settle.txt"), "v2") + const p2 = guardedWrite(task, "settle.txt", "second", "update") + await expect(p2).resolves.toBeUndefined() + + expect(mockedSafeWriteText).toHaveBeenCalledTimes(1) + expect(mockedSafeWriteText).toHaveBeenCalledWith(abs("settle.txt"), "second") + }) + + it("evicts settled chain entries - a later write still serializes in order", async () => { + const reg = new ObservationRegistry() + reg.observe(abs("evict.txt"), "v1") + mockedComputeVersionToken.mockResolvedValue("v1") + const task = createMockTask({ observationRegistry: reg }) + + // A first write settles; its chain entry is evicted with it. + const p1 = guardedWrite(task, "evict.txt", "first", "update") + await expect(p1).resolves.toBeUndefined() + + // Two rapid writes submitted after the eviction must still run one + // at a time in submission order (the eviction must not drop the + // chain for in-flight or just-enqueued links). + const order: string[] = [] + mockedSafeWriteText.mockImplementation(async (_path: string, content: string) => { + order.push(content) + }) + const p2 = guardedWrite(task, "evict.txt", "second", "update") + const p3 = guardedWrite(task, "evict.txt", "third", "update") + await Promise.all([p2, p3]) + + expect(order).toEqual(["second", "third"]) + // Three publishes in total: the settled first write plus the two + // serialized rapid writes. + expect(mockedSafeWriteText).toHaveBeenCalledTimes(3) + }) + + it("writes on different paths are independent (no cross-path serialization)", async () => { + const reg = new ObservationRegistry() + reg.observe(abs("a.txt"), "v1") + reg.observe(abs("b.txt"), "v1") + mockedComputeVersionToken.mockResolvedValue("v1") + const task = createMockTask({ observationRegistry: reg }) + + const p1 = guardedWrite(task, "a.txt", "a", "update") + const p2 = guardedWrite(task, "b.txt", "b", "update") + await Promise.all([p1, p2]) + + expect(mockedSafeWriteText).toHaveBeenCalledTimes(2) + }) + }) + + describe("path resolution", () => { + it("resolves a relative path against task.cwd", async () => { + const reg = new ObservationRegistry() + reg.observe(abs("sub/dir.txt"), "v1") + mockedComputeVersionToken.mockResolvedValue("v1") + const task = createMockTask({ observationRegistry: reg }) + + await guardedWrite(task, "sub/dir.txt", "content", "update") + + expect(mockedSafeWriteText).toHaveBeenCalledWith(abs("sub/dir.txt"), "content") + }) + + it("normalizes an already-absolute input (trailing separator) to the observation key", async () => { + const reg = new ObservationRegistry() + const canonical = abs("sub/dir.txt") + // ReadFileTool observes under path.resolve(task.cwd, relPath) — the + // canonical spelling. A write addressed with a trailing separator used + // to bypass the observation (isAbsolute passthrough) and fail + // "File already exists" / "File not read yet" for a file that was read. + reg.observe(canonical, "v1") + mockedComputeVersionToken.mockResolvedValue("v1") + const task = createMockTask({ observationRegistry: reg }) + + await guardedWrite(task, canonical + "/", "content", "update") + + expect(mockedSafeWriteText).toHaveBeenCalledTimes(1) + expect(mockedSafeWriteText).toHaveBeenCalledWith(canonical, "content") + }) + + it("serializes two spellings of one file through a single chain key", async () => { + const reg = new ObservationRegistry() + const canonical = abs("shared2.txt") + reg.observe(canonical, "v1") + mockedComputeVersionToken.mockResolvedValue("v1") + const task = createMockTask({ observationRegistry: reg }) + + // The first publish changes the on-disk state (new token). + mockedSafeWriteText.mockImplementation(async () => { + mockedComputeVersionToken.mockResolvedValue("v2") + }) + + // Plain spelling vs the trailing-separator spelling: with one chain key + // they are strictly ordered (first matches v1, second sees v2). + const p1 = guardedWrite(task, canonical, "first", "update") + const p2 = guardedWrite(task, canonical + "/", "second", "update") + const [r1, r2] = await Promise.allSettled([p1, p2]) + + if (r1.status !== "fulfilled" || r2.status !== "rejected") { + throw new Error("expected exactly one publish, got " + r1.status + " / " + r2.status) + } + expect(mockedSafeWriteText).toHaveBeenCalledTimes(1) + expect(r2.reason.message).toBe( + "Stale version -- the file changed since you read it (expected v1, current v2); re-read the file, then retry.", + ) + }) + }) + + describe("resetChain", () => { + it("detaches pending links so later writes start a fresh chain", async () => { + const reg = new ObservationRegistry() + reg.observe(abs("x.txt"), "v1") + mockedComputeVersionToken.mockResolvedValue("v1") + const task = createMockTask({ observationRegistry: reg }) + + await guardedWrite(task, "x.txt", "a", "update") + resetChain() + await guardedWrite(task, "x.txt", "b", "update") + + expect(mockedSafeWriteText).toHaveBeenLastCalledWith(abs("x.txt"), "b") + }) + }) +}) diff --git a/src/core/tools/__tests__/readFileTool.spec.ts b/src/core/tools/__tests__/readFileTool.spec.ts index 6c9e177d38..7e2fa3aac7 100644 --- a/src/core/tools/__tests__/readFileTool.spec.ts +++ b/src/core/tools/__tests__/readFileTool.spec.ts @@ -13,10 +13,16 @@ */ import path from "path" +import type { Stats } from "fs" + +import type { LegacyReadFileParams } from "@roo-code/types" import { isBinaryFile } from "isbinaryfile" import { readFileTool, ReadFileTool } from "../ReadFileTool" +import type { Task } from "../../task/Task" +import { ObservationRegistry } from "../../task/observationRegistry" +import { computeVersionToken } from "../../../utils/versionToken" import { formatResponse } from "../../prompts/responses" import { validateImageForProcessing, @@ -136,6 +142,7 @@ interface MockTaskOptions { rooIgnoreAllowed?: boolean maxImageFileSize?: number maxTotalImageSize?: number + observationRegistry?: ObservationRegistry } function createMockTask(options: MockTaskOptions = {}) { @@ -143,6 +150,9 @@ function createMockTask(options: MockTaskOptions = {}) { return { cwd: "/test/workspace", + // Mirror Task: every task always owns an observation registry (A2, #1375). + // Tests asserting on observations pass their own instance via options. + observationRegistry: options.observationRegistry ?? new ObservationRegistry(), api: { getModel: vi.fn().mockReturnValue({ info: { supportsImages }, @@ -187,7 +197,18 @@ describe("ReadFileTool", () => { vi.clearAllMocks() // Default mock implementations - mockedFsStat.mockResolvedValue({ isDirectory: () => false } as any) + // The stat default carries BigIntStats fields (A2, epic #1375): reads now + // token-ize the pre/post stats, so the default must look like a real bigint stat. + // Tests overriding it do so per-call with mockResolvedValue(Once). + mockedFsStat.mockResolvedValue({ + isDirectory: () => false, + dev: BigInt(1), + ino: BigInt(2), + size: BigInt(300), + mtimeNs: BigInt(4_000_000_000n), + ctimeNs: BigInt(5_000_000_000n), + // Cast: the mock only implements the members the tool and versionToken read. + } as unknown as Stats) mockedIsBinaryFile.mockResolvedValue(false) mockedFsReadFile.mockResolvedValue(Buffer.from("test content")) mockedReadWithSlice.mockReturnValue({ @@ -1489,5 +1510,306 @@ describe("ReadFileTool", () => { expect(mockTask.didToolFailInCurrentTurn).toBe(true) }) + + describe("observation registry", () => { + it("records an observation on successful read of an existing file", async () => { + const mockTask = createMockTask({ + observationRegistry: new ObservationRegistry(), + }) + const callbacks = createMockCallbacks() + + // Override the beforeEach default stat mock with proper BigIntStats. + mockedFsStat.mockResolvedValue({ + isDirectory: () => false, + dev: BigInt(1), + ino: BigInt(2), + size: BigInt(300), + mtimeNs: BigInt(4_000_000_000n), + ctimeNs: BigInt(5_000_000_000n), + // Cast: the mock only implements the members the tool and versionToken read. + } as unknown as Stats) + mockedIsBinaryFile.mockResolvedValue(false) + + // Spy on observe to capture the exact key used (Windows path.resolve may use backslashes). + const reg = mockTask.observationRegistry! + const observeSpy = vi.spyOn(reg, "observe") + + // Cast: the mock task only implements the members ReadFileTool.execute touches. + await readFileTool.execute({ path: "existing.ts" }, mockTask as unknown as Task, callbacks) + + // Verify the tool called observe exactly once with a valid token. + expect(observeSpy).toHaveBeenCalledTimes(1) + const [calledPath, calledVersion] = observeSpy.mock.calls[0] + expect(calledPath).toContain("existing.ts") + expect(calledVersion).toMatch(/^\d+:\d+:\d+:\d+:\d+$/) + + // Verify get() returns the same data using the spy-captured key. + const obs = reg.get(calledPath) + expect(obs).toBeDefined() + expect(obs!.version).toBe(calledVersion) + }) + + it("a failed read (absent path) leaves the registry size 0 and does not throw", async () => { + const mockTask = createMockTask({ + observationRegistry: new ObservationRegistry(), + }) + const callbacks = createMockCallbacks() + + mockedFsReadFile.mockRejectedValue(new Error("ENOENT")) + + // Cast: the mock task only implements the members ReadFileTool.execute touches. + await readFileTool.execute({ path: "missing.ts" }, mockTask as unknown as Task, callbacks) + + // observationRegistry is guaranteed present because we passed it in createMockTask. + const reg = mockTask.observationRegistry + expect(reg).toBeDefined() + expect(reg!.size).toBe(0) + }) + + it("records an observation for legacy-format reads of existing files", async () => { + const mockTask = createMockTask({ + observationRegistry: new ObservationRegistry(), + }) + const callbacks = createMockCallbacks() + + mockedFsStat.mockResolvedValue({ + isDirectory: () => false, + dev: BigInt(1), + ino: BigInt(2), + size: BigInt(300), + mtimeNs: BigInt(4_000_000_000n), + ctimeNs: BigInt(5_000_000_000n), + // Cast: the mock only implements the members the tool and versionToken read. + } as unknown as Stats) + mockedIsBinaryFile.mockResolvedValue(false) + + const reg = mockTask.observationRegistry! + const observeSpy = vi.spyOn(reg, "observe") + + // Typed legacy (pre-refactor) params: the multi-file format with the + // _legacyFormat discriminant (see LegacyReadFileParams). + const legacyParams: LegacyReadFileParams = { + files: [{ path: "legacy.ts" }], + _legacyFormat: true, + } + + // Cast: the mock task only implements the members ReadFileTool.execute touches. + await readFileTool.execute(legacyParams, mockTask as unknown as Task, callbacks) + + expect(observeSpy).toHaveBeenCalledTimes(1) + const [calledPath, calledVersion] = observeSpy.mock.calls[0] + expect(calledPath).toContain("legacy.ts") + expect(calledVersion).toMatch(/^\d+:\d+:\d+:\d+:\d+$/) + }) + + it("does not observe when the file mutates between the pre-read and post-read stats", async () => { + const mockTask = createMockTask({ + observationRegistry: new ObservationRegistry(), + }) + const callbacks = createMockCallbacks() + + const preStats = { + isDirectory: () => false, + dev: BigInt(1), + ino: BigInt(2), + size: BigInt(300), + mtimeNs: BigInt(4_000_000_000n), + ctimeNs: BigInt(5_000_000_000n), + } + // A mutation lands mid-read: the post-read stat differs. + const postStats = { ...preStats, size: BigInt(301) } + + // Call order: directory check, pre-read stat, post-read stat. + mockedFsStat + .mockResolvedValueOnce({ isDirectory: () => false } as unknown as Stats) + .mockResolvedValueOnce(preStats as unknown as Stats) + .mockResolvedValueOnce(postStats as unknown as Stats) + mockedIsBinaryFile.mockResolvedValue(false) + + const reg = mockTask.observationRegistry! + const observeSpy = vi.spyOn(reg, "observe") + + // Cast: the mock task only implements the members ReadFileTool.execute touches. + await readFileTool.execute({ path: "mutated.ts" }, mockTask as unknown as Task, callbacks) + + // The read itself succeeded, but the target stays unobserved: the content the + // model received is not the on-disk state, so observing it would let a later + // write match a token the model never saw. + expect(observeSpy).not.toHaveBeenCalled() + expect(reg.size).toBe(0) + expect(mockTask.didToolFailInCurrentTurn).toBe(false) + }) + + it("leaves the target unobserved without failing the read when the pre-read stat fails", async () => { + const mockTask = createMockTask({ + observationRegistry: new ObservationRegistry(), + }) + const callbacks = createMockCallbacks() + + // Directory check OK; the pre-read stat fails (caught, target unobserved). + mockedFsStat + .mockResolvedValueOnce({ isDirectory: () => false } as unknown as Stats) + .mockRejectedValueOnce(new Error("EACCES")) + mockedIsBinaryFile.mockResolvedValue(false) + + const reg = mockTask.observationRegistry! + const observeSpy = vi.spyOn(reg, "observe") + + // Cast: the mock task only implements the members ReadFileTool.execute touches. + await readFileTool.execute({ path: "stat-fail.ts" }, mockTask as unknown as Task, callbacks) + + expect(observeSpy).not.toHaveBeenCalled() + expect(reg.size).toBe(0) + // The read still succeeds — a stat failure never fails the read. + expect(mockTask.didToolFailInCurrentTurn).toBe(false) + expect(callbacks.pushToolResult).toHaveBeenCalled() + }) + + it("leaves the target unobserved without failing the read when the post-read stat fails", async () => { + const mockTask = createMockTask({ + observationRegistry: new ObservationRegistry(), + }) + const callbacks = createMockCallbacks() + + const okStats = { + isDirectory: () => false, + dev: BigInt(1), + ino: BigInt(2), + size: BigInt(300), + mtimeNs: BigInt(4_000_000_000n), + ctimeNs: BigInt(5_000_000_000n), + } + // Directory check and pre-read stat OK; the post-read stat fails. + mockedFsStat + .mockResolvedValueOnce({ isDirectory: () => false } as unknown as Stats) + .mockResolvedValueOnce(okStats as unknown as Stats) + .mockRejectedValueOnce(new Error("EACCES")) + mockedIsBinaryFile.mockResolvedValue(false) + + const reg = mockTask.observationRegistry! + const observeSpy = vi.spyOn(reg, "observe") + + // Cast: the mock task only implements the members ReadFileTool.execute touches. + await readFileTool.execute({ path: "post-stat-fail.ts" }, mockTask as unknown as Task, callbacks) + + expect(observeSpy).not.toHaveBeenCalled() + expect(reg.size).toBe(0) + expect(mockTask.didToolFailInCurrentTurn).toBe(false) + expect(callbacks.pushToolResult).toHaveBeenCalled() + }) + + it("legacy format: does not observe when the file mutates between the pre-read and post-read stats", async () => { + const mockTask = createMockTask({ + observationRegistry: new ObservationRegistry(), + }) + const callbacks = createMockCallbacks() + + const preStats = { + isDirectory: () => false, + dev: BigInt(1), + ino: BigInt(2), + size: BigInt(300), + mtimeNs: BigInt(4_000_000_000n), + ctimeNs: BigInt(5_000_000_000n), + } + // Call order: directory check, pre-read stat, post-read stat (mutated). + mockedFsStat + .mockResolvedValueOnce({ isDirectory: () => false } as unknown as Stats) + .mockResolvedValueOnce(preStats as unknown as Stats) + .mockResolvedValueOnce({ ...preStats, size: BigInt(301) } as unknown as Stats) + mockedIsBinaryFile.mockResolvedValue(false) + + const reg = mockTask.observationRegistry! + const observeSpy = vi.spyOn(reg, "observe") + + const legacyParams: LegacyReadFileParams = { + files: [{ path: "legacy-mutated.ts" }], + _legacyFormat: true, + } + + // Cast: the mock task only implements the members ReadFileTool.execute touches. + await readFileTool.execute(legacyParams, mockTask as unknown as Task, callbacks) + + expect(observeSpy).not.toHaveBeenCalled() + expect(reg.size).toBe(0) + expect(mockTask.didToolFailInCurrentTurn).toBe(false) + }) + + it("legacy format: leaves the target unobserved when a stat fails without failing the read", async () => { + const mockTask = createMockTask({ + observationRegistry: new ObservationRegistry(), + }) + const callbacks = createMockCallbacks() + + // Directory check OK; the pre-read stat fails (caught, target unobserved). + mockedFsStat + .mockResolvedValueOnce({ isDirectory: () => false } as unknown as Stats) + .mockRejectedValueOnce(new Error("EACCES")) + mockedIsBinaryFile.mockResolvedValue(false) + + const reg = mockTask.observationRegistry! + const observeSpy = vi.spyOn(reg, "observe") + + const legacyParams: LegacyReadFileParams = { + files: [{ path: "legacy-stat-fail.ts" }], + _legacyFormat: true, + } + + // Cast: the mock task only implements the members ReadFileTool.execute touches. + await readFileTool.execute(legacyParams, mockTask as unknown as Task, callbacks) + + expect(observeSpy).not.toHaveBeenCalled() + expect(reg.size).toBe(0) + expect(mockTask.didToolFailInCurrentTurn).toBe(false) + expect(callbacks.pushToolResult).toHaveBeenCalled() + }) + it("legacy format: leaves the target unobserved when the post-read stat fails", async () => { + const mockTask = createMockTask({ + observationRegistry: new ObservationRegistry(), + }) + const callbacks = createMockCallbacks() + + const okStats = { + isDirectory: () => false, + dev: BigInt(1), + ino: BigInt(2), + size: BigInt(300), + mtimeNs: BigInt(4_000_000_000n), + ctimeNs: BigInt(5_000_000_000n), + } + // Directory check and pre-read stat OK; the post-read stat fails. + mockedFsStat + .mockResolvedValueOnce({ isDirectory: () => false } as unknown as Stats) + .mockResolvedValueOnce(okStats as unknown as Stats) + .mockRejectedValueOnce(new Error("EACCES")) + mockedIsBinaryFile.mockResolvedValue(false) + + const reg = mockTask.observationRegistry! + const observeSpy = vi.spyOn(reg, "observe") + + const legacyParams: LegacyReadFileParams = { + files: [{ path: "legacy-post-stat-fail.ts" }], + _legacyFormat: true, + } + + // Cast: the mock task only implements the members ReadFileTool.execute touches. + await readFileTool.execute(legacyParams, mockTask as unknown as Task, callbacks) + + expect(observeSpy).not.toHaveBeenCalled() + expect(reg.size).toBe(0) + expect(mockTask.didToolFailInCurrentTurn).toBe(false) + expect(callbacks.pushToolResult).toHaveBeenCalled() + }) + it("two separate Task-owned registries are independent", async () => { + const regA = new ObservationRegistry() + const regB = new ObservationRegistry() + regA.observe("/shared.ts", "v1") + expect(regA.get("/shared.ts")!.version).toBe("v1") + expect(regB.get("/shared.ts")).toBeUndefined() + regB.observe("/shared.ts", "v2") + expect(regA.get("/shared.ts")!.version).toBe("v1") + expect(regB.get("/shared.ts")!.version).toBe("v2") + }) + }) }) }) diff --git a/src/core/tools/__tests__/searchReplaceTool.spec.ts b/src/core/tools/__tests__/searchReplaceTool.spec.ts index 5cf10790d4..5b67d8ed4e 100644 --- a/src/core/tools/__tests__/searchReplaceTool.spec.ts +++ b/src/core/tools/__tests__/searchReplaceTool.spec.ts @@ -59,6 +59,14 @@ vi.mock("../../diff/stats", () => ({ computeDiffStats: vi.fn(() => ({ additions: 1, deletions: 1 })), })) +vi.mock("../../checkpoints", () => ({ + checkpointSave: vi.fn().mockResolvedValue(undefined), +})) + +vi.mock("../../auto-approval", () => ({ + checkAutoApproval: vi.fn().mockResolvedValue({ decision: "ask" }), +})) + vi.mock("vscode", () => ({ window: { showWarningMessage: vi.fn().mockResolvedValue(undefined), @@ -113,7 +121,9 @@ describe("searchReplaceTool", () => { getState: vi.fn().mockResolvedValue({ diagnosticsEnabled: true, writeDelayMs: 1000, - experiments: {}, + // Pin the legacy diff-editor path explicitly: the PREVENT_FOCUS_DISRUPTION default + // flipped to true (L2, plan #33), so these tests must opt out. + experiments: { preventFocusDisruption: false } as Record, }), }), } @@ -166,6 +176,10 @@ describe("searchReplaceTool", () => { fileContent?: string isPartial?: boolean accessAllowed?: boolean + experiments?: Record + // Trial addendum: extra extension-state fields merged into the pinned state + // so the helper's providerRef pin does not clobber per-test state. + state?: Record } = {}, ): Promise { const fileExists = options.fileExists ?? true @@ -176,6 +190,17 @@ describe("searchReplaceTool", () => { mockedFileExistsAtPath.mockResolvedValue(fileExists) mockedFsReadFile.mockResolvedValue(fileContent) mockCline.rooIgnoreController.validateAccess.mockReturnValue(accessAllowed) + mockCline.providerRef.deref.mockReturnValue({ + getState: vi.fn().mockResolvedValue({ + diagnosticsEnabled: true, + writeDelayMs: 1000, + // Trial addendum: pin the legacy diff-editor path for the pre-existing suites; + // the L2 default flipped preventFocusDisruption to true. Tests exercising + // the chat-diff branch opt in via the experiments option. + experiments: options.experiments ?? { preventFocusDisruption: false }, + ...options.state, + }), + }) const nativeArgs: Record = { file_path: testFilePath, @@ -330,6 +355,34 @@ describe("searchReplaceTool", () => { }) }) + describe("focus disruption default (L2: chat-diff is the default approval path)", () => { + it("saves via saveDirectly without opening the diff editor when no experiment value is stored", async () => { + mockAskApproval.mockResolvedValue(true) + // No stored experiment value: the default (flipped to true in L2) resolves + // to the chat-diff path. Trial addendum: the empty experiments object is + // threaded through the helper (its providerRef pin would otherwise clobber + // this test's stored-value scenario). + await executeSearchReplaceTool({}, { experiments: {} }) + + expect(mockCline.diffViewProvider.saveDirectly).toHaveBeenCalled() + expect(mockCline.diffViewProvider.open).not.toHaveBeenCalled() + expect(mockCline.diffViewProvider.saveChanges).not.toHaveBeenCalled() + expect(mockCline.didEditFile).toBe(true) + }) + + it("saves via saveDirectly when the user has explicitly enabled the experiment", async () => { + mockAskApproval.mockResolvedValue(true) + // Explicit stored true: same chat-diff routing as the default. + // Trial addendum: threaded through the helper's experiments option. + await executeSearchReplaceTool({}, { experiments: { preventFocusDisruption: true } }) + + expect(mockCline.diffViewProvider.saveDirectly).toHaveBeenCalled() + expect(mockCline.diffViewProvider.open).not.toHaveBeenCalled() + expect(mockCline.diffViewProvider.saveChanges).not.toHaveBeenCalled() + expect(mockCline.didEditFile).toBe(true) + }) + }) + describe("partial block handling", () => { it("handles partial block without errors after path stabilizes", async () => { // Path stabilization requires two consecutive calls with the same path @@ -439,4 +492,42 @@ describe("searchReplaceTool", () => { expect(mockAskApproval).toHaveBeenCalled() }) }) + + describe("guarded write (S4b, epic #1375)", () => { + const focusDisruption = { preventFocusDisruption: true } + + it("publishes through saveDirectly with edit kind", async () => { + const result = await executeSearchReplaceTool( + { old_string: "Line 2", new_string: "Modified Line 2" }, + { fileContent: "Line 1\nLine 2\nLine 3", experiments: focusDisruption }, + ) + + expect(mockCline.diffViewProvider.saveDirectly).toHaveBeenCalledWith( + testFilePath, + "Line 1\nModified Line 2\nLine 3", + false, + true, + 1000, + "edit", + ) + expect(mockCline.didEditFile).toBe(true) + expect(result).toBe("Tool result message") + expect(mockHandleError).not.toHaveBeenCalled() + }) + + it("surfaces the unobserved edit remediation as a tool error and publishes nothing", async () => { + const guardError = new Error("File not read yet -- read the file, then retry.") + mockCline.diffViewProvider.saveDirectly.mockRejectedValue(guardError) + + const result = await executeSearchReplaceTool( + { old_string: "Line 2", new_string: "Modified Line 2" }, + { fileContent: "Line 1\nLine 2\nLine 3", experiments: focusDisruption }, + ) + + expect(mockHandleError).toHaveBeenCalledWith("search and replace", guardError) + expect(result).toBeUndefined() + expect(mockCline.diffViewProvider.reset).toHaveBeenCalled() + expect(mockCline.didEditFile).toBe(false) + }) + }) }) diff --git a/src/core/tools/__tests__/writeToFileTool.spec.ts b/src/core/tools/__tests__/writeToFileTool.spec.ts index 52a7e3c052..01ed871ab8 100644 --- a/src/core/tools/__tests__/writeToFileTool.spec.ts +++ b/src/core/tools/__tests__/writeToFileTool.spec.ts @@ -8,7 +8,10 @@ import { getReadablePath } from "../../../utils/path" import { unescapeHtmlEntities } from "../../../utils/text-normalization" import { everyLineHasLineNumbers, stripLineNumbers } from "../../../integrations/misc/extract-text" import { ToolUse, ToolResponse, AskApproval, HandleError, PushToolResult } from "../../../shared/tools" +import { checkpointSave } from "../../checkpoints" +import { formatResponse } from "../../prompts/responses" import { writeToFileTool } from "../WriteToFileTool" +import { convertNewFileToUnifiedDiff, sanitizeUnifiedDiff } from "../../diff/stats" vi.mock("path", async () => { const originalPath = await vi.importActual("path") @@ -26,6 +29,13 @@ vi.mock("delay", () => ({ default: vi.fn(), })) +// The focus-disruption save path reads the original file content via fs.readFile. +vi.mock("fs/promises", () => ({ + default: { + readFile: vi.fn().mockResolvedValue("original content"), + }, +})) + vi.mock("../../../utils/fs", () => ({ fileExistsAtPath: vi.fn().mockResolvedValue(false), createDirectoriesForFile: vi.fn().mockResolvedValue([]), @@ -89,6 +99,10 @@ vi.mock("../../ignore/RooIgnoreController", () => ({ }, })) +vi.mock("../../checkpoints", () => ({ + checkpointSave: vi.fn().mockResolvedValue(undefined), +})) + describe("writeToFileTool", () => { // Test data const testFilePath = "test/file.txt" @@ -96,6 +110,10 @@ describe("writeToFileTool", () => { const testContent = "Line 1\nLine 2\nLine 3" const testContentWithMarkdown = "```javascript\nLine 1\nLine 2\n```" + // The exact approval diff the tool computes for a new file (B3a threads it + // into the checkpoint write for the per-step change card). + const newFileApprovalDiff = sanitizeUnifiedDiff(convertNewFileToUnifiedDiff(testContent, testFilePath)) + // Mocked functions with correct types const mockedFileExistsAtPath = fileExistsAtPath as MockedFunction const mockedCreateDirectoriesForFile = createDirectoriesForFile as MockedFunction @@ -137,6 +155,9 @@ describe("writeToFileTool", () => { getState: vi.fn().mockResolvedValue({ diagnosticsEnabled: true, writeDelayMs: 1000, + // Pin the legacy diff-editor path explicitly: the PREVENT_FOCUS_DISRUPTION default + // flipped to true (L2, plan #33), so these tests must opt out. + experiments: { preventFocusDisruption: false }, }), }), } @@ -156,6 +177,11 @@ describe("writeToFileTool", () => { userEdits: null, finalContent: "final content", }), + saveDirectly: vi.fn().mockResolvedValue({ + newProblemsMessage: "", + userEdits: undefined, + finalContent: "saved", + }), scrollToFirstDiff: vi.fn(), updateDiagnosticSettings: vi.fn(), pushToolWriteResult: vi.fn().mockImplementation(async function ( @@ -187,6 +213,7 @@ describe("writeToFileTool", () => { mockCline.say = vi.fn().mockResolvedValue(undefined) mockCline.ask = vi.fn().mockResolvedValue(undefined) mockCline.recordToolError = vi.fn() + mockCline.processQueuedMessages = vi.fn() mockCline.sayAndCreateMissingParamError = vi.fn().mockResolvedValue("Missing param error") mockAskApproval = vi.fn().mockResolvedValue(true) @@ -204,6 +231,11 @@ describe("writeToFileTool", () => { fileExists?: boolean isPartial?: boolean accessAllowed?: boolean + experiments?: Record + // Trial addendum: extra extension-state fields merged into the pinned state + // (perWriteCheckpoints, auto-approval settings, ...) so the helper's + // providerRef pin does not clobber per-test state. + state?: Record } = {}, ): Promise { // Configure mocks based on test scenario @@ -213,6 +245,18 @@ describe("writeToFileTool", () => { mockedFileExistsAtPath.mockResolvedValue(fileExists) mockCline.rooIgnoreController.validateAccess.mockReturnValue(accessAllowed) + mockCline.providerRef.deref.mockReturnValue({ + getState: vi.fn().mockResolvedValue({ + diagnosticsEnabled: true, + writeDelayMs: 1000, + // Trial addendum: pin the legacy diff-editor path for the pre-existing suites; + // the L2 default flipped preventFocusDisruption to true (chat-diff is the + // default approval path). Tests exercising the chat-diff branch opt in via + // the experiments option. + experiments: options.experiments ?? { preventFocusDisruption: false }, + ...options.state, + }), + }) // Create a tool use object const toolUse: ToolUse = { @@ -364,6 +408,34 @@ describe("writeToFileTool", () => { }) }) + describe("focus disruption default (L2: chat-diff is the default approval path)", () => { + it("saves via saveDirectly without opening the diff editor when no experiment value is stored", async () => { + mockAskApproval.mockResolvedValue(true) + // No stored experiment value: the default (flipped to true in L2) resolves + // to the chat-diff path. Trial addendum: the empty experiments object is + // threaded through the helper (its providerRef pin would otherwise clobber + // this test's stored-value scenario). + await executeWriteFileTool({}, { experiments: {} }) + + expect(mockCline.diffViewProvider.saveDirectly).toHaveBeenCalled() + expect(mockCline.diffViewProvider.open).not.toHaveBeenCalled() + expect(mockCline.diffViewProvider.saveChanges).not.toHaveBeenCalled() + expect(mockCline.didEditFile).toBe(true) + }) + + it("saves via saveDirectly when the user has explicitly enabled the experiment", async () => { + mockAskApproval.mockResolvedValue(true) + // Explicit stored true: same chat-diff routing as the default. + // Trial addendum: threaded through the helper's experiments option. + await executeWriteFileTool({}, { experiments: { preventFocusDisruption: true } }) + + expect(mockCline.diffViewProvider.saveDirectly).toHaveBeenCalled() + expect(mockCline.diffViewProvider.open).not.toHaveBeenCalled() + expect(mockCline.diffViewProvider.saveChanges).not.toHaveBeenCalled() + expect(mockCline.didEditFile).toBe(true) + }) + }) + describe("file operations", () => { it("successfully creates new files with full workflow", async () => { await executeWriteFileTool({}, { fileExists: false }) @@ -450,6 +522,58 @@ describe("writeToFileTool", () => { }) }) + describe("guarded write (S4b, epic #1375)", () => { + const focusDisruption = { preventFocusDisruption: true } + + it("publishes through saveDirectly with create kind when the write is approved", async () => { + const result = await executeWriteFileTool({}, { fileExists: true, experiments: focusDisruption }) + + expect(mockCline.diffViewProvider.saveDirectly).toHaveBeenCalledWith( + testFilePath, + testContent, + false, + true, + 1000, + "create", + ) + expect(mockCline.diffViewProvider.saveChanges).not.toHaveBeenCalled() + expect(mockCline.fileContextTracker.trackFileContext).toHaveBeenCalledWith(testFilePath, "roo_edited") + expect(mockCline.didEditFile).toBe(true) + expect(mockCline.consecutiveMistakeCount).toBe(0) + expect(result).toBe("Tool result message") + expect(mockHandleError).not.toHaveBeenCalled() + }) + + it("surfaces the unobserved-existing remediation as a tool error and publishes nothing", async () => { + const guardError = new Error( + `File already exists at ${absoluteFilePath} and was not read before this write -- read the file first, then retry.`, + ) + mockCline.diffViewProvider.saveDirectly.mockRejectedValue(guardError) + + const result = await executeWriteFileTool({}, { fileExists: true, experiments: focusDisruption }) + + expect(mockHandleError).toHaveBeenCalledWith("writing file", guardError) + expect(result).toBeUndefined() + expect(mockCline.diffViewProvider.reset).toHaveBeenCalled() + expect(mockCline.diffViewProvider.saveChanges).not.toHaveBeenCalled() + expect(mockCline.didEditFile).toBe(false) + }) + + it("surfaces the stale-version remediation as a tool error and publishes nothing", async () => { + const guardError = new Error( + "Stale version -- the file changed since you read it (expected v1, current v2); re-read the file, then retry.", + ) + mockCline.diffViewProvider.saveDirectly.mockRejectedValue(guardError) + + const result = await executeWriteFileTool({}, { fileExists: true, experiments: focusDisruption }) + + expect(mockHandleError).toHaveBeenCalledWith("writing file", guardError) + expect(result).toBeUndefined() + expect(mockCline.diffViewProvider.reset).toHaveBeenCalled() + expect(mockCline.didEditFile).toBe(false) + }) + }) + describe("error handling", () => { it("handles general file operation errors", async () => { mockCline.diffViewProvider.open.mockRejectedValue(new Error("General error")) @@ -472,4 +596,135 @@ describe("writeToFileTool", () => { expect(mockHandleError).toHaveBeenCalledWith("handling partial write_to_file", expect.any(Error)) }) }) + + describe("per-write checkpoints (B1)", () => { + const mockedCheckpointSave = checkpointSave as MockedFunction + + it("records one suppressed checkpoint after a successful write (default-on)", async () => { + await executeWriteFileTool({}) + + expect(mockedCheckpointSave).toHaveBeenCalledOnce() + // B2: the write info threads the path, operation, and the approval + // diff stats (3 added lines, 0 removed) into the checkpoint hook. + // B3a: the approval diff itself is threaded verbatim for the + // per-step change card. + expect(mockedCheckpointSave).toHaveBeenCalledWith(mockCline, false, true, { + path: testFilePath, + operation: "create", + diffStats: { additions: 3, deletions: 0 }, + diff: newFileApprovalDiff, + }) + }) + + it("does not record a checkpoint when perWriteCheckpoints is disabled", async () => { + // Trial addendum: thread the state through the helper (its providerRef pin + // would otherwise clobber this test's extension state). + await executeWriteFileTool({}, { state: { perWriteCheckpoints: false } }) + + expect(mockCline.consecutiveMistakeCount).toBe(0) + expect(mockedCheckpointSave).not.toHaveBeenCalled() + }) + + it("does not record a checkpoint when the write fails", async () => { + mockCline.diffViewProvider.open.mockRejectedValue(new Error("write failed")) + + await executeWriteFileTool({}) + + expect(mockHandleError).toHaveBeenCalledWith("writing file", expect.any(Error)) + expect(mockedCheckpointSave).not.toHaveBeenCalled() + }) + + it("waits for the per-write checkpoint before the tool completes", async () => { + let checkpointStarted = false + let releaseCheckpoint: () => void = () => {} + mockedCheckpointSave.mockImplementationOnce(() => { + checkpointStarted = true + return new Promise((resolve) => { + releaseCheckpoint = () => resolve(undefined) + }) + }) + const processQueuedSpy = vi.fn() + mockCline.processQueuedMessages = processQueuedSpy + + const toolPromise = executeWriteFileTool({}) + + // Advance microtasks until the tool reaches the checkpoint call (all + // preceding awaits are mocked resolutions, no real timers involved). + for (let i = 0; i < 50 && !checkpointStarted; i++) { + await Promise.resolve() + } + expect(checkpointStarted).toBe(true) + + let settled = false + void toolPromise.then(() => { + settled = true + }) + + // The tool must not complete while the checkpoint is still + // staging/committing: a later write started by the task loop would + // otherwise collapse into the same (or a missing) commit. + await new Promise((resolve) => setTimeout(resolve, 20)) + expect(settled).toBe(false) + expect(processQueuedSpy).not.toHaveBeenCalled() + + releaseCheckpoint() + await toolPromise + expect(settled).toBe(true) + expect(processQueuedSpy).toHaveBeenCalledOnce() + }) + + it("threads write info with approval diff stats when the prevent-focus-disruption experiment is enabled", async () => { + // Trial addendum: opt into the chat-diff branch via the helper's experiments + // option (the helper now pins the legacy path by default). + await executeWriteFileTool({}, { experiments: { preventFocusDisruption: true } }) + + // The experiment branch saves directly (no diff view) and still + // journals the write through the same single checkpoint hook, carrying + // the approval diff for the per-step change card (B3a). + expect(mockCline.diffViewProvider.saveDirectly).toHaveBeenCalledWith( + testFilePath, + testContent, + false, + true, + 1000, + // Trial addendum: S4b appends the guarded-write kind to the save call. + "create", + ) + expect(mockedCheckpointSave).toHaveBeenCalledWith(mockCline, false, true, { + path: testFilePath, + operation: "create", + diffStats: { additions: 3, deletions: 0 }, + diff: newFileApprovalDiff, + }) + }) + + it("omits diff stats from the checkpoint write when the approval diff is empty", async () => { + // Writing identical content to an existing file produces an empty + // approval diff, so the checkpoint write carries no diffStats. + vi.mocked(formatResponse.createPrettyPatch).mockReturnValueOnce("") + + await executeWriteFileTool({}, { fileExists: true }) + + expect(mockedCheckpointSave).toHaveBeenCalledOnce() + expect(mockedCheckpointSave).toHaveBeenCalledWith(mockCline, false, true, { + path: testFilePath, + operation: "update", + }) + }) + + it("threads autoApproved into the checkpoint write for auto-approved steps", async () => { + // B3a: when the step is auto-approved the checkpoint write carries + // autoApproved so checkpointSave can force the compact change card. + // Trial addendum: thread the auto-approval state through the helper. + await executeWriteFileTool({}, { state: { autoApprovalEnabled: true, alwaysAllowWrite: true } }) + + expect(mockedCheckpointSave).toHaveBeenCalledWith(mockCline, false, true, { + path: testFilePath, + operation: "create", + diffStats: { additions: 3, deletions: 0 }, + diff: newFileApprovalDiff, + autoApproved: true, + }) + }) + }) }) diff --git a/src/core/tools/apply-patch/apply.ts b/src/core/tools/apply-patch/apply.ts index 4ab377f732..64009d27bf 100644 --- a/src/core/tools/apply-patch/apply.ts +++ b/src/core/tools/apply-patch/apply.ts @@ -29,6 +29,15 @@ export interface ApplyPatchFileChange { originalContent?: string /** New content (for add/update) */ newContent?: string + /** + * B3a: the unified approval diff for this file (computed by the tool + * handler), reused by the per-step change card. + */ + diff?: string + /** B3a: { added, removed } stats of the approval diff, when computable. */ + diffStats?: { added: number; removed: number } + /** B3a: whether this file's approval was auto-approved. */ + autoApproved?: boolean } /** diff --git a/src/core/tools/guardedWrite.ts b/src/core/tools/guardedWrite.ts new file mode 100644 index 0000000000..596cc41035 --- /dev/null +++ b/src/core/tools/guardedWrite.ts @@ -0,0 +1,260 @@ +/** + * Guarded-write compare-and-swap core (upstream epic #1375, phase A4a). + * + * Wraps the S3 safeWriteText publish primitive behind version-token guards so + * that every write is deterministic: + * + * - an unobserved target may only be created when it is absent + * (createIfAbsent); + * - an observed target is published only when the on-disk version token still + * matches the token recorded at read time (replaceIfVersion); + * - an edit-style write requires a prior observation (unobservedEditGuard). + * + * A per-absolute-path FIFO chain of tail promises orders concurrent + * in-process writes to the same path: the first matching write wins, the rest + * fail stale. Observations come from the task's S2 ObservationRegistry. + */ + +import * as fs from "fs/promises" +import * as path from "path" + +import { safeWriteText } from "../../services/file-safety/safeWriteText" +import { computeVersionToken } from "../../utils/versionToken" +import type { Task } from "../task/Task" + +// -- Types ------------------------------------------------------------------ + +/** Write kind that drives guard selection. */ +export type GuardedWriteKind = "create" | "update" | "edit" + +/** Internal error thrown when a guard rejects a write. */ +class GuardRejectedError extends Error { + constructor( + message: string, + readonly path: string, + ) { + super(message) + this.name = "GuardRejectedError" + } +} + +// -- Per-path tail-promise chain -------------------------------------------- + +/** + * Per-absolute-path FIFO chain of pending guarded writes (tail promise per + * path). Every write enqueues onto the current tail for its path, so + * concurrent writes to the same path run one at a time in submission order. + * + * The chain never leaks a rejection through itself: each link settles, a + * rejected link is skipped by the next writer (a failed write must not block + * later writes to the same path), and every caller receives its own link + * promise to handle. + * + * Settled entries are evicted (below), so a long-lived extension does not + * accumulate a map entry per distinct written path. + */ +const pendingChains = new Map>() + +/** + * Enqueue a write operation on the per-path FIFO chain. + * + * Returns the promise for this link; it always settles. A prior link that + * rejected is skipped, not propagated. The map entry for this link is + * deleted once it settles — but only while it is still the current tail for + * the path, so a replacement enqueued in the meantime keeps ownership. + */ +function enqueue(pathKey: string, fn: () => Promise): Promise { + const prev = pendingChains.get(pathKey) ?? Promise.resolve() + const next = prev.then(fn, fn) + pendingChains.set(pathKey, next) + void next.then( + () => { + if (pendingChains.get(pathKey) === next) { + pendingChains.delete(pathKey) + } + }, + () => { + if (pendingChains.get(pathKey) === next) { + pendingChains.delete(pathKey) + } + }, + ) + return next +} + +// -- Guard primitives -------------------------------------------------------- + +/** + * Extract a Node errno code (e.g. "ENOENT") from a thrown value, or + * undefined when the value carries none. + */ +function errorCode(error: unknown): string | undefined { + return typeof error === "object" && error !== null && "code" in error + ? (error as { code?: string }).code + : undefined +} + +/** True when the path is absent on disk (fs.access reports ENOENT). */ +async function fileIsAbsent(absolutePath: string): Promise { + try { + await fs.access(absolutePath) + return false + } catch (error: unknown) { + return errorCode(error) === "ENOENT" + } +} + +/** + * Publish content only if the target file does not exist. + * + * Rejects with a loud remediation error when the file already exists: the + * write was issued for a file that was never read, so the caller must read + * the file first, then retry. + */ +export async function createIfAbsent(absolutePath: string, content: string): Promise { + try { + await fs.access(absolutePath) + } catch (error: unknown) { + if (errorCode(error) !== "ENOENT") { + // A real I/O failure (EACCES, EIO, ...) -- not a guard verdict. + throw error + } + await safeWriteText(absolutePath, content) + return + } + + throw new GuardRejectedError( + "File already exists at " + + absolutePath + + " and was not read before this write -- read the file first, then retry.", + absolutePath, + ) +} + +/** + * Publish content only if the current on-disk version token equals + * expectedVersion (the token observed at read time). + * + * On a match the content is published via the S3 safeWriteText primitive; on + * a mismatch the write is rejected stale with a re-read-then-retry + * remediation suffix. + */ +export async function replaceIfVersion(absolutePath: string, expectedVersion: string, content: string): Promise { + let currentVersion: string + try { + currentVersion = await computeVersionToken(absolutePath) + } catch (error: unknown) { + if (errorCode(error) === "ENOENT") { + // The observed file was deleted after the read: the version recorded + // at read time no longer exists on disk. Normalize the raw ENOENT + // into the guard's re-read-then-retry contract so the caller gets a + // remediation it can act on, not a raw errno. + throw new GuardRejectedError( + "File was deleted after it was read -- the version recorded at read time (" + + expectedVersion + + ") no longer exists; re-read the file, then retry.", + absolutePath, + ) + } + // A real I/O failure (EACCES, EIO, ...) -- not a guard verdict. + throw error + } + + if (currentVersion === expectedVersion) { + await safeWriteText(absolutePath, content) + return + } + + throw new GuardRejectedError( + "Stale version -- the file changed since you read it (expected " + + expectedVersion + + ", current " + + currentVersion + + "); re-read the file, then retry.", + absolutePath, + ) +} + +/** + * Unobserved-edit guard: an edit-style write without a prior observation is + * rejected before any I/O. The literal-match / patch logic stays with the + * tools in S4b; this guard only verifies that a read happened first. + * + * Returns Promise because the rejection is total: this function + * never resolves. + */ +export async function unobservedEditGuard(absolutePath: string): Promise { + throw new GuardRejectedError("File not read yet -- read the file, then retry.", absolutePath) +} + +// -- Public API -------------------------------------------------------------- + +/** + * Resolve a relative or absolute path against task.cwd. + * + * path.resolve also normalizes an already-absolute input (collapsing "." / ".." + * segments and trailing separators), so the key always matches the + * ObservationRegistry key recorded at read time (ReadFileTool observes under + * path.resolve(task.cwd, relPath)) and two spellings of one file share one + * FIFO chain. + */ +function resolveAbsolutePath(task: Task, relPathOrAbsolute: string): string { + return path.resolve(task.cwd, relPathOrAbsolute) +} + +/** + * Guarded write entry point. + * + * 1. Resolves the absolute path against task.cwd. + * 2. Consults the task's S2 observation registry to pick the guard: + * - unobserved + create/update: createIfAbsent (rejects if it exists); + * - observed + create on a file that vanished after the read: recreate; + * - observed otherwise: replaceIfVersion (CAS on the S1 version token); + * - unobserved + edit: unobservedEditGuard. + * 3. Runs the chosen guard on the per-path FIFO chain so concurrent writes to + * the same path are deterministically ordered. + */ +export async function guardedWrite( + task: Task, + relPathOrAbsolute: string, + content: string, + kind: GuardedWriteKind = "update", +): Promise { + const absolutePath = resolveAbsolutePath(task, relPathOrAbsolute) + + return enqueue(absolutePath, async () => { + const obs = task.observationRegistry.get(absolutePath) + + if (obs === undefined) { + // Edit-style writes require a prior read: no observation, no write. + if (kind === "edit") { + await unobservedEditGuard(absolutePath) + } + // Never read: only an absent target may be created. (The edit guard + // above rejects before reaching this line.) + await createIfAbsent(absolutePath, content) + return + } + + if (kind === "edit") { + await replaceIfVersion(absolutePath, obs.version, content) + return + } + + // kind is "create" or "update": a "create" on a file that vanished + // after the read recreates it; otherwise the version recorded at read + // time must still match the on-disk token. + if (kind === "create" && (await fileIsAbsent(absolutePath))) { + await createIfAbsent(absolutePath, content) + } else { + await replaceIfVersion(absolutePath, obs.version, content) + } + }) +} + +/** + * Reset the per-path tail-promise chains (test hook). + */ +export function resetChain(): void { + pendingChains.clear() +} diff --git a/src/core/webview/ClineProvider.ts b/src/core/webview/ClineProvider.ts index 4621cb3fc4..52bb209922 100644 --- a/src/core/webview/ClineProvider.ts +++ b/src/core/webview/ClineProvider.ts @@ -51,6 +51,8 @@ import { ORGANIZATION_ALLOW_ALL, DEFAULT_MODES, DEFAULT_CHECKPOINT_TIMEOUT_SECONDS, + DEFAULT_PER_WRITE_CHECKPOINTS, + DEFAULT_CHANGE_CARD_DETAIL, getModelId, isRetiredProvider, providerIdentifiers, @@ -2556,6 +2558,8 @@ export class ClineProvider ttsSpeed, enableCheckpoints, checkpointTimeout, + perWriteCheckpoints, + changeCardDetail, soundVolume, writeDelayMs, diffFuzzyThreshold, @@ -2715,6 +2719,8 @@ export class ClineProvider ttsSpeed: ttsSpeed ?? 1.0, enableCheckpoints: enableCheckpoints ?? true, checkpointTimeout: checkpointTimeout ?? DEFAULT_CHECKPOINT_TIMEOUT_SECONDS, + perWriteCheckpoints: perWriteCheckpoints ?? DEFAULT_PER_WRITE_CHECKPOINTS, + changeCardDetail: changeCardDetail ?? DEFAULT_CHANGE_CARD_DETAIL, shouldShowAnnouncement: telemetrySetting !== "unset" && lastShownAnnouncementId !== this.latestAnnouncementId, allowedCommands: mergedAllowedCommands, @@ -2951,6 +2957,8 @@ export class ClineProvider ttsSpeed: stateValues.ttsSpeed ?? 1.0, enableCheckpoints: stateValues.enableCheckpoints ?? true, checkpointTimeout: stateValues.checkpointTimeout ?? DEFAULT_CHECKPOINT_TIMEOUT_SECONDS, + perWriteCheckpoints: stateValues.perWriteCheckpoints ?? DEFAULT_PER_WRITE_CHECKPOINTS, + changeCardDetail: stateValues.changeCardDetail ?? DEFAULT_CHANGE_CARD_DETAIL, soundVolume: stateValues.soundVolume, writeDelayMs: stateValues.writeDelayMs ?? DEFAULT_WRITE_DELAY_MS, diffFuzzyThreshold: stateValues.diffFuzzyThreshold ?? DEFAULT_DIFF_FUZZY_THRESHOLD, diff --git a/src/core/webview/__tests__/ClineProvider.spec.ts b/src/core/webview/__tests__/ClineProvider.spec.ts index 731124cccc..6977d30605 100644 --- a/src/core/webview/__tests__/ClineProvider.spec.ts +++ b/src/core/webview/__tests__/ClineProvider.spec.ts @@ -720,6 +720,8 @@ describe("ClineProvider", () => { soundEnabled: false, ttsEnabled: false, enableCheckpoints: false, + perWriteCheckpoints: false, + changeCardDetail: "summary", writeDelayMs: 1000, mcpEnabled: true, mode: defaultModeSlug, @@ -1401,6 +1403,81 @@ describe("ClineProvider", () => { expect(state.destructiveCommandGuardEnabled).toBe(false) }) + test("getState returns the saved per-write checkpoints setting", async () => { + await provider.contextProxy.setValue("perWriteCheckpoints", false) + + const state = await provider.getState() + + expect(state.perWriteCheckpoints).toBe(false) + }) + + test("getState defaults per-write checkpoints to true when unset", async () => { + const state = await provider.getState() + + expect(state.perWriteCheckpoints).toBe(true) + }) + + test("getStateToPostToWebview returns the saved per-write checkpoints setting", async () => { + await provider.resolveWebviewView(mockWebviewView) + await provider.contextProxy.setValue("perWriteCheckpoints", true) + + const state = await provider.getStateToPostToWebview() + + expect(state.perWriteCheckpoints).toBe(true) + }) + + test("getStateToPostToWebview returns false when per-write checkpoints is saved as false", async () => { + // The default is also true, so only an explicit false proves that the + // stored value (rather than the default) reaches the webview state. + await provider.resolveWebviewView(mockWebviewView) + await provider.contextProxy.setValue("perWriteCheckpoints", false) + + const state = await provider.getStateToPostToWebview() + + expect(state.perWriteCheckpoints).toBe(false) + }) + + test("getStateToPostToWebview defaults per-write checkpoints to true when unset", async () => { + await provider.resolveWebviewView(mockWebviewView) + + const state = await provider.getStateToPostToWebview() + + expect(state.perWriteCheckpoints).toBe(true) + }) + + test("getState returns the saved changeCardDetail setting", async () => { + await provider.contextProxy.setValue("changeCardDetail", "full") + + const state = await provider.getState() + + expect(state.changeCardDetail).toBe("full") + }) + + test("getState defaults changeCardDetail to summary when unset", async () => { + const state = await provider.getState() + + expect(state.changeCardDetail).toBe("summary") + }) + + test("getStateToPostToWebview returns the saved changeCardDetail setting", async () => { + // The default is "summary", so only an explicit "full" proves that the + // stored value (rather than the default) reaches the webview state. + await provider.resolveWebviewView(mockWebviewView) + await provider.contextProxy.setValue("changeCardDetail", "full") + + const state = await provider.getStateToPostToWebview() + + expect(state.changeCardDetail).toBe("full") + }) + + test("getStateToPostToWebview defaults changeCardDetail to summary when unset", async () => { + await provider.resolveWebviewView(mockWebviewView) + + const state = await provider.getStateToPostToWebview() + + expect(state.changeCardDetail).toBe("summary") + }) + test("language is set to VSCode language", async () => { // Mock VSCode language as Spanish ;(vscode.env as any).language = "pt-BR" @@ -1409,14 +1486,36 @@ describe("ClineProvider", () => { expect(state.language).toBe("pt-BR") }) - test("writeDelayMs defaults to 1000ms", async () => { + test("writeDelayMs defaults to DEFAULT_WRITE_DELAY_MS", async () => { // Mock globalState.get to return undefined for writeDelayMs ;(mockContext.globalState.get as any).mockImplementation((key: string) => { return key === "writeDelayMs" ? undefined : null }) const state = await provider.getState() - expect(state.writeDelayMs).toBe(1000) + expect(state.writeDelayMs).toBe(DEFAULT_WRITE_DELAY_MS) + }) + + test("getStateToPostToWebview returns the persisted writeDelayMs value", async () => { + await provider.resolveWebviewView(mockWebviewView) + + // Simulate the updateSettings handler storing the value. + await provider.contextProxy.setValue("writeDelayMs", 500) + + const state = await provider.getStateToPostToWebview() + + expect(state.writeDelayMs).toBe(500) + }) + + test("getStateToPostToWebview defaults writeDelayMs to DEFAULT_WRITE_DELAY_MS when unset", async () => { + await provider.resolveWebviewView(mockWebviewView) + + // Ensure the setting is not persisted. + await provider.contextProxy.setValue("writeDelayMs", undefined) + + const state = await provider.getStateToPostToWebview() + + expect(state.writeDelayMs).toBe(DEFAULT_WRITE_DELAY_MS) }) test("getState applies fallback defaults for write, diff, and terminal settings", async () => { diff --git a/src/core/webview/__tests__/webviewMessageHandler.rollback.spec.ts b/src/core/webview/__tests__/webviewMessageHandler.rollback.spec.ts new file mode 100644 index 0000000000..238e1daad5 --- /dev/null +++ b/src/core/webview/__tests__/webviewMessageHandler.rollback.spec.ts @@ -0,0 +1,374 @@ +// npx vitest run src/core/webview/__tests__/webviewMessageHandler.rollback.spec.ts +import { describe, expect, it, vi, beforeEach } from "vitest" + +import type { ExtensionMessage, WebviewMessage } from "@roo-code/types" + +import { webviewMessageHandler } from "../webviewMessageHandler" +import { restoreLatestFile, rollbackFile, rollbackStep } from "../../checkpoints/rollback" +import type { Task } from "../../task/Task" +import type { ClineProvider } from "../ClineProvider" + +// The rollback cases only call these two provider methods, so the provider +// double below is cast once at this boundary; the spy is shared so results +// can be asserted after the handler runs. +vi.mock("../../checkpoints/rollback", () => ({ + rollbackFile: vi.fn(), + rollbackStep: vi.fn(), + restoreLatestFile: vi.fn(), +})) + +vi.mock("vscode", () => ({ + window: { + showErrorMessage: vi.fn(), + }, + workspace: { + workspaceFolders: undefined, + }, +})) + +// The no-task failure posts localized copy; the extension i18n loader only +// populates resources outside tests, so the spec pins the English values the +// handler asks for (path is relative to this file: ../../../i18n = src/i18n). +vi.mock("../../../i18n", () => ({ + changeLanguage: vi.fn(), + t: (key: string) => { + const values: Record = { + "common:errors.message.no_active_task_to_roll_back": "No active task to roll back from", + "common:errors.message.no_active_task_to_restore": "No active task to restore from", + } + return values[key] ?? key + }, +})) + +// Structural mock: the handler only needs the task identity for these cases. +const mockTask = {} as Task +const postMessageToWebview = vi.fn(async (_message: ExtensionMessage) => undefined) + +function makeProvider(task: Task | undefined): ClineProvider { + const provider = { + getCurrentTask: () => task, + postMessageToWebview, + } + // Cast at the spec boundary: the rollback cases only read getCurrentTask() + // and observe postMessageToWebview calls on the structural double. + return provider as unknown as ClineProvider +} + +const provider = makeProvider(mockTask) + +describe("webviewMessageHandler - change card rollback", () => { + beforeEach(() => { + vi.clearAllMocks() + }) + + describe("checkpointRollbackFile", () => { + it("restores the file and posts the success outcome back to the webview", async () => { + vi.mocked(rollbackFile).mockResolvedValueOnce({ filePath: "src/a.ts", success: true }) + + await webviewMessageHandler(provider, { + type: "checkpointRollbackFile", + payload: { cardTs: 1000, checkpointId: "abc123", filePath: "src/a.ts" }, + }) + + expect(rollbackFile).toHaveBeenCalledWith(mockTask, "abc123", "src/a.ts") + expect(postMessageToWebview).toHaveBeenCalledWith({ + type: "checkpointRollbackResult", + checkpointRollbackResult: { cardTs: 1000, filePath: "src/a.ts", success: true }, + }) + }) + + it("posts the error outcome when the restore fails", async () => { + vi.mocked(rollbackFile).mockResolvedValueOnce({ + filePath: "src/a.ts", + success: false, + error: "checkpoint not found", + }) + + await webviewMessageHandler(provider, { + type: "checkpointRollbackFile", + payload: { cardTs: 1000, checkpointId: "abc123", filePath: "src/a.ts" }, + }) + + expect(postMessageToWebview).toHaveBeenCalledWith({ + type: "checkpointRollbackResult", + checkpointRollbackResult: { + cardTs: 1000, + filePath: "src/a.ts", + success: false, + error: "checkpoint not found", + }, + }) + }) + + it("posts a correlated failure result when there is no current task", async () => { + const emptyProvider = makeProvider(undefined) + + await webviewMessageHandler(emptyProvider, { + type: "checkpointRollbackFile", + payload: { cardTs: 1000, checkpointId: "abc123", filePath: "src/a.ts" }, + }) + + expect(rollbackFile).not.toHaveBeenCalled() + // The requesting card must clear its pending state, so the handler + // posts a correlated failure instead of nothing. + expect(postMessageToWebview).toHaveBeenCalledWith({ + type: "checkpointRollbackResult", + checkpointRollbackResult: { + cardTs: 1000, + filePath: "src/a.ts", + success: false, + error: "No active task to roll back from", + }, + }) + }) + + it("posts a correlated failure when the rollback itself throws", async () => { + vi.mocked(rollbackFile).mockRejectedValueOnce(new Error("git restore failed")) + + await webviewMessageHandler(provider, { + type: "checkpointRollbackFile", + payload: { cardTs: 1000, checkpointId: "abc123", filePath: "src/a.ts" }, + }) + + // The card must not stay pending: the handler turns the throw into a + // correlated failure result instead of dropping the message. + expect(postMessageToWebview).toHaveBeenCalledWith({ + type: "checkpointRollbackResult", + checkpointRollbackResult: { + cardTs: 1000, + filePath: "src/a.ts", + success: false, + error: "Rollback failed: git restore failed", + }, + }) + }) + }) + + describe("checkpointRollbackStep", () => { + it("restores every step file and posts the aggregated outcome", async () => { + vi.mocked(rollbackStep).mockResolvedValueOnce({ + checkpointId: "abc123", + files: [ + { filePath: "src/a.ts", success: true }, + { filePath: "src/b.ts", success: true }, + ], + }) + + await webviewMessageHandler(provider, { + type: "checkpointRollbackStep", + payload: { cardTs: 1000, checkpointId: "abc123", filePaths: ["src/a.ts", "src/b.ts"] }, + }) + + expect(rollbackStep).toHaveBeenCalledWith(mockTask, ["src/a.ts", "src/b.ts"], "abc123") + expect(postMessageToWebview).toHaveBeenCalledWith({ + type: "checkpointRollbackResult", + checkpointRollbackResult: { + cardTs: 1000, + success: true, + files: [ + { filePath: "src/a.ts", success: true }, + { filePath: "src/b.ts", success: true }, + ], + }, + }) + }) + + it("reports success false with the first failing file's error when a step file fails", async () => { + vi.mocked(rollbackStep).mockResolvedValueOnce({ + checkpointId: "abc123", + files: [ + { filePath: "src/a.ts", success: true }, + { filePath: "src/b.ts", success: false, error: "boom" }, + ], + }) + + await webviewMessageHandler(provider, { + type: "checkpointRollbackStep", + payload: { cardTs: 1000, filePaths: ["src/a.ts", "src/b.ts"] }, + }) + + // Without an explicit step checkpoint id the journal lookup is used. + expect(rollbackStep).toHaveBeenCalledWith(mockTask, ["src/a.ts", "src/b.ts"], undefined) + expect(postMessageToWebview).toHaveBeenCalledWith({ + type: "checkpointRollbackResult", + checkpointRollbackResult: { + cardTs: 1000, + success: false, + error: "boom", + files: [ + { filePath: "src/a.ts", success: true }, + { filePath: "src/b.ts", success: false, error: "boom" }, + ], + }, + }) + }) + + it("posts a correlated failure result when there is no current task", async () => { + const emptyProvider = makeProvider(undefined) + + await webviewMessageHandler(emptyProvider, { + type: "checkpointRollbackStep", + payload: { cardTs: 1000, filePaths: ["src/a.ts"] }, + }) + + expect(rollbackStep).not.toHaveBeenCalled() + expect(postMessageToWebview).toHaveBeenCalledWith({ + type: "checkpointRollbackResult", + checkpointRollbackResult: { + cardTs: 1000, + success: false, + error: "No active task to roll back from", + }, + }) + }) + + it("posts a correlated failure when the step rollback itself throws", async () => { + vi.mocked(rollbackStep).mockRejectedValueOnce(new Error("journal unreadable")) + + await webviewMessageHandler(provider, { + type: "checkpointRollbackStep", + payload: { cardTs: 1000, filePaths: ["src/a.ts"] }, + }) + + expect(postMessageToWebview).toHaveBeenCalledWith({ + type: "checkpointRollbackResult", + checkpointRollbackResult: { + cardTs: 1000, + success: false, + error: "Rollback failed: journal unreadable", + }, + }) + }) + + it("ignores payloads that do not match the schema", async () => { + await webviewMessageHandler(provider, { + // Malformed on purpose (only the webview produces this message): the cast + // lets the spec reach the handler's safeParse rejection without `any`. + type: "checkpointRollbackFile", + payload: { cardTs: 1000 } as unknown as WebviewMessage["payload"], + }) + await webviewMessageHandler(provider, { + type: "checkpointRollbackStep", + payload: { cardTs: 1000, filePaths: [] }, + }) + await webviewMessageHandler(provider, { + type: "checkpointRestoreLatestFile", + payload: { cardTs: 1000 } as unknown as WebviewMessage["payload"], + }) + + expect(rollbackFile).not.toHaveBeenCalled() + expect(rollbackStep).not.toHaveBeenCalled() + expect(restoreLatestFile).not.toHaveBeenCalled() + expect(postMessageToWebview).not.toHaveBeenCalled() + }) + }) + + describe("checkpointRestoreLatestFile", () => { + it("restores the file to its latest recorded version and posts the success outcome", async () => { + vi.mocked(restoreLatestFile).mockResolvedValueOnce({ filePath: "src/a.ts", success: true }) + + await webviewMessageHandler(provider, { + type: "checkpointRestoreLatestFile", + payload: { cardTs: 1000, filePath: "src/a.ts" }, + }) + + expect(restoreLatestFile).toHaveBeenCalledWith(mockTask, "src/a.ts") + expect(postMessageToWebview).toHaveBeenCalledWith({ + type: "checkpointRollbackResult", + checkpointRollbackResult: { + cardTs: 1000, + kind: "restore-latest", + filePath: "src/a.ts", + success: true, + }, + }) + }) + + it("flags a no-op restore-latest so the card can report it", async () => { + vi.mocked(restoreLatestFile).mockResolvedValueOnce({ filePath: "src/a.ts", success: true, noOp: true }) + + await webviewMessageHandler(provider, { + type: "checkpointRestoreLatestFile", + payload: { cardTs: 1000, filePath: "src/a.ts" }, + }) + + expect(postMessageToWebview).toHaveBeenCalledWith({ + type: "checkpointRollbackResult", + checkpointRollbackResult: { + cardTs: 1000, + kind: "restore-latest", + filePath: "src/a.ts", + success: true, + noOp: true, + }, + }) + }) + + it("posts the error outcome when the restore fails", async () => { + vi.mocked(restoreLatestFile).mockResolvedValueOnce({ + filePath: "src/a.ts", + success: false, + error: "Checkpoints are not enabled for this task", + }) + + await webviewMessageHandler(provider, { + type: "checkpointRestoreLatestFile", + payload: { cardTs: 1000, filePath: "src/a.ts" }, + }) + + expect(postMessageToWebview).toHaveBeenCalledWith({ + type: "checkpointRollbackResult", + checkpointRollbackResult: { + cardTs: 1000, + kind: "restore-latest", + filePath: "src/a.ts", + success: false, + error: "Checkpoints are not enabled for this task", + }, + }) + }) + + it("posts a correlated failure result when there is no current task", async () => { + const emptyProvider = makeProvider(undefined) + + await webviewMessageHandler(emptyProvider, { + type: "checkpointRestoreLatestFile", + payload: { cardTs: 1000, filePath: "src/a.ts" }, + }) + + expect(restoreLatestFile).not.toHaveBeenCalled() + // The requesting card must clear its pending state, so the handler + // posts a correlated failure instead of nothing. + expect(postMessageToWebview).toHaveBeenCalledWith({ + type: "checkpointRollbackResult", + checkpointRollbackResult: { + cardTs: 1000, + kind: "restore-latest", + filePath: "src/a.ts", + success: false, + error: "No active task to restore from", + }, + }) + }) + + it("posts a correlated failure when the restore itself throws", async () => { + vi.mocked(restoreLatestFile).mockRejectedValueOnce(new Error("git checkout failed")) + + await webviewMessageHandler(provider, { + type: "checkpointRestoreLatestFile", + payload: { cardTs: 1000, filePath: "src/a.ts" }, + }) + + expect(postMessageToWebview).toHaveBeenCalledWith({ + type: "checkpointRollbackResult", + checkpointRollbackResult: { + cardTs: 1000, + kind: "restore-latest", + filePath: "src/a.ts", + success: false, + error: "Restore failed: git checkout failed", + }, + }) + }) + }) +}) diff --git a/src/core/webview/webviewMessageHandler.ts b/src/core/webview/webviewMessageHandler.ts index 0dad65a480..2ba76f70c7 100644 --- a/src/core/webview/webviewMessageHandler.ts +++ b/src/core/webview/webviewMessageHandler.ts @@ -19,6 +19,9 @@ import { TelemetryEventName, RooCodeSettings, ExperimentId, + checkpointRollbackFilePayloadSchema, + checkpointRollbackStepPayloadSchema, + checkpointRestoreLatestFilePayloadSchema, checkoutDiffPayloadSchema, checkoutRestorePayloadSchema, getCompletionCheckpoint, @@ -1598,6 +1601,169 @@ export const webviewMessageHandler = async ( break } + case "checkpointRollbackFile": { + // B3b: restore one change-card file to its step checkpoint and report + // the outcome back to the requesting card (correlated by cardTs). + const result = checkpointRollbackFilePayloadSchema.safeParse(message.payload) + + if (result.success) { + const task = provider.getCurrentTask() + + if (task) { + // Lazy import: the rollback module pulls the checkpoint service and the + // editor integrations (DiffViewProvider) into the import graph. Loading it + // only when a rollback is requested keeps specs that mock `vscode` minimally + // from executing editor module-scope code at import time. + try { + const { rollbackFile } = await import("../checkpoints/rollback") + const outcome = await rollbackFile(task, result.data.checkpointId, result.data.filePath) + await provider.postMessageToWebview({ + type: "checkpointRollbackResult", + checkpointRollbackResult: { + cardTs: result.data.cardTs, + filePath: outcome.filePath, + success: outcome.success, + ...(outcome.error ? { error: outcome.error } : {}), + }, + }) + } catch (error) { + // Correlated failure: a throw between the request and the result post + // (import, journal read, git restore) would otherwise leave the + // requesting card pending forever. + await provider.postMessageToWebview({ + type: "checkpointRollbackResult", + checkpointRollbackResult: { + cardTs: result.data.cardTs, + filePath: result.data.filePath, + success: false, + error: `Rollback failed: ${error instanceof Error ? error.message : String(error)}`, + }, + }) + } + } else { + // No active task: the rollback cannot run. Post the correlated + // failure so the requesting card can clear its pending state + // instead of waiting on a result that will never arrive. + await provider.postMessageToWebview({ + type: "checkpointRollbackResult", + checkpointRollbackResult: { + cardTs: result.data.cardTs, + filePath: result.data.filePath, + success: false, + error: t("common:errors.message.no_active_task_to_roll_back"), + }, + }) + } + } + + break + } + case "checkpointRollbackStep": { + // B3b: restore every file of a change-card step to the step checkpoint. + const result = checkpointRollbackStepPayloadSchema.safeParse(message.payload) + + if (result.success) { + const task = provider.getCurrentTask() + + if (task) { + // Lazy import (see the checkpointRollbackFile case above). + try { + const { rollbackStep } = await import("../checkpoints/rollback") + const outcome = await rollbackStep(task, result.data.filePaths, result.data.checkpointId) + const firstFailure = outcome.files.find((file) => !file.success) + await provider.postMessageToWebview({ + type: "checkpointRollbackResult", + checkpointRollbackResult: { + cardTs: result.data.cardTs, + success: outcome.files.every((file) => file.success), + ...(firstFailure ? { error: firstFailure.error } : {}), + files: outcome.files, + }, + }) + } catch (error) { + // Correlated failure (see the checkpointRollbackFile case). + await provider.postMessageToWebview({ + type: "checkpointRollbackResult", + checkpointRollbackResult: { + cardTs: result.data.cardTs, + success: false, + error: `Rollback failed: ${error instanceof Error ? error.message : String(error)}`, + }, + }) + } + } else { + // No active task: post the correlated failure so the requesting + // card can clear its pending state. + await provider.postMessageToWebview({ + type: "checkpointRollbackResult", + checkpointRollbackResult: { + cardTs: result.data.cardTs, + success: false, + error: t("common:errors.message.no_active_task_to_roll_back"), + }, + }) + } + } + + break + } + case "checkpointRestoreLatestFile": { + // B3b: restore one change-card file to the latest recorded version of + // that file (the forward direction to a rollback) and report the + // outcome back to the requesting card (correlated by cardTs, kind + // "restore-latest"). + const result = checkpointRestoreLatestFilePayloadSchema.safeParse(message.payload) + + if (result.success) { + const task = provider.getCurrentTask() + + if (task) { + // Lazy import (see the checkpointRollbackFile case above). + try { + const { restoreLatestFile } = await import("../checkpoints/rollback") + const outcome = await restoreLatestFile(task, result.data.filePath) + await provider.postMessageToWebview({ + type: "checkpointRollbackResult", + checkpointRollbackResult: { + cardTs: result.data.cardTs, + kind: "restore-latest", + filePath: outcome.filePath, + success: outcome.success, + ...(outcome.noOp ? { noOp: true } : {}), + ...(outcome.error ? { error: outcome.error } : {}), + }, + }) + } catch (error) { + // Correlated failure (see the checkpointRollbackFile case). + await provider.postMessageToWebview({ + type: "checkpointRollbackResult", + checkpointRollbackResult: { + cardTs: result.data.cardTs, + kind: "restore-latest", + filePath: result.data.filePath, + success: false, + error: `Restore failed: ${error instanceof Error ? error.message : String(error)}`, + }, + }) + } + } else { + // No active task: post the correlated failure so the requesting + // card can clear its pending state. + await provider.postMessageToWebview({ + type: "checkpointRollbackResult", + checkpointRollbackResult: { + cardTs: result.data.cardTs, + kind: "restore-latest", + filePath: result.data.filePath, + success: false, + error: t("common:errors.message.no_active_task_to_restore"), + }, + }) + } + } + + break + } case "completionCheckpointDiff": { const currentCline = provider.getCurrentTask() const checkpoint = currentCline ? resolveCompletionCheckpoint(currentCline) : undefined diff --git a/src/eslint-suppressions.json b/src/eslint-suppressions.json index 36cbfeac5b..69c02c71ed 100644 --- a/src/eslint-suppressions.json +++ b/src/eslint-suppressions.json @@ -976,7 +976,7 @@ }, "core/tools/__tests__/readFileTool.spec.ts": { "@typescript-eslint/no-explicit-any": { - "count": 98 + "count": 97 } }, "core/tools/__tests__/runSlashCommandTool.spec.ts": { @@ -1171,7 +1171,7 @@ }, "integrations/editor/__tests__/DiffViewProvider.spec.ts": { "@typescript-eslint/no-explicit-any": { - "count": 310 + "count": 306 } }, "integrations/editor/__tests__/EditorUtils.spec.ts": { @@ -1721,7 +1721,7 @@ }, "utils/safeWriteJson.ts": { "@typescript-eslint/no-explicit-any": { - "count": 4 + "count": 3 } }, "utils/tts.ts": { diff --git a/src/i18n/locales/ca/common.json b/src/i18n/locales/ca/common.json index 24ae3f310c..9500c6e35b 100644 --- a/src/i18n/locales/ca/common.json +++ b/src/i18n/locales/ca/common.json @@ -102,6 +102,8 @@ }, "message": { "no_active_task_to_delete": "No hi ha cap tasca activa de la qual eliminar missatges", + "no_active_task_to_roll_back": "No hi ha cap tasca activa per desfer", + "no_active_task_to_restore": "No hi ha cap tasca activa per restaurar", "invalid_timestamp_for_deletion": "Marca de temps del missatge no vàlida per a l'eliminació", "cannot_delete_missing_timestamp": "No es pot eliminar el missatge: falta la marca de temps", "cannot_delete_invalid_timestamp": "No es pot eliminar el missatge: marca de temps no vàlida", diff --git a/src/i18n/locales/de/common.json b/src/i18n/locales/de/common.json index 54fa0b3c22..b1affc0257 100644 --- a/src/i18n/locales/de/common.json +++ b/src/i18n/locales/de/common.json @@ -99,6 +99,8 @@ }, "message": { "no_active_task_to_delete": "Keine aktive Aufgabe, aus der Nachrichten gelöscht werden können", + "no_active_task_to_roll_back": "Keine aktive Aufgabe, von der aus zurückgerollt werden kann", + "no_active_task_to_restore": "Keine aktive Aufgabe, von der aus wiederhergestellt werden kann", "invalid_timestamp_for_deletion": "Ungültiger Nachrichten-Zeitstempel zum Löschen", "cannot_delete_missing_timestamp": "Nachricht kann nicht gelöscht werden: fehlender Zeitstempel", "cannot_delete_invalid_timestamp": "Nachricht kann nicht gelöscht werden: ungültiger Zeitstempel", diff --git a/src/i18n/locales/en/common.json b/src/i18n/locales/en/common.json index 516a3d4f88..8fd0c2049e 100644 --- a/src/i18n/locales/en/common.json +++ b/src/i18n/locales/en/common.json @@ -100,6 +100,8 @@ }, "message": { "no_active_task_to_delete": "No active task to delete messages from", + "no_active_task_to_roll_back": "No active task to roll back from", + "no_active_task_to_restore": "No active task to restore from", "invalid_timestamp_for_deletion": "Invalid message timestamp for deletion", "cannot_delete_missing_timestamp": "Cannot delete message: missing timestamp", "cannot_delete_invalid_timestamp": "Cannot delete message: invalid timestamp", diff --git a/src/i18n/locales/es/common.json b/src/i18n/locales/es/common.json index 71dc994516..9dcf4cd27b 100644 --- a/src/i18n/locales/es/common.json +++ b/src/i18n/locales/es/common.json @@ -99,6 +99,8 @@ }, "message": { "no_active_task_to_delete": "No hay tarea activa de la cual eliminar mensajes", + "no_active_task_to_roll_back": "No hay ninguna tarea activa desde la que deshacer los cambios", + "no_active_task_to_restore": "No hay ninguna tarea activa desde la que restaurar los cambios", "invalid_timestamp_for_deletion": "Marca de tiempo del mensaje no válida para eliminación", "cannot_delete_missing_timestamp": "No se puede eliminar el mensaje: falta marca de tiempo", "cannot_delete_invalid_timestamp": "No se puede eliminar el mensaje: marca de tiempo no válida", diff --git a/src/i18n/locales/fr/common.json b/src/i18n/locales/fr/common.json index 87009ee988..e6cc7e3cb7 100644 --- a/src/i18n/locales/fr/common.json +++ b/src/i18n/locales/fr/common.json @@ -99,6 +99,8 @@ }, "message": { "no_active_task_to_delete": "Aucune tâche active pour supprimer des messages", + "no_active_task_to_roll_back": "Aucune tâche active pour annuler les modifications", + "no_active_task_to_restore": "Aucune tâche active pour restaurer les modifications", "invalid_timestamp_for_deletion": "Horodatage du message invalide pour la suppression", "cannot_delete_missing_timestamp": "Impossible de supprimer le message : horodatage manquant", "cannot_delete_invalid_timestamp": "Impossible de supprimer le message : horodatage invalide", diff --git a/src/i18n/locales/hi/common.json b/src/i18n/locales/hi/common.json index f4bd1c3055..4e4933134f 100644 --- a/src/i18n/locales/hi/common.json +++ b/src/i18n/locales/hi/common.json @@ -99,6 +99,8 @@ }, "message": { "no_active_task_to_delete": "संदेशों को हटाने के लिए कोई सक्रिय कार्य नहीं", + "no_active_task_to_roll_back": "वापस करने के लिए कोई सक्रिय कार्य नहीं है", + "no_active_task_to_restore": "पुनर्स्थापित करने के लिए कोई सक्रिय कार्य नहीं है", "invalid_timestamp_for_deletion": "हटाने के लिए अमान्य संदेश टाइमस्टैम्प", "cannot_delete_missing_timestamp": "संदेश हटाया नहीं जा सकता: टाइमस्टैम्प गुम है", "cannot_delete_invalid_timestamp": "संदेश हटाया नहीं जा सकता: अमान्य टाइमस्टैम्प", diff --git a/src/i18n/locales/id/common.json b/src/i18n/locales/id/common.json index bcee321af5..47e37f4323 100644 --- a/src/i18n/locales/id/common.json +++ b/src/i18n/locales/id/common.json @@ -99,6 +99,8 @@ }, "message": { "no_active_task_to_delete": "Tidak ada tugas aktif untuk menghapus pesan", + "no_active_task_to_roll_back": "Tidak ada tugas aktif untuk di-rollback", + "no_active_task_to_restore": "Tidak ada tugas aktif untuk dipulihkan", "invalid_timestamp_for_deletion": "Timestamp pesan tidak valid untuk penghapusan", "cannot_delete_missing_timestamp": "Tidak dapat menghapus pesan: timestamp tidak ada", "cannot_delete_invalid_timestamp": "Tidak dapat menghapus pesan: timestamp tidak valid", diff --git a/src/i18n/locales/it/common.json b/src/i18n/locales/it/common.json index 395be16b84..5c3920390b 100644 --- a/src/i18n/locales/it/common.json +++ b/src/i18n/locales/it/common.json @@ -99,6 +99,8 @@ }, "message": { "no_active_task_to_delete": "Nessuna attività attiva da cui eliminare messaggi", + "no_active_task_to_roll_back": "Nessun task attivo da cui annullare le modifiche", + "no_active_task_to_restore": "Nessun task attivo da cui ripristinare le modifiche", "invalid_timestamp_for_deletion": "Timestamp del messaggio non valido per l'eliminazione", "cannot_delete_missing_timestamp": "Impossibile eliminare il messaggio: timestamp mancante", "cannot_delete_invalid_timestamp": "Impossibile eliminare il messaggio: timestamp non valido", diff --git a/src/i18n/locales/ja/common.json b/src/i18n/locales/ja/common.json index 7dccfcd837..2fb3351331 100644 --- a/src/i18n/locales/ja/common.json +++ b/src/i18n/locales/ja/common.json @@ -99,6 +99,8 @@ }, "message": { "no_active_task_to_delete": "メッセージを削除するアクティブなタスクがありません", + "no_active_task_to_roll_back": "ロールバックできるアクティブなタスクがありません", + "no_active_task_to_restore": "復元できるアクティブなタスクがありません", "invalid_timestamp_for_deletion": "削除用のメッセージタイムスタンプが無効です", "cannot_delete_missing_timestamp": "メッセージを削除できません：タイムスタンプがありません", "cannot_delete_invalid_timestamp": "メッセージを削除できません：タイムスタンプが無効です", diff --git a/src/i18n/locales/ko/common.json b/src/i18n/locales/ko/common.json index 0ca65be687..23eb829e21 100644 --- a/src/i18n/locales/ko/common.json +++ b/src/i18n/locales/ko/common.json @@ -99,6 +99,8 @@ }, "message": { "no_active_task_to_delete": "메시지를 삭제할 활성 작업이 없습니다", + "no_active_task_to_roll_back": "되돌릴 활성 작업이 없습니다", + "no_active_task_to_restore": "복원할 활성 작업이 없습니다", "invalid_timestamp_for_deletion": "삭제를 위한 메시지 타임스탬프가 유효하지 않습니다", "cannot_delete_missing_timestamp": "메시지를 삭제할 수 없습니다: 타임스탬프가 없습니다", "cannot_delete_invalid_timestamp": "메시지를 삭제할 수 없습니다: 타임스탬프가 유효하지 않습니다", diff --git a/src/i18n/locales/nl/common.json b/src/i18n/locales/nl/common.json index a38415edfd..21cafafad4 100644 --- a/src/i18n/locales/nl/common.json +++ b/src/i18n/locales/nl/common.json @@ -99,6 +99,8 @@ }, "message": { "no_active_task_to_delete": "Geen actieve taak om berichten uit te verwijderen", + "no_active_task_to_roll_back": "Geen actieve taak om terug te draaien", + "no_active_task_to_restore": "Geen actieve taak om te herstellen", "invalid_timestamp_for_deletion": "Ongeldig bericht tijdstempel voor verwijdering", "cannot_delete_missing_timestamp": "Kan bericht niet verwijderen: tijdstempel ontbreekt", "cannot_delete_invalid_timestamp": "Kan bericht niet verwijderen: ongeldig tijdstempel", diff --git a/src/i18n/locales/pl/common.json b/src/i18n/locales/pl/common.json index ff898e8987..41ef2d7a70 100644 --- a/src/i18n/locales/pl/common.json +++ b/src/i18n/locales/pl/common.json @@ -99,6 +99,8 @@ }, "message": { "no_active_task_to_delete": "Brak aktywnego zadania do usunięcia wiadomości", + "no_active_task_to_roll_back": "Brak aktywnego zadania do cofnięcia", + "no_active_task_to_restore": "Brak aktywnego zadania do przywrócenia", "invalid_timestamp_for_deletion": "Nieprawidłowy znacznik czasu wiadomości do usunięcia", "cannot_delete_missing_timestamp": "Nie można usunąć wiadomości: brak znacznika czasu", "cannot_delete_invalid_timestamp": "Nie można usunąć wiadomości: nieprawidłowy znacznik czasu", diff --git a/src/i18n/locales/pt-BR/common.json b/src/i18n/locales/pt-BR/common.json index d3c31ed2dd..d76488a7e5 100644 --- a/src/i18n/locales/pt-BR/common.json +++ b/src/i18n/locales/pt-BR/common.json @@ -103,6 +103,8 @@ }, "message": { "no_active_task_to_delete": "Nenhuma tarefa ativa para excluir mensagens", + "no_active_task_to_roll_back": "Nenhuma tarefa ativa para desfazer", + "no_active_task_to_restore": "Nenhuma tarefa ativa para restaurar", "invalid_timestamp_for_deletion": "Timestamp da mensagem inválido para exclusão", "cannot_delete_missing_timestamp": "Não é possível excluir mensagem: timestamp ausente", "cannot_delete_invalid_timestamp": "Não é possível excluir mensagem: timestamp inválido", diff --git a/src/i18n/locales/ru/common.json b/src/i18n/locales/ru/common.json index 08d2e2aa2c..57b06a5325 100644 --- a/src/i18n/locales/ru/common.json +++ b/src/i18n/locales/ru/common.json @@ -99,6 +99,8 @@ }, "message": { "no_active_task_to_delete": "Нет активной задачи для удаления сообщений", + "no_active_task_to_roll_back": "Нет активного задания для отката", + "no_active_task_to_restore": "Нет активного задания для восстановления", "invalid_timestamp_for_deletion": "Недействительная временная метка сообщения для удаления", "cannot_delete_missing_timestamp": "Невозможно удалить сообщение: отсутствует временная метка", "cannot_delete_invalid_timestamp": "Невозможно удалить сообщение: недействительная временная метка", diff --git a/src/i18n/locales/tr/common.json b/src/i18n/locales/tr/common.json index 716ccbc6de..3b49e740fd 100644 --- a/src/i18n/locales/tr/common.json +++ b/src/i18n/locales/tr/common.json @@ -99,6 +99,8 @@ }, "message": { "no_active_task_to_delete": "Mesaj silinecek aktif görev yok", + "no_active_task_to_roll_back": "Geri alınacak etkin görev yok", + "no_active_task_to_restore": "Geri yüklenecek etkin görev yok", "invalid_timestamp_for_deletion": "Silme için geçersiz mesaj zaman damgası", "cannot_delete_missing_timestamp": "Mesaj silinemiyor: zaman damgası eksik", "cannot_delete_invalid_timestamp": "Mesaj silinemiyor: geçersiz zaman damgası", diff --git a/src/i18n/locales/vi/common.json b/src/i18n/locales/vi/common.json index 69c6343c31..3f5a5abb58 100644 --- a/src/i18n/locales/vi/common.json +++ b/src/i18n/locales/vi/common.json @@ -99,6 +99,8 @@ }, "message": { "no_active_task_to_delete": "Không có nhiệm vụ hoạt động để xóa tin nhắn", + "no_active_task_to_roll_back": "Không có tác vụ hoạt động để hoàn tác", + "no_active_task_to_restore": "Không có tác vụ hoạt động để khôi phục", "invalid_timestamp_for_deletion": "Dấu thời gian tin nhắn không hợp lệ để xóa", "cannot_delete_missing_timestamp": "Không thể xóa tin nhắn: thiếu dấu thời gian", "cannot_delete_invalid_timestamp": "Không thể xóa tin nhắn: dấu thời gian không hợp lệ", diff --git a/src/i18n/locales/zh-CN/common.json b/src/i18n/locales/zh-CN/common.json index 3600f0aa7c..0cac675500 100644 --- a/src/i18n/locales/zh-CN/common.json +++ b/src/i18n/locales/zh-CN/common.json @@ -104,6 +104,8 @@ }, "message": { "no_active_task_to_delete": "没有可删除消息的活跃任务", + "no_active_task_to_roll_back": "没有可回滚的活动任务", + "no_active_task_to_restore": "没有可恢复的活动任务", "invalid_timestamp_for_deletion": "删除操作的消息时间戳无效", "cannot_delete_missing_timestamp": "无法删除消息：缺少时间戳", "cannot_delete_invalid_timestamp": "无法删除消息：时间戳无效", diff --git a/src/i18n/locales/zh-TW/common.json b/src/i18n/locales/zh-TW/common.json index c635769891..a11e2ac30f 100644 --- a/src/i18n/locales/zh-TW/common.json +++ b/src/i18n/locales/zh-TW/common.json @@ -98,6 +98,8 @@ }, "message": { "no_active_task_to_delete": "沒有可刪除訊息的活躍工作", + "no_active_task_to_roll_back": "沒有可還原的活動任務", + "no_active_task_to_restore": "沒有可復原的活動任務", "invalid_timestamp_for_deletion": "刪除操作的訊息時間戳無效", "cannot_delete_missing_timestamp": "無法刪除訊息：缺少時間戳", "cannot_delete_invalid_timestamp": "無法刪除訊息：時間戳無效", diff --git a/src/integrations/editor/DiffViewProvider.ts b/src/integrations/editor/DiffViewProvider.ts index bb3368f063..6bdb021df1 100644 --- a/src/integrations/editor/DiffViewProvider.ts +++ b/src/integrations/editor/DiffViewProvider.ts @@ -18,6 +18,7 @@ import { arePathsEqual, getReadablePath } from "../../utils/path" import { formatResponse } from "../../core/prompts/responses" import { diagnosticsToProblemsString, getNewDiagnostics } from "../diagnostics" import { Task } from "../../core/task/Task" +import { guardedWrite, type GuardedWriteKind } from "../../core/tools/guardedWrite" import { DecorationController } from "./DecorationController" @@ -1136,6 +1137,10 @@ export class DiffViewProvider { * @param relPath - Relative path to the file * @param content - Content to write to the file * @param openFile - Whether to show the file in editor (false = open in memory only for diagnostics) + * @param writeKind - Guarded-write kind that selects the S4a guard for this publish. + * Defaults to "create" because this method always publishes a complete file + * content: an unobserved target may only be created when absent, and an + * observed target must still carry the version token recorded at read time. * @returns Result of the save operation including any new problems detected */ async saveDirectly( @@ -1144,6 +1149,7 @@ export class DiffViewProvider { openFile: boolean = true, diagnosticsEnabled: boolean = true, writeDelayMs: number = DEFAULT_WRITE_DELAY_MS, + writeKind: GuardedWriteKind = "create", ): Promise<{ newProblemsMessage: string | undefined userEdits: string | undefined @@ -1151,12 +1157,25 @@ export class DiffViewProvider { }> { const absolutePath = path.resolve(this.cwd, relPath) - // Get diagnostics before editing the file - this.preDiagnostics = vscode.languages.getDiagnostics() - - // Write the content directly to the file + // Get diagnostics before editing the file. Capture the snapshot locally: + // overlapping saveDirectly calls (multi-file edits) must not let a later + // call overwrite this one's baseline before its diagnostics tail runs. + const preDiagnostics = vscode.languages.getDiagnostics() + this.preDiagnostics = preDiagnostics + + // Publish through the S4 guarded-write API (epic #1375): an unobserved + // write to an existing file and a stale observed version are rejected + // with a re-read-then-retry remediation instead of overwriting the file. + // Concurrent in-process writes to the same path are already ordered by + // the guard's per-path FIFO chain, so no additional locking is added here. + const task = this.taskRef.deref() + if (!task) { + // Fail closed: without the owning task the observation registry is + // unreachable and the write cannot be guarded. + throw new Error("Cannot guard the write: the owning task is no longer available") + } await createDirectoriesForFile(absolutePath) - await fs.writeFile(absolutePath, content, "utf-8") + await guardedWrite(task, relPath, content, writeKind) // Open the document to ensure diagnostics are loaded // When openFile is false (PREVENT_FOCUS_DISRUPTION enabled), we only open in memory @@ -1175,23 +1194,64 @@ export class DiffViewProvider { await doc.save() } - // Force a small delay to ensure diagnostics are triggered - await new Promise((resolve) => setTimeout(resolve, 100)) + // The 100 ms diagnostics-settle wait is carried by the + // emitPostSaveDiagnostics tail (inMemoryDocument) instead of here: + // blocking the save path delayed every openFile=false save even when + // diagnostics were disabled or the write delay was 0. } - let newProblemsMessage = "" - + // L1 (A2): resolve without awaiting the LSP diagnostics settle. The + // diagnostics check becomes a fire-and-forget tail that emits any new + // problems via the existing "error" ClineSay type; the returned + // newProblemsMessage is therefore always undefined. if (diagnosticsEnabled) { - // Add configurable delay to allow linters time to process - const safeDelayMs = Math.max(0, writeDelayMs) + // The method's outer try/catch guarantees it never rejects, so the + // fire-and-forget call needs no .catch wrapper. + void this.emitPostSaveDiagnostics(relPath, writeDelayMs, preDiagnostics, !openFile) + } - try { - await delay(safeDelayMs) - } catch (error) { - console.warn(`Failed to apply write delay: ${error}`) - } + // Store the results for formatFileWriteResponse + this.newProblemsMessage = undefined + this.userEdits = undefined + this.relPath = relPath + this.newContent = content - const postDiagnostics = vscode.languages.getDiagnostics() + return { + newProblemsMessage: undefined, + userEdits: undefined, + finalContent: content, + } + } + + // L1 (A2): fire-and-forget post-save diagnostics. After the write delay, + // collects new Error-severity problems and emits them via the existing + // "error" ClineSay type (only Error-severity diagnostics reach this branch; + // "error" carries no task-failure semantics in core). Abort-safe: say() + // rejects when the task is aborted, so the whole body sits inside a + // try/catch that degrades to a console.warn — the tail can never reject. + private async emitPostSaveDiagnostics( + relPath: string, + writeDelayMs: number, + preDiagnostics: [vscode.Uri, vscode.Diagnostic[]][], + inMemoryDocument = false, + ): Promise { + try { + // Add configurable delay to allow linters time to process. When the + // document was opened in memory (openFile=false), the tail also + // carries the 100 ms diagnostics-settle wait that used to block + // saveDirectly. delay() never rejects, so no catch is required here. + const safeDelayMs = Math.max(0, writeDelayMs) + (inMemoryDocument ? 100 : 0) + await delay(safeDelayMs) + + // Filter to the saved file: saveDirectly resolves before this tail + // completes, so in a multi-file write sequence (e.g. apply_patch) + // a later file's problems must not be attributed to this relPath. + const savedFilePath = path.resolve(this.cwd, relPath) + // arePathsEqual: case-insensitive on Windows, where a relPath whose + // casing differs from the diagnostic URI is still the same file. + const postDiagnostics = vscode.languages + .getDiagnostics() + .filter(([uri]) => arePathsEqual(uri.fsPath, savedFilePath)) // Get diagnostic settings from state const task = this.taskRef.deref() @@ -1200,27 +1260,20 @@ export class DiffViewProvider { const maxDiagnosticMessages = state?.maxDiagnosticMessages ?? 50 const newProblems = await diagnosticsToProblemsString( - getNewDiagnostics(this.preDiagnostics, postDiagnostics), + getNewDiagnostics(preDiagnostics, postDiagnostics), [vscode.DiagnosticSeverity.Error], this.cwd, includeDiagnosticMessages, maxDiagnosticMessages, ) - newProblemsMessage = - newProblems.length > 0 ? `\n\nNew problems detected after saving the file:\n${newProblems}` : "" - } - - // Store the results for formatFileWriteResponse - this.newProblemsMessage = newProblemsMessage - this.userEdits = undefined - this.relPath = relPath - this.newContent = content - - return { - newProblemsMessage, - userEdits: undefined, - finalContent: content, + if (newProblems.length > 0) { + await task?.say("error", `New problems detected after saving file: ${relPath}\n\n${newProblems}`) + } + } catch (error) { + // Abort-safe: never let a post-save diagnostic emit become an + // unhandled rejection (say() rejects when the task is aborted). + console.warn(`Post-save diagnostics emit failed: ${error}`) } } } diff --git a/src/integrations/editor/__tests__/DiffViewProvider.spec.ts b/src/integrations/editor/__tests__/DiffViewProvider.spec.ts index aee88f4061..58db78f533 100644 --- a/src/integrations/editor/__tests__/DiffViewProvider.spec.ts +++ b/src/integrations/editor/__tests__/DiffViewProvider.spec.ts @@ -2,9 +2,17 @@ import { DiffViewProvider, DIFF_VIEW_URI_SCHEME, DIFF_VIEW_LABEL_CHANGES } from import * as vscode from "vscode" import * as path from "path" import delay from "delay" +import { DEFAULT_WRITE_DELAY_MS } from "@roo-code/types" import { makeRange, makeTextDocument, makeTextEditor, makeUri } from "../../../test-utils/vscode" +import * as fs from "fs/promises" + +import { computeVersionToken } from "../../../utils/versionToken" +import { safeWriteText } from "../../../services/file-safety/safeWriteText" +import { ObservationRegistry } from "../../../core/task/observationRegistry" +import type { Task } from "../../../core/task/Task" + // Mock delay vi.mock("delay", () => ({ default: vi.fn().mockResolvedValue(undefined), @@ -15,6 +23,20 @@ vi.mock("fs/promises", () => ({ readFile: vi.fn().mockResolvedValue("file content"), writeFile: vi.fn().mockResolvedValue(undefined), access: vi.fn().mockResolvedValue(undefined), + mkdir: vi.fn().mockResolvedValue(undefined), + rename: vi.fn().mockResolvedValue(undefined), + unlink: vi.fn().mockResolvedValue(undefined), +})) + +// Mock safeWriteText (used by saveDirectly) +vi.mock("../../../services/file-safety/safeWriteText", () => ({ + safeWriteText: vi.fn().mockResolvedValue(undefined), +})) + +// Mock the S1 version token (used by the S4 guarded write); the real +// computeVersionToken needs fs.stat, which is not part of the fs/promises mock above. +vi.mock("../../../utils/versionToken", () => ({ + computeVersionToken: vi.fn(), })) // Mock utils @@ -25,7 +47,15 @@ vi.mock("../../../utils/fs", () => ({ // Mock path vi.mock("path", () => ({ resolve: vi.fn((cwd, relPath) => `${cwd}/${relPath}`), + normalize: vi.fn((p: string) => p), + isAbsolute: vi.fn((p: string) => p.startsWith("/")), basename: vi.fn((path) => path.split("/").pop()), + dirname: vi.fn((path) => path.split("/").slice(0, -1).join("/") || "/"), + join: (...args: string[]) => args.join("/"), + // diagnosticsToProblemsString formats its output header via + // path.relative(cwd, uri.fsPath).toPosix(); the object-with-toPosix shape + // mirrors the repo's own diagnostics.spec.ts mock. + relative: vi.fn((cwd: string, p: string) => ({ toPosix: () => p.replace(`${cwd}/`, "") })), })) // Mock vscode @@ -149,8 +179,11 @@ describe("DiffViewProvider", () => { return mockWorkspaceEdit as any }) - // Create a mock Task instance + // Create a mock Task instance. The guarded write (S4b) consults the task's + // S2 observation registry, so the mock carries a real (in-memory) instance. mockTask = { + cwd: mockCwd, + observationRegistry: new ObservationRegistry(), providerRef: { deref: vi.fn().mockReturnValue({ getState: vi.fn().mockResolvedValue({ @@ -164,6 +197,8 @@ describe("DiffViewProvider", () => { }), }), }, + // L1: saveDirectly's fire-and-forget diagnostics tail emits via say(). + say: vi.fn().mockResolvedValue(true), } diffViewProvider = new DiffViewProvider(mockCwd, mockTask) @@ -783,6 +818,13 @@ describe("DiffViewProvider", () => { // Mock vscode functions vi.mocked(vscode.window.showTextDocument).mockResolvedValue({} as any) vi.mocked(vscode.languages.getDiagnostics).mockReturnValue([]) + + // Baseline for the single-writer flow these tests encode: the file was read + // before the write, so the observation registry holds the version token the + // guarded write recomputes and compares, and the target exists on disk. + mockTask.observationRegistry.observe(`${mockCwd}/test.ts`, "v1") + vi.mocked(computeVersionToken).mockResolvedValue("v1") + vi.mocked(fs.access).mockResolvedValue(undefined) }) it("should write content directly to file without opening diff view", async () => { @@ -791,9 +833,9 @@ describe("DiffViewProvider", () => { const result = await diffViewProvider.saveDirectly("test.ts", "new content", true, true, 2000) - // Verify file was written - const fs = await import("fs/promises") - expect(fs.writeFile).toHaveBeenCalledWith(`${mockCwd}/test.ts`, "new content", "utf-8") + // Verify file was written via safeWriteText + const { safeWriteText } = await import("../../../services/file-safety/safeWriteText") + expect(safeWriteText).toHaveBeenCalledWith(`${mockCwd}/test.ts`, "new content") // Verify file was opened without focus expect(vscode.window.showTextDocument).toHaveBeenCalledWith( @@ -801,25 +843,31 @@ describe("DiffViewProvider", () => { { preview: false, preserveFocus: true }, ) - // Verify diagnostics were checked after delay + // L1: saveDirectly resolves before the fire-and-forget diagnostics + // tail runs; flush one macrotask tick so the mocked delay (and the + // post-write getDiagnostics) have been reached before asserting. + await new Promise((resolve) => setTimeout(resolve, 0)) + + // Verify the tail applied the configured write delay expect(mockDelay).toHaveBeenCalledWith(2000) expect(vscode.languages.getDiagnostics).toHaveBeenCalled() - // Verify result - expect(result.newProblemsMessage).toBe("") + // Verify result: L1 no longer returns a problems message + expect(result.newProblemsMessage).toBeUndefined() expect(result.userEdits).toBeUndefined() expect(result.finalContent).toBe("new content") }) - it("should not open file when openWithoutFocus is false", async () => { + it("should not open file when openWithoutFocus is false (focus not stolen: in-memory document only)", async () => { await diffViewProvider.saveDirectly("test.ts", "new content", false, true, 1000) - // Verify file was written - const fs = await import("fs/promises") - expect(fs.writeFile).toHaveBeenCalledWith(`${mockCwd}/test.ts`, "new content", "utf-8") + // Verify file was written via safeWriteText + const { safeWriteText } = await import("../../../services/file-safety/safeWriteText") + expect(safeWriteText).toHaveBeenCalledWith(`${mockCwd}/test.ts`, "new content") - // Verify file was NOT opened + // Verify file was NOT opened in the editor, and the document is loaded in memory only expect(vscode.window.showTextDocument).not.toHaveBeenCalled() + expect(vscode.workspace.openTextDocument).toHaveBeenCalledWith(vscode.Uri.file(`${mockCwd}/test.ts`)) }) it("should skip diagnostics when diagnosticsEnabled is false", async () => { @@ -829,14 +877,18 @@ describe("DiffViewProvider", () => { await diffViewProvider.saveDirectly("test.ts", "new content", true, false, 1000) - // Verify file was written - const fs = await import("fs/promises") - expect(fs.writeFile).toHaveBeenCalledWith(`${mockCwd}/test.ts`, "new content", "utf-8") + // Verify file was written via safeWriteText + const { safeWriteText } = await import("../../../services/file-safety/safeWriteText") + expect(safeWriteText).toHaveBeenCalledWith(`${mockCwd}/test.ts`, "new content") // Verify delay was NOT called expect(mockDelay).not.toHaveBeenCalled() // getDiagnostics is called once for pre-diagnostics, but not for post-diagnostics expect(vscode.languages.getDiagnostics).toHaveBeenCalledTimes(1) + + // L1: no diagnostics tail is launched, so nothing is ever emitted + await new Promise((resolve) => setTimeout(resolve, 0)) + expect(mockTask.say).not.toHaveBeenCalled() }) it("should handle negative delay values", async () => { @@ -845,6 +897,9 @@ describe("DiffViewProvider", () => { await diffViewProvider.saveDirectly("test.ts", "new content", true, true, -500) + // L1: the tail runs after resolve; flush one macrotask tick first. + await new Promise((resolve) => setTimeout(resolve, 0)) + // Verify delay was called with 0 (safe minimum) expect(mockDelay).toHaveBeenCalledWith(0) }) @@ -852,11 +907,234 @@ describe("DiffViewProvider", () => { it("should store results for formatFileWriteResponse", async () => { await diffViewProvider.saveDirectly("test.ts", "new content", true, true, 1000) - // Verify internal state was updated - expect((diffViewProvider as any).newProblemsMessage).toBe("") - expect((diffViewProvider as any).userEdits).toBeUndefined() - expect((diffViewProvider as any).relPath).toBe("test.ts") - expect((diffViewProvider as any).newContent).toBe("new content") + // Verify internal state was updated (L1: the problems message is no + // longer stored; it is emitted asynchronously via say("error")) + expect(diffViewProvider["newProblemsMessage"]).toBeUndefined() + expect(diffViewProvider["userEdits"]).toBeUndefined() + expect(diffViewProvider["relPath"]).toBe("test.ts") + expect(diffViewProvider["newContent"]).toBe("new content") + }) + + it("resolves immediately and emits new problems via say('error') after the write delay", async () => { + const mockDelay = vi.mocked(delay) + mockDelay.mockClear() + vi.mocked(vscode.languages.getDiagnostics).mockClear() + + // Pre-write diagnostics are empty; the post-write snapshot (read by + // the fire-and-forget tail) reports one new Error-severity problem. + // vscode.workspace.fs.stat is an unimplemented vi.fn() mock, so + // diagnosticsToProblemsString takes its "(unavailable)" fallback + // branch and still formats the line. + const newDiag: vscode.Diagnostic = { + severity: vscode.DiagnosticSeverity.Error, + range: new vscode.Range(0, 0, 0, 1), + message: "boom", + } + const postDiagnostics: [vscode.Uri, vscode.Diagnostic[]][] = [[makeUri(`${mockCwd}/test.ts`), [newDiag]]] + vi.mocked(vscode.languages.getDiagnostics).mockReturnValueOnce([]).mockReturnValue(postDiagnostics) + + const result = await diffViewProvider.saveDirectly("test.ts", "new content", true, true, 100) + + // L1: saveDirectly resolves before the tail emits anything. + expect(result.newProblemsMessage).toBeUndefined() + expect(mockTask.say).not.toHaveBeenCalled() + + // Flush the fire-and-forget tail (the mocked delay resolves immediately). + await new Promise((resolve) => setTimeout(resolve, 0)) + + expect(mockDelay).toHaveBeenCalledWith(100) + expect(mockTask.say).toHaveBeenCalledTimes(1) + // The existing "error" ClineSay type is used, with the new-problems text. + expect(mockTask.say).toHaveBeenCalledWith( + "error", + expect.stringContaining("New problems detected after saving file: test.ts"), + ) + expect(mockTask.say.mock.calls[0]?.[1]).toContain("boom") + }) + + it("attributes only the saved file's new problems to the saved file", async () => { + vi.mocked(vscode.languages.getDiagnostics).mockClear() + + // Multi-file write sequence: a later file's new error must not be + // attributed to this tail's relPath by the workspace-wide snapshot. + const ownDiag: vscode.Diagnostic = { + severity: vscode.DiagnosticSeverity.Error, + range: new vscode.Range(0, 0, 0, 1), + message: "own-problem", + } + const otherDiag: vscode.Diagnostic = { + severity: vscode.DiagnosticSeverity.Error, + range: new vscode.Range(0, 0, 0, 1), + message: "other-file-problem", + } + const postDiagnostics: [vscode.Uri, vscode.Diagnostic[]][] = [ + [makeUri(`${mockCwd}/test.ts`), [ownDiag]], + [makeUri(`${mockCwd}/other.ts`), [otherDiag]], + ] + vi.mocked(vscode.languages.getDiagnostics).mockReturnValueOnce([]).mockReturnValue(postDiagnostics) + + await diffViewProvider.saveDirectly("test.ts", "new content", true, true, 100) + + // Flush the fire-and-forget tail. + await new Promise((resolve) => setTimeout(resolve, 0)) + + expect(mockTask.say).toHaveBeenCalledTimes(1) + expect(mockTask.say.mock.calls[0]?.[1]).toContain("own-problem") + expect(mockTask.say.mock.calls[0]?.[1]).not.toContain("other-file-problem") + }) + + it("attributes diagnostics to the saved file when the URI casing differs (Windows)", async () => { + const platformSpy = vi.spyOn(process, "platform", "get").mockReturnValue("win32") + vi.mocked(vscode.languages.getDiagnostics).mockClear() + + // The diagnostic URI uses different casing than the saved relPath: + // on Windows this is still the same file (arePathsEqual). + const newDiag: vscode.Diagnostic = { + severity: vscode.DiagnosticSeverity.Error, + range: new vscode.Range(0, 0, 0, 1), + message: "case-mismatch-problem", + } + const postDiagnostics: [vscode.Uri, vscode.Diagnostic[]][] = [[makeUri(`${mockCwd}/Test.ts`), [newDiag]]] + vi.mocked(vscode.languages.getDiagnostics).mockReturnValueOnce([]).mockReturnValue(postDiagnostics) + + await diffViewProvider.saveDirectly("test.ts", "new content", true, true, 100) + + // Flush the fire-and-forget tail. + await new Promise((resolve) => setTimeout(resolve, 0)) + + expect(mockTask.say).toHaveBeenCalledTimes(1) + expect(mockTask.say.mock.calls[0]?.[1]).toContain("case-mismatch-problem") + platformSpy.mockRestore() + }) + + it("does not block the save on a diagnostics settle delay when diagnostics are disabled", async () => { + // openFile=false used to await a 100 ms settle delay even when + // diagnostics were disabled; that delay now lives in the tail, which + // does not run at all when diagnosticsEnabled is false. + vi.mocked(vscode.languages.getDiagnostics).mockClear() + vi.mocked(vscode.languages.getDiagnostics).mockReturnValue([]) + const mockDelay = vi.mocked(delay) + mockDelay.mockClear() + + const result = await diffViewProvider.saveDirectly("test.ts", "new content", false, false) + + expect(result.finalContent).toBe("new content") + expect(mockDelay).not.toHaveBeenCalled() + expect(mockTask.say).not.toHaveBeenCalled() + }) + + it("carries the in-memory settle delay in the tail for openFile=false saves", async () => { + vi.mocked(vscode.languages.getDiagnostics).mockClear() + vi.mocked(vscode.languages.getDiagnostics).mockReturnValue([]) + const mockDelay = vi.mocked(delay) + mockDelay.mockClear() + + await diffViewProvider.saveDirectly("test.ts", "new content", false, true, 100) + + // writeDelayMs (100) + the 100 ms in-memory diagnostics settle, both + // applied by the tail instead of the save path. + expect(mockDelay).toHaveBeenCalledWith(200) + }) + + it("never calls say when there are no new problems", async () => { + vi.mocked(vscode.languages.getDiagnostics).mockClear() + vi.mocked(vscode.languages.getDiagnostics).mockReturnValue([]) + + await diffViewProvider.saveDirectly("test.ts", "new content", true, true, 50) + + // Flush the fire-and-forget tail (pre/post snapshots are both empty, + // so diagnosticsToProblemsString returns "" and nothing is emitted). + await new Promise((resolve) => setTimeout(resolve, 0)) + + expect(mockTask.say).not.toHaveBeenCalled() + }) + + it("logs a warning instead of an unhandled rejection when the post-save say() is aborted", async () => { + const consoleWarnSpy = vi.spyOn(console, "warn").mockImplementation(() => {}) + vi.mocked(vscode.languages.getDiagnostics).mockClear() + + // One new Error-severity problem so the tail reaches say(). + const newDiag: vscode.Diagnostic = { + severity: vscode.DiagnosticSeverity.Error, + range: new vscode.Range(0, 0, 0, 1), + message: "boom", + } + const postDiagnostics: [vscode.Uri, vscode.Diagnostic[]][] = [[makeUri(`${mockCwd}/test.ts`), [newDiag]]] + vi.mocked(vscode.languages.getDiagnostics).mockReturnValueOnce([]).mockReturnValue(postDiagnostics) + + // The task is aborted while the diagnostics tail is emitting: say() rejects. + mockTask.say.mockRejectedValueOnce(new Error("aborted")) + + await diffViewProvider.saveDirectly("test.ts", "new content", true, true, 0) + // Flush the fire-and-forget tail. + await new Promise((resolve) => setTimeout(resolve, 0)) + + // The method's outer catch swallows the rejection with a warning. + expect(consoleWarnSpy).toHaveBeenCalledWith(expect.stringContaining("Post-save diagnostics emit failed:")) + + consoleWarnSpy.mockRestore() + }) + + describe("guarded write (S4b, epic #1375)", () => { + const enoent = () => Object.assign(new Error("ENOENT: no such file or directory"), { code: "ENOENT" }) + + it("rejects an unobserved write to an existing file with the read-first remediation", async () => { + mockTask.observationRegistry.clear() + + await expect(diffViewProvider.saveDirectly("test.ts", "new content", true, false, 0)).rejects.toThrow( + "File already exists at /mock/cwd/test.ts and was not read before this write -- read the file first, then retry.", + ) + expect(safeWriteText).not.toHaveBeenCalled() + }) + + it("creates an unobserved file when the target is absent", async () => { + mockTask.observationRegistry.clear() + vi.mocked(fs.access).mockRejectedValue(enoent()) + + await diffViewProvider.saveDirectly("test.ts", "new content", true, false, 0) + + expect(safeWriteText).toHaveBeenCalledWith(`${mockCwd}/test.ts`, "new content") + }) + + it("rejects an observed write whose version token is stale", async () => { + // The file changed on disk after the read that recorded "v1". + vi.mocked(computeVersionToken).mockResolvedValue("v2") + + const result = diffViewProvider.saveDirectly("test.ts", "new content", true, false, 0) + + await expect(result).rejects.toThrow("Stale version") + await expect(result).rejects.toThrow("re-read the file, then retry.") + expect(safeWriteText).not.toHaveBeenCalled() + }) + + it("rejects an unobserved edit-kind write before any I/O", async () => { + mockTask.observationRegistry.clear() + + await expect( + diffViewProvider.saveDirectly("test.ts", "new content", true, false, 0, "edit"), + ).rejects.toThrow("File not read yet -- read the file, then retry.") + expect(safeWriteText).not.toHaveBeenCalled() + expect(fs.access).not.toHaveBeenCalled() + }) + + it("recreates an observed file that vanished after the read", async () => { + vi.mocked(fs.access).mockRejectedValue(enoent()) + + await diffViewProvider.saveDirectly("test.ts", "new content", true, false, 0) + + expect(safeWriteText).toHaveBeenCalledWith(`${mockCwd}/test.ts`, "new content") + }) + + it("fails closed when the owning task has been collected", async () => { + // A real WeakRef cannot be forced to deref to undefined deterministically + // (GC timing), so a structural stub stands in for the collected reference. + diffViewProvider["taskRef"] = { deref: () => undefined } as unknown as WeakRef + + await expect(diffViewProvider.saveDirectly("test.ts", "new content", true, false, 0)).rejects.toThrow( + "Cannot guard the write: the owning task is no longer available", + ) + expect(safeWriteText).not.toHaveBeenCalled() + }) }) }) @@ -918,8 +1196,8 @@ describe("DiffViewProvider", () => { const result = await diffViewProvider.saveChanges() - // Verify default behavior (enabled=true, delay=2000ms) - expect(mockDelay).toHaveBeenCalledWith(1000) + // Verify default behavior (enabled=true, delay falls back to DEFAULT_WRITE_DELAY_MS) + expect(mockDelay).toHaveBeenCalledWith(DEFAULT_WRITE_DELAY_MS) expect(vscode.languages.getDiagnostics).toHaveBeenCalled() expect(result.newProblemsMessage).toBe("") }) diff --git a/src/services/checkpoints/ShadowCheckpointService.ts b/src/services/checkpoints/ShadowCheckpointService.ts index 3e3d3d0653..a4d5a56756 100644 --- a/src/services/checkpoints/ShadowCheckpointService.ts +++ b/src/services/checkpoints/ShadowCheckpointService.ts @@ -407,6 +407,106 @@ export abstract class ShadowCheckpointService extends EventEmitter { } } + /** + * Restore a single file to its state at `commitHash` without moving the + * branch or truncating the checkpoint list (unlike + * {@link restoreCheckpoint}). + * + * If the file did not exist at `commitHash`, it is removed from the + * working tree instead — rolling a file back to before it was created. + */ + public async restoreFile(commitHash: string, filePath: string): Promise { + try { + this.log(`[${this.constructor.name}#restoreFile] restoring ${filePath} from ${commitHash}`) + + if (!this.git) { + throw new Error("Shadow git repo not initialized") + } + + // Git pathspecs are always POSIX: normalize a native (Windows + // backslash) path before the git calls. Without this, `cat-file -e` on + // a backslashed path never matches, the file is treated as absent at + // the checkpoint, and the delete branch below would remove a file the + // checkpoint actually contains. The local fs.rm join keeps the native + // form, since the OS treats both separators interchangeably there. + const gitPath = filePath.toPosix() + + // Constrain the path to the workspace before either branch: `..` + // segments would otherwise normalize (path.join / git pathspec) to a + // location outside `this.workspaceDir`, and the delete branch could + // remove an unrelated file. `path.resolve` normalizes the segments; + // the trailing-separator prefix check is the containment guard. + const resolvedTarget = path.resolve(this.workspaceDir, filePath) + const workspaceRoot = this.workspaceDir.endsWith(path.sep) + ? this.workspaceDir + : this.workspaceDir + path.sep + if (resolvedTarget !== this.workspaceDir && !resolvedTarget.startsWith(workspaceRoot)) { + throw new Error(`restoreFile target is outside the workspace: ${filePath}`) + } + + // The lexical check cannot see through a symlinked ancestor: a link + // inside the workspace pointing outside it passes the prefix check + // while the real target resolves elsewhere. Re-check containment on + // the resolved (real) paths whenever the target file currently + // exists — that is the destructive case. A target that does not exist + // cannot be deleted, and the checkout branch only writes files the + // task-owned shadow repo recorded. Resolving both sides keeps a + // legitimate symlinked workspace root working. + if (await fileExistsAtPath(resolvedTarget)) { + const realWorkspaceRoot = await fs.realpath(this.workspaceDir) + const realRoot = realWorkspaceRoot.endsWith(path.sep) ? realWorkspaceRoot : realWorkspaceRoot + path.sep + const realTarget = await fs.realpath(resolvedTarget) + if (realTarget !== realWorkspaceRoot && !realTarget.startsWith(realRoot)) { + throw new Error(`restoreFile target resolves outside the workspace: ${filePath}`) + } + } + + const start = Date.now() + const existed = await this.fileExistsInCommit(commitHash, gitPath) + + if (existed) { + await this.git.checkout([commitHash, "--", gitPath]) + } else { + await fs.rm(path.join(this.workspaceDir, filePath), { force: true }) + } + + const duration = Date.now() - start + this.emit("restore", { type: "restore", commitHash, duration }) + this.log(`[${this.constructor.name}#restoreFile] restored ${filePath} in ${duration}ms`) + } catch (e) { + const error = e instanceof Error ? e : new Error(String(e)) + this.log(`[${this.constructor.name}#restoreFile] failed to restore file: ${error.message}`) + this.emit("error", { type: "error", error }) + throw error + } + } + + /** Whether `filePath` exists in the tree of `commitHash`. */ + private async fileExistsInCommit(commitHash: string, filePath: string): Promise { + // A failed lookup is not evidence the file is absent. If the commit object + // itself cannot be read (invalid hash, corrupt or missing shadow repo), + // falling through to the restoreFile delete branch would remove a live + // file. Verify the object first and fail the restore loudly instead. + // + // Verification must be evidence-based: simple-git's raw() only rejects + // when git writes a fatal to stderr, and `git cat-file -e ` + // fails *silently* (exit 1, no output) — a silent resolution would be + // read as "valid commit". `rev-parse --verify` emits the resolved id on + // success and a stderr fatal on every failure mode, so the reject is + // reliable. + try { + await this.git!.raw(["rev-parse", "--verify", `${commitHash}^{commit}`]) + } catch { + throw new Error(`Checkpoint unavailable: ${commitHash}`) + } + try { + await this.git!.raw(["cat-file", "-e", `${commitHash}:${filePath}`]) + return true + } catch { + return false + } + } + public async getDiff({ from, to }: { from?: string; to?: string }): Promise { if (!this.git) { throw new Error("Shadow git repo not initialized") diff --git a/src/services/checkpoints/__tests__/ShadowCheckpointService.spec.ts b/src/services/checkpoints/__tests__/ShadowCheckpointService.spec.ts index 1710cc97e8..32be9b67e9 100644 --- a/src/services/checkpoints/__tests__/ShadowCheckpointService.spec.ts +++ b/src/services/checkpoints/__tests__/ShadowCheckpointService.spec.ts @@ -165,6 +165,170 @@ describe.each([[RepoPerTaskCheckpointService, "RepoPerTaskCheckpointService"]])( }) }) + describe(`${klass.name}#restoreFile`, () => { + it("restores a modified file to a previous checkpoint without touching other files", async () => { + await fs.writeFile(testFile, "Ahoy, world!") + const commit1 = await service.saveCheckpoint("First checkpoint") + expect(commit1?.commit).toBeTruthy() + + const newFile = path.join(service.workspaceDir, "new.txt") + await fs.writeFile(newFile, "New file content") + const commit2 = await service.saveCheckpoint("Second checkpoint") + expect(commit2?.commit).toBeTruthy() + + // Drift both files, then roll back only test.txt to commit 1. + await fs.writeFile(testFile, "Changed after checkpoint") + await fs.writeFile(newFile, "Also changed") + + await service.restoreFile(commit1!.commit, "test.txt") + + expect(await fs.readFile(testFile, "utf-8")).toBe("Ahoy, world!") + // new.txt is not part of the restore and keeps its drifted content. + expect(await fs.readFile(newFile, "utf-8")).toBe("Also changed") + }) + + it("deletes a file that did not exist at the checkpoint", async () => { + await fs.writeFile(testFile, "Ahoy, world!") + const commit1 = await service.saveCheckpoint("First checkpoint") + expect(commit1?.commit).toBeTruthy() + + const newFile = path.join(service.workspaceDir, "new.txt") + await fs.writeFile(newFile, "Created after the checkpoint") + + await service.restoreFile(commit1!.commit, "new.txt") + + expect(await fileExistsAtPath(newFile)).toBe(false) + // The untouched file keeps its checkpoint-1 content. + expect(await fs.readFile(testFile, "utf-8")).toBe("Ahoy, world!") + }) + + it("accepts a native backslashed path on Windows (git pathspecs are POSIX)", async () => { + if (process.platform !== "win32") { + // Off-Windows the journal paths are already POSIX, so there is + // nothing to normalize; the toPosix() call is still exercised + // as a no-op by every other restoreFile test. + return + } + + // A nested file, so the relative path actually contains a + // separator that Windows writes as a backslash. + const subFile = path.join(path.dirname(testFile), "subdir", "inner.txt") + await fs.mkdir(path.dirname(subFile), { recursive: true }) + await fs.writeFile(subFile, "Ahoy, world!") + const commit1 = await service.saveCheckpoint("First checkpoint") + expect(commit1?.commit).toBeTruthy() + + await fs.writeFile(subFile, "Drifted") + + // The change journal and the webview hand backslashed paths to the + // rollback service on Windows; restoreFile must normalize them for + // the git calls. Without the normalization, `cat-file -e` would + // never match and the delete branch would remove a file the + // checkpoint actually contains. + await service.restoreFile(commit1!.commit, "subdir" + path.sep + "inner.txt") + + expect(await fs.readFile(subFile, "utf-8")).toBe("Ahoy, world!") + }) + + it("rejects a restore target that escapes the workspace", async () => { + // `..` segments must be normalized and contained, so neither the + // checkout nor the delete branch can touch a file outside the + // workspace (CWE-22). + await fs.writeFile(testFile, "Ahoy, world!") + const commit1 = await service.saveCheckpoint("First checkpoint") + expect(commit1?.commit).toBeTruthy() + + await expect(service.restoreFile(commit1!.commit, path.join("..", "escape.txt"))).rejects.toThrow( + /outside the workspace/, + ) + }) + + it("rejects an unavailable checkpoint instead of deleting the live file", async () => { + // A corrupt journal entry can hand the rollback service a checkpoint + // id that does not resolve to a shadow-repo object. The lookup + // failure must not be read as "file absent at the checkpoint" — + // that would route the restore into the delete branch and remove a + // live file. The restore must fail loudly and leave the file intact. + await fs.writeFile(testFile, "Ahoy, world!") + await service.saveCheckpoint("First checkpoint") + + await expect( + service.restoreFile("0000000000000000000000000000000000000000", "test.txt"), + ).rejects.toThrow("Checkpoint unavailable") + + expect(await fileExistsAtPath(testFile)).toBe(true) + expect(await fs.readFile(testFile, "utf-8")).toBe("Ahoy, world!") + }) + + it("rejects a target that escapes the workspace through a symlinked ancestor", async () => { + // A lexical prefix check passes for a path that goes through a + // symlink inside the workspace pointing outside it; the resolved + // (real) target must be contained as well, or the delete/checkout + // branch would mutate a file the task never owned (CWE-22). + if (process.platform === "win32") { + // Creating symlinks needs elevated privileges on Windows; the + // lexical-containment test above still covers the portable case. + return + } + await fs.writeFile(testFile, "Ahoy, world!") + const commit1 = await service.saveCheckpoint("First checkpoint") + expect(commit1?.commit).toBeTruthy() + + const outsideDir = path.join(tmpDir, `outside-${Date.now()}`) + await fs.mkdir(outsideDir, { recursive: true }) + const outsideFile = path.join(outsideDir, "sneaky.txt") + await fs.writeFile(outsideFile, "outside") + + // A directory link inside the workspace pointing at the outside dir: + // lexically `link/sneaky.txt` is inside the workspace. + await fs.symlink(outsideDir, path.join(service.workspaceDir, "link"), "dir") + + // "resolves outside the workspace" is the real-path (symlink) + // guard's message; the bare "outside the workspace" would also + // match the lexical guard's error. + await expect(service.restoreFile(commit1!.commit, path.join("link", "sneaky.txt"))).rejects.toThrow( + /resolves outside the workspace/, + ) + + // The outside file survives: the restore failed before any mutation. + expect(await fileExistsAtPath(outsideFile)).toBe(true) + expect(await fs.readFile(outsideFile, "utf-8")).toBe("outside") + }) + + it("emits a restore event when a file is restored", async () => { + const restoreListener = vi.fn() + service.on("restore", restoreListener) + + await fs.writeFile(testFile, "Ahoy, world!") + const commit1 = await service.saveCheckpoint("First checkpoint") + expect(commit1?.commit).toBeTruthy() + + await fs.writeFile(testFile, "Drifted") + await service.restoreFile(commit1!.commit, "test.txt") + + expect(restoreListener).toHaveBeenCalledWith( + expect.objectContaining({ type: "restore", commitHash: commit1!.commit }), + ) + }) + + it("throws and emits an error event when the shadow repo is not initialized", async () => { + // A service that never ran initShadowGit has no git handle; restoreFile + // must fail cleanly and surface the error event. + const raw = await klass.create({ + taskId, + shadowDir: path.join(tmpDir, `noinit-${Date.now()}`), + workspaceDir: path.join(tmpDir, `ws-noinit-${Date.now()}`), + log: () => {}, + }) + + const errorListener = vi.fn() + raw.on("error", errorListener) + + await expect(raw.restoreFile("sha-x", "test.txt")).rejects.toThrow("Shadow git repo not initialized") + expect(errorListener).toHaveBeenCalledWith(expect.objectContaining({ type: "error" })) + }) + }) + describe(`${klass.name}#saveCheckpoint`, () => { it("creates a checkpoint if there are pending changes", async () => { await fs.writeFile(testFile, "Ahoy, world!") diff --git a/src/services/file-safety/__tests__/safeWriteText.spec.ts b/src/services/file-safety/__tests__/safeWriteText.spec.ts new file mode 100644 index 0000000000..4accc2b71e --- /dev/null +++ b/src/services/file-safety/__tests__/safeWriteText.spec.ts @@ -0,0 +1,614 @@ +import * as fs from "fs/promises" +import * as fsSync from "fs" +import { execFile } from "child_process" +import type { ChildProcess } from "child_process" +import * as path from "path" + +import { safeWriteText, type SafeWriteTextOptions } from "../safeWriteText" + +// Full mock for fs/promises — all methods are vi.fn() stubs +vi.mock("fs/promises", () => ({ + mkdir: vi.fn(), + access: vi.fn(), + rename: vi.fn(), + unlink: vi.fn(), + realpath: vi.fn(), +})) + +// Full mock for fs — all sync methods are vi.fn() stubs. Stats is a bare +// class stub so tests can build minimal Stats stand-ins via its prototype. +vi.mock("fs", () => ({ + openSync: vi.fn(), + writeSync: vi.fn(), + closeSync: vi.fn(), + mkdirSync: vi.fn(), + fsyncSync: vi.fn(), + chmodSync: vi.fn(), + fchmodSync: vi.fn(), + statSync: vi.fn(), + Stats: class Stats {}, +})) + +// Mock child_process.execFile (callback-based — must invoke callback to resolve) +vi.mock("child_process", () => ({ + execFile: vi.fn((cmd, args, opts, cb) => { + if (typeof cb === "function") cb(null) + }), +})) + +// Minimal stand-in for the ChildProcess that callback-form execFile returns. +const fakeChild = { kill: () => true } as unknown as ChildProcess + +// Helper that mirrors safeWriteText's path resolution exactly +function _resolvedTarget(filePath: string): string { + return path.resolve(filePath) +} +function _dirPath(filePath: string): string { + return path.dirname(_resolvedTarget(filePath)) +} +function _stagingDir(dir: string): string { + return path.join(dir, ".file-safety-staging") +} + +// Minimal Stats stand-in: the SUT only reads `.mode` from it. +function _stats(mode: number): fsSync.Stats { + const s = Object.create(fsSync.Stats.prototype) as fsSync.Stats + Object.assign(s, { mode }) + return s +} + +// ── Test 1: staging file created then cleaned after success ──────────────── + +describe("safeWriteText", () => { + beforeEach(() => { + vi.resetAllMocks() + // After resetAllMocks, vi.fn() returns undefined — restore promise defaults. + vi.mocked(fs.mkdir).mockResolvedValue(undefined) + vi.mocked(fs.access).mockResolvedValue(undefined) + vi.mocked(fs.rename).mockResolvedValue(undefined) + vi.mocked(fs.unlink).mockResolvedValue(undefined) + // Existing-target default: a regular 0o644 file. + vi.mocked(fsSync.statSync).mockReturnValue(_stats(0o644)) + // Default sync-write behaviour: report that all requested bytes were + // written. The Buffer overload passes (fd, buffer, offset, length), + // so the fourth argument is the requested length. + vi.mocked(fsSync.writeSync).mockImplementation((...args: unknown[]) => + typeof args[3] === "number" ? args[3] : 0, + ) + }) + + describe("staging and cleanup", () => { + it("creates a temp file in the staging dir, fsyncs it, renames to target, and cleans up on success", async () => { + const targetPath = "/tmp/test-dir/target.txt" + vi.mocked(fs.realpath).mockResolvedValue(targetPath) + vi.mocked(fsSync.openSync).mockReturnValue(1) // fd=1 + vi.mocked(fsSync.closeSync).mockReturnValue(undefined) + + await safeWriteText(targetPath, "hello world", { platform: "linux" }) + + // staging dir was created with private permissions — use + // stringContaining to handle Windows path resolution + expect(fsSync.mkdirSync).toHaveBeenCalledWith(expect.stringContaining(".file-safety-staging"), { + recursive: true, + mode: 0o700, + }) + // a pre-existing staging dir is repaired to private permissions too + expect(fsSync.chmodSync).toHaveBeenCalledWith(expect.stringContaining(".file-safety-staging"), 0o700) + + // temp file was opened for writing with the existing target's mode + // (default 0o644 from the statSync default mock) + expect(fsSync.openSync).toHaveBeenCalledWith(expect.stringContaining("safeWriteText_"), "w", 0o644) + + // content was written as a buffer (partial-write loop, full write) + expect(fsSync.writeSync).toHaveBeenCalledWith(1, Buffer.from("hello world", "utf8"), 0, 11) + + // fsync (sync form) was called on the fd + expect(fsSync.fsyncSync).toHaveBeenCalledWith(1) + + // file was closed + expect(fsSync.closeSync).toHaveBeenCalledWith(1) + + // atomic rename happened — realpath mock returns targetPath, so that's the dest + expect(fs.rename).toHaveBeenCalledWith(expect.stringContaining("safeWriteText_"), targetPath) + + // no unlink of temp (it's now the committed file; DACL skipped via platform:linux) + expect(fs.unlink).not.toHaveBeenCalled() + }) + }) + + // ── Test 2: fsync ordering ─────────────────────────────────────────────── + + describe("fsync ordering", () => { + it("calls fsync on the fd before close, and rename after close", async () => { + const targetPath = "/tmp/test-dir/target.txt" + vi.mocked(fs.realpath).mockResolvedValue(targetPath) + vi.mocked(fsSync.openSync).mockReturnValue(1) + + await safeWriteText(targetPath, "data", { platform: "linux" }) + + // Verify call order: openSync(temp) → writeSync → fsyncSync(temp) + // → closeSync(temp) → rename. On POSIX the parent directory is then + // opened and fsynced after the commit rename, so openSync/fsyncSync/ + // closeSync each have a second (directory) call. + expect(vi.mocked(fsSync.openSync).mock.calls.length).toBe(2) + expect(vi.mocked(fsSync.writeSync).mock.calls.length).toBe(1) + expect(vi.mocked(fsSync.fsyncSync).mock.calls.length).toBe(2) + expect(vi.mocked(fsSync.closeSync).mock.calls.length).toBe(2) + + // the temp file was fully closed before the commit rename + expect(vi.mocked(fsSync.closeSync).mock.calls[0][0]).toBe(1) + expect(fs.rename).toHaveBeenCalled() + }) + }) + + // ── Test 3: simulated failure between write and rename leaves target intact ── + + describe("crash/torn-write safety", () => { + it("simulated failure between fsync and rename leaves the target byte-identical and no temp left behind", async () => { + const targetPath = "/tmp/test-dir/target.txt" + vi.mocked(fs.realpath).mockResolvedValue(targetPath) + vi.mocked(fsSync.openSync).mockReturnValue(1) + vi.mocked(fs.rename).mockRejectedValue(new Error("ENOSPC")) + + await expect(safeWriteText(targetPath, "new data", { platform: "linux" })).rejects.toThrow("ENOSPC") + + // rename was attempted (the failure point) + expect(fs.rename).toHaveBeenCalledWith(expect.stringContaining("safeWriteText_"), targetPath) + + // temp file was cleaned up on failure + expect(fs.unlink).toHaveBeenCalledWith(expect.stringContaining("safeWriteText_")) + + // backup was NOT created (backup:false by default), so target is untouched + // The only rename call was temp→target, not a rollback rename + expect(fs.rename).toHaveBeenCalledTimes(1) + }) + + it("a post-commit backup cleanup failure is non-fatal: the target stays committed and no temp is left behind", async () => { + const targetPath = "/tmp/test-dir/target.txt" + vi.mocked(fs.realpath).mockResolvedValue(targetPath) + vi.mocked(fsSync.openSync).mockReturnValue(1) + // The post-commit backup unlink (SUT step 6) fails — the write must + // still succeed; an orphaned backup is the documented acceptable + // outcome, so the failure is swallowed instead of rolling back. + vi.mocked(fs.unlink).mockRejectedValueOnce(new Error("EPERM")) + + await safeWriteText(targetPath, "data", { backup: true, platform: "linux" }) + + // the commit rename (temp -> target) still happened + expect(fs.rename).toHaveBeenNthCalledWith(2, expect.stringContaining("safeWriteText_"), targetPath) + + // the failing cleanup was the post-commit backup unlink + expect(fs.unlink).toHaveBeenCalledWith(expect.stringContaining("safeWriteText.bak_")) + + // no rollback rename: the committed target is not restored from the backup + expect(fs.rename).toHaveBeenCalledTimes(2) + + // the staging temp was already committed by the rename; nothing + // temp-shaped is unlinked afterwards + expect(fs.unlink).not.toHaveBeenCalledWith(expect.stringContaining("safeWriteText_")) + }) + }) + + // ── Test 4: backup:true keeps old safeWriteJson semantics incl. rollback ── + + describe("backup:true", () => { + it("renames target -> backup before commit, deletes backup on success", async () => { + const targetPath = "/tmp/test-dir/target.txt" + vi.mocked(fs.realpath).mockResolvedValue(targetPath) + vi.mocked(fsSync.openSync).mockReturnValue(1) + + await safeWriteText(targetPath, "new data", { backup: true }) + + // target was accessed (exists check) + expect(fs.access).toHaveBeenCalledWith(targetPath) + + // first rename: target -> backup + expect(fs.rename).toHaveBeenNthCalledWith(1, targetPath, expect.stringContaining("safeWriteText.bak_")) + + // second rename: temp -> target (realpath mock returns targetPath) + expect(fs.rename).toHaveBeenNthCalledWith(2, expect.stringContaining("safeWriteText_"), targetPath) + + // backup was deleted on success + expect(fs.unlink).toHaveBeenCalledWith(expect.stringContaining("safeWriteText.bak_")) + }) + + it("rollback: on failure after rename target->backup, restores backup to target", async () => { + const targetPath = "/tmp/test-dir/target.txt" + vi.mocked(fs.realpath).mockResolvedValue(targetPath) + vi.mocked(fsSync.openSync).mockReturnValue(1) + // first rename (target->backup) succeeds, second fails + let callCount = 0 + vi.mocked(fs.rename).mockImplementation(async () => { + callCount++ + if (callCount === 1) return // target -> backup + throw new Error("ENOSPC") // temp -> target fails + }) + + await expect(safeWriteText(targetPath, "new data", { backup: true })).rejects.toThrow("ENOSPC") + + // rollback rename is the 3rd call (after target->backup and temp->target failure) + expect(fs.rename).toHaveBeenNthCalledWith(3, expect.stringContaining("safeWriteText.bak_"), targetPath) + + // temp was cleaned up on failure + expect(fs.unlink).toHaveBeenCalledWith(expect.stringContaining("safeWriteText_")) + }) + + it("backup:true when target does not exist: no backup created, just commit", async () => { + const targetPath = "/tmp/test-dir/target.txt" + vi.mocked(fs.realpath).mockResolvedValue(targetPath) + vi.mocked(fsSync.openSync).mockReturnValue(1) + // fs.access resolves for dirPath check, but rejects for target check (backup path) + vi.mocked(fs.access).mockImplementation(async (p) => { + if (typeof p === "string" && p.endsWith("target.txt")) throw { code: "ENOENT" } + }) + + await safeWriteText(targetPath, "new data", { backup: true, platform: "linux" }) + + // no backup rename (target didn't exist) + expect(fs.access).toHaveBeenCalledWith(targetPath) + + // only one rename: temp -> target + expect(fs.rename).toHaveBeenCalledTimes(1) + expect(fs.rename).toHaveBeenCalledWith(expect.stringContaining("safeWriteText_"), targetPath) + + // no unlink (no backup to delete; DACL skipped via platform:linux) + expect(fs.unlink).not.toHaveBeenCalled() + }) + }) + + // ── Test 5: win32 DACL path ────────────────────────────────────────────── + + describe("win32 DACL", () => { + it.skipIf(process.platform !== "win32")( + "copies target DACL onto staging file via icacls before rename on Windows", + async () => { + const targetPath = "/tmp/test-dir/target.txt" + vi.mocked(fs.realpath).mockResolvedValue(targetPath) + await safeWriteText(targetPath, "data", { platform: "win32" }) + + // icacls dump + restore were called (execFile is callback-based mock) + expect(execFile).toHaveBeenCalledTimes(2) + }, + ) + + it("non-win32: DACL path is unreachable when platform is not win32", async () => { + const targetPath = "/tmp/test-dir/target.txt" + vi.mocked(fs.realpath).mockResolvedValue(targetPath) + vi.mocked(fsSync.openSync).mockReturnValue(1) + + await safeWriteText(targetPath, "data", { platform: "linux" }) + + // icacls was NOT called on non-win32 + expect(execFile).not.toHaveBeenCalled() + }) + + it("win32 DACL failure falls back to plain rename (never fails the write)", async () => { + const targetPath = "/tmp/test-dir/target.txt" + vi.mocked(fs.realpath).mockResolvedValue(targetPath) + vi.mocked(fsSync.openSync).mockReturnValue(1) + // icacls dump fails — the callback-based mock must invoke cb with an error. + vi.mocked(execFile).mockImplementation((_cmd, _args, _opts, cb) => { + if (typeof cb === "function") cb(new Error("icacls error"), "", "") + return fakeChild + }) + + await safeWriteText(targetPath, "data", { platform: "win32" }) + + // write succeeded despite icacls failure (fallback to plain rename) + expect(fs.rename).toHaveBeenCalled() + }) + + it("win32 DACL save args are [targetPath, /save, dumpPath, /T] before backup rename", async () => { + const targetPath = "/tmp/test-dir/target.txt" + vi.mocked(fs.realpath).mockResolvedValue(targetPath) + vi.mocked(fsSync.openSync).mockReturnValue(1) + + await safeWriteText(targetPath, "data", { backup: true, platform: "win32" }) + + // icacls was called twice (save + restore) + expect(execFile).toHaveBeenCalledTimes(2) + + // First call: save DACL from target before backup rename + const firstCall = vi.mocked(execFile).mock.calls[0] + expect(firstCall[0]).toBe("icacls") + expect(firstCall[1]).toEqual([targetPath, "/save", expect.stringContaining(".acl.tmp"), "/T"]) + + // Second call: restore DACL onto directory after commit rename + const secondCall = vi.mocked(execFile).mock.calls[1] + expect(secondCall[0]).toBe("icacls") + expect(secondCall[1]).toEqual([ + expect.stringContaining("/tmp/test-dir"), + "/restore", + expect.stringContaining(".acl.tmp"), + ]) + + // dump file was unlinked after restore + expect(fs.unlink).toHaveBeenCalledWith(expect.stringContaining(".acl.tmp")) + }) + + it("win32 DACL: dump is unlinked even when restore fails", async () => { + const targetPath = "/tmp/test-dir/target.txt" + vi.mocked(fs.realpath).mockResolvedValue(targetPath) + vi.mocked(fsSync.openSync).mockReturnValue(1) + + // icacls save succeeds, restore fails + let callCount = 0 + vi.mocked(execFile).mockImplementation((_cmd, _args, _opts, cb) => { + callCount++ + if (typeof cb === "function") { + cb(callCount === 1 ? null : new Error("icacls restore error"), "", "") + } + return fakeChild + }) + + await safeWriteText(targetPath, "data", { platform: "win32" }) + + // write succeeded despite restore failure (best-effort) + expect(fs.rename).toHaveBeenCalled() + + // dump file was still unlinked in finally + expect(fs.unlink).toHaveBeenCalledWith(expect.stringContaining(".acl.tmp")) + }) + + it("win32 DACL: when target does not exist, no save/restore/dump", async () => { + const targetPath = "/tmp/test-dir/target.txt" + vi.mocked(fs.realpath).mockResolvedValue(targetPath) + vi.mocked(fsSync.openSync).mockReturnValue(1) + + // fs.access rejects for targetPath (ENOENT), but resolves for dirPath + vi.mocked(fs.access).mockImplementation(async (p) => { + if (typeof p === "string" && p.endsWith("target.txt")) throw { code: "ENOENT" } + return undefined + }) + + await safeWriteText(targetPath, "data", { platform: "win32" }) + + // icacls was NOT called (target absent → skip DACL entirely) + expect(execFile).not.toHaveBeenCalled() + + // no dump file created or unlinked + expect(fs.unlink).not.toHaveBeenCalled() + }) + }) + + // ── Test 6: pre-written temp path (tempPath option) ────────────────────── + + describe("pre-written temp path", () => { + it("uses the provided tempPath, fsyncs it, and renames to target", async () => { + const targetPath = "/tmp/test-dir/target.txt" + vi.mocked(fs.realpath).mockResolvedValue(targetPath) + vi.mocked(fsSync.openSync).mockReturnValue(1) + + const customTempPath = "/tmp/custom-temp.tmp" + + // platform:linux skips DACL entirely so this test focuses on tempPath only + await safeWriteText(targetPath, "", { tempPath: customTempPath, platform: "linux" }) + + // openSync was called on the custom temp path (r+ mode for fsync) + expect(fsSync.openSync).toHaveBeenCalledWith(customTempPath, "r+") + + // fsync was called + expect(fsSync.fsyncSync).toHaveBeenCalledWith(1) + + // rename happened — realpath mock returns targetPath + expect(fs.rename).toHaveBeenCalledWith(customTempPath, targetPath) + + // no unlink of custom temp (caller's concern; DACL skipped via platform:linux) + expect(fs.unlink).not.toHaveBeenCalled() + + // a caller-supplied tempPath must not create the staging directory + expect(fsSync.mkdirSync).not.toHaveBeenCalled() + }) + + it("applies the existing target's mode to a caller-supplied tempPath before publishing", async () => { + const targetPath = "/tmp/test-dir/target.txt" + vi.mocked(fs.realpath).mockResolvedValue(targetPath) + vi.mocked(fsSync.statSync).mockReturnValue(_stats(0o600)) + vi.mocked(fsSync.openSync).mockReturnValue(2) + + const customTempPath = "/tmp/custom-temp.tmp" + + await safeWriteText(targetPath, "", { tempPath: customTempPath, platform: "linux" }) + + // the caller-staged temp is fchmod'd to the restrictive target mode so + // the atomic rename cannot widen a 0o600 target (CWE-732 regression) + expect(fsSync.fchmodSync).toHaveBeenCalledWith(2, 0o600) + expect(fsSync.openSync).toHaveBeenCalledWith(customTempPath, "r+") + expect(fs.rename).toHaveBeenCalledWith(customTempPath, targetPath) + }) + + it("keeps the temp's default mode when the target does not exist yet (ENOENT)", async () => { + const targetPath = "/tmp/test-dir/target.txt" + vi.mocked(fs.realpath).mockResolvedValue(targetPath) + const enoent = Object.assign(new Error("ENOENT: no such file or directory"), { code: "ENOENT" }) + vi.mocked(fsSync.statSync).mockImplementation(() => { + throw enoent + }) + vi.mocked(fsSync.openSync).mockReturnValue(2) + + const customTempPath = "/tmp/custom-temp.tmp" + + await safeWriteText(targetPath, "", { tempPath: customTempPath, platform: "linux" }) + + // no existing target, so nothing to preserve and no fchmod on the temp + expect(fsSync.fchmodSync).not.toHaveBeenCalled() + expect(fs.rename).toHaveBeenCalledWith(customTempPath, targetPath) + }) + + it("opens the temp before applying a read-only target's mode (0o444 does not block the open)", async () => { + const targetPath = "/tmp/test-dir/target.txt" + vi.mocked(fs.realpath).mockResolvedValue(targetPath) + vi.mocked(fsSync.statSync).mockReturnValue(_stats(0o444)) + vi.mocked(fsSync.openSync).mockReturnValue(3) + + const customTempPath = "/tmp/custom-temp.tmp" + + await safeWriteText(targetPath, "", { tempPath: customTempPath, platform: "linux" }) + + // a 0o444 target must not make openSync(tempPath, "r+") fail: the mode + // is applied with fchmodSync on the already-open fd, after the open + expect(fsSync.openSync).toHaveBeenCalledWith(customTempPath, "r+") + expect(fsSync.fchmodSync).toHaveBeenCalledWith(3, 0o444) + const openIdx = vi.mocked(fsSync.openSync).mock.invocationCallOrder[0] + const fchmodIdx = vi.mocked(fsSync.fchmodSync).mock.invocationCallOrder[0] + expect(openIdx).toBeLessThan(fchmodIdx) + expect(fs.rename).toHaveBeenCalledWith(customTempPath, targetPath) + }) + }) + + // ── Test 7: symlink handling (Finding 4 regression test) ───────────────── + + describe("symlink handling", () => { + it("a write through a symlink commits onto the resolved referent, never the link path", async () => { + const linkPath = "/tmp/links/link.txt" + const referentPath = "/tmp/targets/target.txt" + vi.mocked(fs.realpath).mockResolvedValue(referentPath) + vi.mocked(fsSync.openSync).mockReturnValue(1) + + await safeWriteText(linkPath, "new-content", { platform: "linux" }) + + // The commit rename must target the realpath result (the referent), never the link itself — + // that is what guarantees a write through a symlink replaces the referent's content + // and preserves the link. + expect(fs.rename).toHaveBeenCalledWith(expect.stringContaining("safeWriteText_"), referentPath) + expect(fs.rename).not.toHaveBeenCalledWith(expect.anything(), linkPath) + }) + + it("when realpath reports ENOENT (target absent), uses the given path as-is", async () => { + const targetPath = "/tmp/test-dir/target.txt" + vi.mocked(fs.realpath).mockRejectedValue(Object.assign(new Error("ENOENT"), { code: "ENOENT" })) + vi.mocked(fsSync.openSync).mockReturnValue(1) + + await safeWriteText(targetPath, "data", { platform: "linux" }) + + // rename still happened with the fallback path (path.resolve on /tmp → C:\tmp) + const resolvedFallback = _resolvedTarget(targetPath) + expect(fs.rename).toHaveBeenCalledWith(expect.stringContaining("safeWriteText_"), resolvedFallback) + }) + }) + + // ── Test 8: review fixes (permissions, partial writes, resolution, durability) ── + + describe("review fixes", () => { + it("preserves the target's restrictive mode and tolerates a failed staging-dir permission repair", async () => { + const targetPath = "/tmp/test-dir/target.txt" + vi.mocked(fs.realpath).mockResolvedValue(targetPath) + vi.mocked(fsSync.openSync).mockReturnValue(1) + vi.mocked(fsSync.statSync).mockReturnValue(_stats(0o600)) + // a pre-existing staging dir may fail its best-effort permission repair + vi.mocked(fsSync.chmodSync).mockImplementationOnce(() => { + throw new Error("EACCES") + }) + + await safeWriteText(targetPath, "secret", { platform: "linux" }) + + // the staging file inherits the target's 0o600 mode and the write commits + expect(fsSync.openSync).toHaveBeenCalledWith(expect.stringContaining("safeWriteText_"), "w", 0o600) + expect(fs.rename).toHaveBeenCalledWith(expect.stringContaining("safeWriteText_"), targetPath) + }) + + it("falls back to the 0o644 default when the target does not exist yet", async () => { + const targetPath = "/tmp/test-dir/target.txt" + vi.mocked(fs.realpath).mockResolvedValue(targetPath) + vi.mocked(fsSync.openSync).mockReturnValue(1) + vi.mocked(fsSync.statSync).mockImplementation(() => { + throw Object.assign(new Error("ENOENT"), { code: "ENOENT" }) + }) + + await safeWriteText(targetPath, "fresh", { platform: "linux" }) + + expect(fsSync.openSync).toHaveBeenCalledWith(expect.stringContaining("safeWriteText_"), "w", 0o644) + }) + + it("loops on short writes until the full content is durable before fsync", async () => { + const targetPath = "/tmp/test-dir/target.txt" + const content = "0123456789" // 10 bytes + vi.mocked(fs.realpath).mockResolvedValue(targetPath) + vi.mocked(fsSync.openSync).mockReturnValue(1) + const buffer = Buffer.from(content, "utf8") + // first write (offset 0) reports 4 bytes (short write); the loop continues + vi.mocked(fsSync.writeSync).mockImplementation((...args: unknown[]) => + args[2] === 0 ? 4 : typeof args[3] === "number" ? args[3] : 0, + ) + + await safeWriteText(targetPath, content, { platform: "linux" }) + + // [0,10) reports 4 bytes, then [4,10) writes the remaining 6 + expect(fsSync.writeSync).toHaveBeenCalledTimes(2) + expect(fsSync.writeSync).toHaveBeenNthCalledWith(1, 1, buffer, 0, 10) + expect(fsSync.writeSync).toHaveBeenNthCalledWith(2, 1, buffer, 4, 6) + expect(fsSync.fsyncSync).toHaveBeenCalledWith(1) + expect(fs.rename).toHaveBeenCalledWith(expect.stringContaining("safeWriteText_"), targetPath) + }) + + it("fsyncs the parent directory after the commit rename on POSIX", async () => { + const targetPath = "/tmp/test-dir/target.txt" + vi.mocked(fs.realpath).mockResolvedValue(targetPath) + // temp fd=1 then parent-dir fd=2 - distinct fds prove the ordering + vi.mocked(fsSync.openSync).mockReturnValueOnce(1).mockReturnValue(2) + + await safeWriteText(targetPath, "data", { platform: "linux" }) + + // the directory fsync (fd 2) happens only after the file fsync (fd 1); + // the dir path assertion is path-agnostic (stringContaining) because + // path.dirname renders the same input differently on Windows + expect(fsSync.openSync).toHaveBeenCalledWith(expect.stringContaining("test-dir"), "r") + expect(fsSync.fsyncSync).toHaveBeenNthCalledWith(1, 1) + expect(fsSync.fsyncSync).toHaveBeenNthCalledWith(2, 2) + expect(fsSync.closeSync).toHaveBeenCalledWith(2) + }) + + it("treats a failed parent-directory fsync as best-effort", async () => { + const targetPath = "/tmp/test-dir/target.txt" + vi.mocked(fs.realpath).mockResolvedValue(targetPath) + vi.mocked(fsSync.openSync) + .mockReturnValueOnce(1) + .mockImplementationOnce(() => { + throw new Error("EBADF") + }) + + // the content rename already committed; a missing directory fsync is not fatal + await safeWriteText(targetPath, "data", { platform: "linux" }) + + expect(fs.rename).toHaveBeenCalledWith(expect.stringContaining("safeWriteText_"), targetPath) + }) + + it("propagates realpath errors (EACCES and code-less) instead of the fallback path", async () => { + const targetPath = "/tmp/test-dir/target.txt" + const eacces = Object.assign(new Error("EACCES: permission denied"), { code: "EACCES" }) + vi.mocked(fs.realpath).mockRejectedValueOnce(eacces) + await expect(safeWriteText(targetPath, "data", { platform: "linux" })).rejects.toBe(eacces) + expect(fs.rename).not.toHaveBeenCalled() + + const plain = new Error("resolution failed") + vi.mocked(fs.realpath).mockRejectedValueOnce(plain) + await expect(safeWriteText(targetPath, "data", { platform: "linux" })).rejects.toBe(plain) + expect(fs.rename).not.toHaveBeenCalled() + }) + + it("backup:true propagates access errors (EACCES and code-less) instead of skipping the backup", async () => { + const targetPath = "/tmp/test-dir/target.txt" + const eacces = Object.assign(new Error("EACCES"), { code: "EACCES" }) + const plain = new Error("access failed") + vi.mocked(fs.realpath).mockResolvedValue(targetPath) + vi.mocked(fsSync.openSync).mockReturnValue(1) + // each write accesses dirPath then target; only the target access rejects + const rejectTarget = (error: Error) => async (p: unknown) => { + if (typeof p === "string" && p.endsWith("target.txt")) throw error + } + vi.mocked(fs.access) + .mockImplementationOnce(rejectTarget(eacces)) + .mockImplementationOnce(rejectTarget(eacces)) + .mockImplementationOnce(rejectTarget(plain)) + .mockImplementationOnce(rejectTarget(plain)) + + await expect(safeWriteText(targetPath, "data", { backup: true, platform: "linux" })).rejects.toEqual( + expect.objectContaining({ code: "EACCES" }), + ) + await expect(safeWriteText(targetPath, "data", { backup: true, platform: "linux" })).rejects.toThrow( + "access failed", + ) + expect(fs.rename).not.toHaveBeenCalled() + }) + }) +}) diff --git a/src/services/file-safety/safeWriteText.ts b/src/services/file-safety/safeWriteText.ts new file mode 100644 index 0000000000..71871032e5 --- /dev/null +++ b/src/services/file-safety/safeWriteText.ts @@ -0,0 +1,307 @@ +import * as fs from "fs/promises" +import * as fsSync from "fs" +import * as path from "path" +import { execFile } from "child_process" + +/** + * Options for safeWriteText atomic text publish primitive. + */ +export interface SafeWriteTextOptions { + /** + * When true, preserve the old-file semantics: rename target -> backup first, + * after commit rename delete the backup; on failure roll the backup back to + * the target path. When false (default) the atomic rename simply replaces + * the target -- crash-safe window is zero. + */ + backup?: boolean + + /** + * Platform override for testing. When omitted the real process.platform + * value is used. Set to "win32" or "linux" / "darwin" from tests so that + * both branches are reachable without needing a real Windows runner. + */ + platform?: string + + /** + * Custom execFile runner for testing (e.g. vi.fn). When omitted the real + * child_process.execFile is used. + */ + execFileRunner?: typeof execFile + + /** + * Pre-written temp path to use for the commit phase. When provided, + * safeWriteText skips creating its own staging file and uses this path + * instead (it still fsyncs before rename). Useful when a caller has + * already written data to a temp file via a custom stream. + */ + tempPath?: string +} + +// -- helpers --------------------------------------------------------------- + +/** Generate a unique temp file name in the given directory. */ +function _tempName(dir: string, prefix: string): string { + return path.join(dir, "." + prefix + "_" + Date.now() + "_" + Math.random().toString(36).substring(2) + ".tmp") +} + +/** Create a private staging sub-directory inside *dir* so that multiple + * concurrent writes never collide on their temp names. */ +function _stagingDir(dir: string): string { + const sd = path.join(dir, ".file-safety-staging") + // mode:0o700 protects a freshly created staging dir; the best-effort chmod + // repairs a pre-existing one (mkdirSync with recursive:true never chmods an + // existing directory), so staged temp files are never group/world readable. + fsSync.mkdirSync(sd, { recursive: true, mode: 0o700 }) + try { + fsSync.chmodSync(sd, 0o700) + } catch { + // best-effort: chmod denied or unavailable; a fresh dir was still + // created with the requested mode + } + return sd +} + +/** + * fsync a file descriptor so its data is durable before the atomic rename. + * Uses the sync form because this repo's @types/node does not declare + * fs.promises.fsync; the staging file is small, so the blocking window is bounded. + */ +function _fsyncFile(fd: number): void { + fsSync.fsyncSync(fd) +} + +/** Save the DACL of *srcPath* to a dump file on Windows. + * Returns true when the dump was written successfully; false otherwise. + * Never throws — callers treat failure as "skip DACL handling". */ +async function _saveDaclWindows(srcPath: string, dumpPath: string, execFileRunner?: typeof execFile): Promise { + const runner = execFileRunner ?? execFile + try { + await new Promise((resolve, reject) => { + runner("icacls", [srcPath, "/save", dumpPath, "/T"], { windowsHide: true }, (err) => + err ? reject(err) : resolve(), + ) + }) + return true + } catch { + return false + } +} + +/** Restore a DACL dump onto *dirPath* on Windows. + * Best-effort: content is already committed, so failure is non-fatal. */ +async function _restoreDaclWindows(dirPath: string, dumpPath: string, execFileRunner?: typeof execFile): Promise { + const runner = execFileRunner ?? execFile + try { + await new Promise((resolve, reject) => { + runner("icacls", [dirPath, "/restore", dumpPath], { windowsHide: true }, (err) => + err ? reject(err) : resolve(), + ) + }) + } catch { + // best-effort; content already committed + } +} + +// -- public API ------------------------------------------------------------ + +/** + * Atomic text publish primitive. + * + * 1. Write content to a temp file in a private per-write staging subdir + * (same volume -> atomic rename guaranteed). + * 2. fsync the temp file, then close it. + * 3. win32 only: if target exists save its DACL dump BEFORE backup rename. + * 4. Optionally rename target -> backup (when backup:true). + * 5. Atomic rename temp -> target. + * 6. win32 only: restore DACL onto the directory AFTER commit rename. + * 7. On success: delete backup (if any) and unlink DACL dump. + * 8. On failure: rollback backup to target path; clean up temp + dump. + */ + +/** + * Resolve the publish target: the symlink referent when the given path is an + * existing symlink, the path itself otherwise. Only ENOENT (target absent yet) + * may fall back to the given path; any other resolution error (EACCES, EIO, ...) + * propagates so a broken or unreadable symlink is never written through its + * link path. Callers that stage a temp file themselves must stage it beside + * the resolved path: the commit is a rename onto the referent, and a rename + * across filesystems fails with EXDEV. + */ +export async function resolvePublishTarget(absoluteFilePath: string): Promise { + return fs.realpath(absoluteFilePath).catch((error: unknown) => { + const code = + typeof error === "object" && error !== null && "code" in error + ? (error as { code?: string }).code + : undefined + if (code !== "ENOENT") throw error + return absoluteFilePath + }) +} + +export async function safeWriteText(filePath: string, content: string, options?: SafeWriteTextOptions): Promise { + const absoluteFilePath = path.resolve(filePath) + + // Resolve the symlink referent (see resolvePublishTarget). + const targetPath = await resolvePublishTarget(absoluteFilePath) + const dirPath = path.dirname(targetPath) + + // Ensure parent directory exists (mirrors safeWriteJson behaviour). + await fs.mkdir(dirPath, { recursive: true }) + await fs.access(dirPath) + + // Create the staging directory only when we generate the temp file there; + // callers supplying their own tempPath (e.g. safeWriteJson) must not be left + // with an empty .file-safety-staging directory behind. + const tempPath = options?.tempPath ?? _tempName(_stagingDir(dirPath), "safeWriteText") + + let backupPath: string | null = null + let releaseBackupOnSuccess = false + let daclDumpPath: string | null = null // tracked for cleanup in finally + + try { + // -- Step 1: write content to staging temp file ------------------- + if (!options?.tempPath) { + // Preserve the existing target's permissions: the staging file must + // not be published wider than the file it replaces (a 0o600 target + // must not become 0o644 through the atomic rename). + let targetMode = 0o644 // default for a fresh target + try { + targetMode = fsSync.statSync(targetPath).mode & 0o777 + } catch { + // target does not exist yet - keep the default + } + const fd = fsSync.openSync(tempPath, "w", targetMode) + try { + // Loop until every byte is written: writeSync can report a short + // (partial) write, and publishing a truncated staging file would + // commit corrupt content. + const buffer = Buffer.from(content, "utf8") + let offset = 0 + while (offset < buffer.length) { + offset += fsSync.writeSync(fd, buffer, offset, buffer.length - offset) + } + _fsyncFile(fd) + } finally { + fsSync.closeSync(fd) + } + } else { + // Preserve the existing target's mode (CWE-732): the caller-staged + // temp carries its own creation mode, and publishing it as-is would + // widen a restrictive target (e.g. 0o600 -> 0o644) through rename. + // The mode is applied with fchmodSync on the open fd (AFTER openSync): + // chmodSync on the path before the open would make a read-only target + // (0o400/0o444) fail openSync(tempPath, "r+") with EACCES. + let targetMode: number | null = null + try { + targetMode = fsSync.statSync(targetPath).mode & 0o777 + } catch { + // target does not exist yet - keep the temp's default mode + } + const fd = fsSync.openSync(tempPath, "r+") + try { + if (targetMode !== null) { + fsSync.fchmodSync(fd, targetMode) + } + _fsyncFile(fd) + } finally { + fsSync.closeSync(fd) + } + } + + // -- Step 2 (win32): save DACL BEFORE backup rename --------------- + const platform = options?.platform ?? process.platform + if (platform === "win32") { + try { + await fs.access(targetPath) // target exists? + daclDumpPath = targetPath + ".acl.tmp" + const saved = await _saveDaclWindows(targetPath, daclDumpPath, options?.execFileRunner) + if (!saved) { + daclDumpPath = null // skip DACL handling entirely + } + } catch { + // target does not exist or access failed — no DACL handling + daclDumpPath = null + } + } + + try { + // -- Step 3 (backup:true): rename target -> backup -------------- + if (options?.backup) { + try { + await fs.access(targetPath) + backupPath = _tempName(dirPath, "safeWriteText.bak") + await fs.rename(targetPath, backupPath) + releaseBackupOnSuccess = true + } catch (err: unknown) { + const code = + typeof err === "object" && err !== null && "code" in err + ? (err as { code?: string }).code + : undefined + if (code !== "ENOENT") throw err + } + } + + // -- Step 4: atomic rename temp -> target --------------------- + await fs.rename(tempPath, targetPath) + + // -- Step 4b (POSIX): fsync the parent directory so the directory entry + // changed by the commit rename is durable, not just the file content. + if (platform !== "win32") { + try { + const dirFd = fsSync.openSync(dirPath, "r") + try { + _fsyncFile(dirFd) + } finally { + fsSync.closeSync(dirFd) + } + } catch { + // best-effort: the content rename already committed + } + } + + // -- Step 5 (win32): restore DACL AFTER commit rename --------- + if (platform === "win32" && daclDumpPath !== null) { + const restoredDir = path.dirname(targetPath) + await _restoreDaclWindows(restoredDir, daclDumpPath, options?.execFileRunner) + } + + // -- Step 6 (backup:true): delete backup on success ----------- + if (releaseBackupOnSuccess && backupPath) { + try { + await fs.unlink(backupPath) + } catch { + // non-fatal — orphaned backup is acceptable + } + } + } finally { + // Unlink DACL dump regardless of success/failure in this span. + if (daclDumpPath !== null) { + await fs.unlink(daclDumpPath).catch(() => {}) + } + } + + // tempPath is now the committed file; no cleanup needed. + } catch (originalError: unknown) { + // -- Rollback / cleanup on failure ---------------------------------- + if (backupPath && releaseBackupOnSuccess) { + try { + await fs.rename(backupPath, targetPath) + } catch { + // rollback failed — do not mask original error + } + } + + // Always clean up the staging temp file on failure. + try { + await fs.unlink(tempPath).catch(() => {}) + } catch { + // cleanup failure is non-fatal + } + + if (daclDumpPath !== null) { + await fs.unlink(daclDumpPath).catch(() => {}) + } + + throw originalError + } +} diff --git a/src/services/mcp/McpHub.ts b/src/services/mcp/McpHub.ts index 1374e430fe..42786cfaa5 100644 --- a/src/services/mcp/McpHub.ts +++ b/src/services/mcp/McpHub.ts @@ -506,13 +506,32 @@ export class McpHub { ) const fileExists = await fileExistsAtPath(mcpSettingsFilePath) if (!fileExists) { - await fs.writeFile( + // Create the default settings file under the advisory lock. The merge + // callback preserves any config a concurrent process wrote between the + // existence check above and the locked read, instead of blindly + // truncating it (see #1371). + await safeWriteJson( mcpSettingsFilePath, - `{ - "mcpServers": { - - } -}`, + { mcpServers: {} }, + { + prettyPrint: true, + merge: (existing) => { + const parsed = existing as { mcpServers?: unknown } | null + // Arrays satisfy `typeof === "object"` but are not a valid + // mcpServers map; preserve only a plain object, otherwise the + // file would be rewritten with a value McpSettingsSchema + // rejects on the next load. + if ( + parsed && + parsed.mcpServers && + !Array.isArray(parsed.mcpServers) && + typeof parsed.mcpServers === "object" + ) { + return existing + } + return { mcpServers: {} } + }, + }, ) } return mcpSettingsFilePath diff --git a/src/services/mcp/__tests__/McpHub.spec.ts b/src/services/mcp/__tests__/McpHub.spec.ts index 96589d8dd6..116c87c1ec 100644 --- a/src/services/mcp/__tests__/McpHub.spec.ts +++ b/src/services/mcp/__tests__/McpHub.spec.ts @@ -1,4 +1,5 @@ import * as fs from "fs/promises" +import * as path from "path" import type { Mock } from "vitest" import type { ExtensionContext, Uri } from "vscode" @@ -34,12 +35,44 @@ import { safeWriteJson } from "../../../utils/safeWriteJson" // Mock safeWriteJson vi.mock("../../../utils/safeWriteJson", () => ({ - safeWriteJson: vi.fn(async (filePath, data) => { - // Instead of trying to write to the file system, just call fs.writeFile mock - // This avoids the complex file locking and temp file operations - const fs = await import("fs/promises") - return fs.writeFile(filePath, JSON.stringify(data), "utf8") - }), + safeWriteJson: vi.fn( + async ( + filePath: string, + data: unknown, + options?: { merge?: (existing: unknown, incoming: unknown) => unknown }, + ) => { + // Instead of trying to write to the file system, just call fs.writeFile mock + // This avoids the complex file locking and temp file operations. + // When a merge callback is provided, honor it: read the current on-disk + // content via the fs.readFile mock (simulating the read under the lock) + // and let the callback decide the final value. + let value = data + if (options?.merge) { + let existing: unknown = null + try { + const fs = await import("fs/promises") + existing = JSON.parse(await fs.readFile(filePath, "utf8")) + } catch (error) { + // Mirror the production safeWriteJson merge contract: only ENOENT + // and SyntaxError are recoverable; an EACCES or I/O failure must + // reject before the merge callback runs. + // unknown-safe narrowing: no cast on the caught value (the "in" + // check narrows to object & Record<"code", unknown>). + const code = + error && typeof error === "object" && "code" in error && typeof error.code === "string" + ? error.code + : undefined + if (!(error instanceof SyntaxError) && code !== "ENOENT") { + throw error + } + existing = null + } + value = options.merge(existing, data) + } + const fs = await import("fs/promises") + return fs.writeFile(filePath, JSON.stringify(value), "utf8") + }, + ), })) vi.mock("delay", () => ({ default: vi.fn().mockResolvedValue(undefined) })) @@ -212,6 +245,122 @@ describe("McpHub", () => { watchSpy.mockRestore() }) + describe("getMcpSettingsFilePath", () => { + it("preserves a config written by a concurrent process during initial creation (#1371)", async () => { + const settingsPath = path.join("/mock/settings/path", "mcp_settings.json") + const concurrentConfig = { + mcpServers: { + "concurrent-server": { type: "stdio", command: "node", args: ["server.js"] }, + }, + } + + // Window A's existence check sees the settings file as absent... + // (One-shot overrides: the factory defaults apply to all other tests.) + vi.mocked(fs.access).mockRejectedValueOnce( + Object.assign(new Error("ENOENT: no such file or directory"), { code: "ENOENT" }), + ) + // ...but by the time the locked read runs (safeWriteJson merge), + // window B's config is already on disk. + vi.mocked(fs.readFile).mockResolvedValueOnce(JSON.stringify(concurrentConfig)) + + const returnedPath = await mcpHub.getMcpSettingsFilePath() + + expect(returnedPath).toBe(settingsPath) + // The creation write must carry the concurrent config, not the empty stub. + expect(fs.writeFile).toHaveBeenCalledTimes(1) + const [writtenPath, writtenData] = vi.mocked(fs.writeFile).mock.calls[0] + expect(writtenPath).toBe(settingsPath) + expect(JSON.parse(writtenData as string)).toEqual(concurrentConfig) + }) + + it("writes the default stub when no settings file exists yet", async () => { + const settingsPath = path.join("/mock/settings/path", "mcp_settings.json") + + // Existence check and the locked read both see an absent file. + vi.mocked(fs.access).mockRejectedValueOnce( + Object.assign(new Error("ENOENT: no such file or directory"), { code: "ENOENT" }), + ) + vi.mocked(fs.readFile).mockRejectedValueOnce( + Object.assign(new Error("ENOENT: no such file or directory"), { code: "ENOENT" }), + ) + + const returnedPath = await mcpHub.getMcpSettingsFilePath() + + expect(returnedPath).toBe(settingsPath) + expect(fs.writeFile).toHaveBeenCalledTimes(1) + const [writtenPath, writtenData] = vi.mocked(fs.writeFile).mock.calls[0] + expect(writtenPath).toBe(settingsPath) + expect(JSON.parse(writtenData as string)).toEqual({ mcpServers: {} }) + }) + + it("writes the default stub when the existing content has no mcpServers object", async () => { + const settingsPath = path.join("/mock/settings/path", "mcp_settings.json") + + // Existence check sees an absent file, but the locked read finds content + // that does not carry a mcpServers object (e.g. a torn or foreign write). + vi.mocked(fs.access).mockRejectedValueOnce( + Object.assign(new Error("ENOENT: no such file or directory"), { code: "ENOENT" }), + ) + vi.mocked(fs.readFile).mockResolvedValueOnce(JSON.stringify({ someOtherKey: true })) + + await mcpHub.getMcpSettingsFilePath() + + expect(fs.writeFile).toHaveBeenCalledTimes(1) + const [, writtenData] = vi.mocked(fs.writeFile).mock.calls[0] + expect(JSON.parse(writtenData as string)).toEqual({ mcpServers: {} }) + }) + + it("writes the default stub when the existing mcpServers value is not an object", async () => { + const settingsPath = path.join("/mock/settings/path", "mcp_settings.json") + + vi.mocked(fs.access).mockRejectedValueOnce( + Object.assign(new Error("ENOENT: no such file or directory"), { code: "ENOENT" }), + ) + vi.mocked(fs.readFile).mockResolvedValueOnce(JSON.stringify({ mcpServers: "corrupted" })) + + await mcpHub.getMcpSettingsFilePath() + + expect(fs.writeFile).toHaveBeenCalledTimes(1) + const [, writtenData] = vi.mocked(fs.writeFile).mock.calls[0] + expect(JSON.parse(writtenData as string)).toEqual({ mcpServers: {} }) + }) + + it("writes the default stub when the existing mcpServers value is an array", async () => { + const settingsPath = path.join("/mock/settings/path", "mcp_settings.json") + + // Arrays satisfy `typeof === "object"`; an mcpServers map must be a + // plain object, so an array is invalid and replaced by the stub + // instead of being preserved and rejected by McpSettingsSchema later. + vi.mocked(fs.access).mockRejectedValueOnce( + Object.assign(new Error("ENOENT: no such file or directory"), { code: "ENOENT" }), + ) + vi.mocked(fs.readFile).mockResolvedValueOnce(JSON.stringify({ mcpServers: [] })) + + await mcpHub.getMcpSettingsFilePath() + + expect(fs.writeFile).toHaveBeenCalledTimes(1) + const [, writtenData] = vi.mocked(fs.writeFile).mock.calls[0] + expect(JSON.parse(writtenData as string)).toEqual({ mcpServers: {} }) + }) + + it("rejects creation when the locked read fails with an I/O error (EACCES)", async () => { + const settingsPath = path.join("/mock/settings/path", "mcp_settings.json") + + vi.mocked(fs.access).mockRejectedValueOnce( + Object.assign(new Error("ENOENT: no such file or directory"), { code: "ENOENT" }), + ) + // The locked read fails with a real I/O error (not ENOENT): the + // safeWriteJson mock mirrors the production contract — reject before + // the merge callback runs instead of treating the file as absent. + vi.mocked(fs.readFile).mockRejectedValueOnce( + Object.assign(new Error("EACCES: permission denied"), { code: "EACCES" }), + ) + + await expect(mcpHub.getMcpSettingsFilePath()).rejects.toThrow("EACCES: permission denied") + expect(fs.writeFile).not.toHaveBeenCalled() + }) + }) + describe("Discriminated union type handling", () => { it("should create connected connections with proper type", async () => { // Mock StdioClientTransport diff --git a/src/shared/__tests__/experiments-preventFocusDisruption.spec.ts b/src/shared/__tests__/experiments-preventFocusDisruption.spec.ts index e9f96c7ce7..44a6e237c0 100644 --- a/src/shared/__tests__/experiments-preventFocusDisruption.spec.ts +++ b/src/shared/__tests__/experiments-preventFocusDisruption.spec.ts @@ -5,13 +5,13 @@ describe("PREVENT_FOCUS_DISRUPTION experiment", () => { expect(EXPERIMENT_IDS.PREVENT_FOCUS_DISRUPTION).toBe("preventFocusDisruption") }) - it("should have PREVENT_FOCUS_DISRUPTION in experimentConfigsMap", () => { + it("should have PREVENT_FOCUS_DISRUPTION enabled by default (chat-diff is the default approval path)", () => { expect(experimentConfigsMap.PREVENT_FOCUS_DISRUPTION).toBeDefined() - expect(experimentConfigsMap.PREVENT_FOCUS_DISRUPTION.enabled).toBe(false) + expect(experimentConfigsMap.PREVENT_FOCUS_DISRUPTION.enabled).toBe(true) }) - it("should have PREVENT_FOCUS_DISRUPTION in experimentDefault", () => { - expect(experimentDefault.preventFocusDisruption).toBe(false) + it("should have PREVENT_FOCUS_DISRUPTION enabled in experimentDefault", () => { + expect(experimentDefault.preventFocusDisruption).toBe(true) }) it("should correctly check if PREVENT_FOCUS_DISRUPTION is enabled", () => { @@ -23,8 +23,8 @@ describe("PREVENT_FOCUS_DISRUPTION experiment", () => { const enabledConfig = { preventFocusDisruption: true } expect(experiments.isEnabled(enabledConfig, EXPERIMENT_IDS.PREVENT_FOCUS_DISRUPTION)).toBe(true) - // Test when experiment is not in config (should use default) + // Test when experiment is not in config (should use default — now enabled) const emptyConfig = {} - expect(experiments.isEnabled(emptyConfig, EXPERIMENT_IDS.PREVENT_FOCUS_DISRUPTION)).toBe(false) + expect(experiments.isEnabled(emptyConfig, EXPERIMENT_IDS.PREVENT_FOCUS_DISRUPTION)).toBe(true) }) }) diff --git a/src/shared/__tests__/experiments.spec.ts b/src/shared/__tests__/experiments.spec.ts index f2261a5c09..6e2f75090c 100644 --- a/src/shared/__tests__/experiments.spec.ts +++ b/src/shared/__tests__/experiments.spec.ts @@ -6,10 +6,10 @@ import { EXPERIMENT_IDS, experimentConfigsMap, experiments as Experiments } from describe("experiments", () => { describe("PREVENT_FOCUS_DISRUPTION", () => { - it("is configured correctly", () => { + it("is configured correctly (chat-diff is the default approval path)", () => { expect(EXPERIMENT_IDS.PREVENT_FOCUS_DISRUPTION).toBe("preventFocusDisruption") expect(experimentConfigsMap.PREVENT_FOCUS_DISRUPTION).toMatchObject({ - enabled: false, + enabled: true, }) }) }) diff --git a/src/shared/experiments.ts b/src/shared/experiments.ts index ae538b9138..f8fb132394 100644 --- a/src/shared/experiments.ts +++ b/src/shared/experiments.ts @@ -19,7 +19,10 @@ interface ExperimentConfig { } export const experimentConfigsMap: Record = { - PREVENT_FOCUS_DISRUPTION: { enabled: false }, + // Chat-diff is the default approval path (plan issue #33, L2): writes are saved + // without opening/refocusing the diff editor. The diff-editor path remains + // available by toggling this experiment off (storage key unchanged). + PREVENT_FOCUS_DISRUPTION: { enabled: true }, IMAGE_GENERATION: { enabled: false }, RUN_SLASH_COMMAND: { enabled: false }, CUSTOM_TOOLS: { enabled: false }, diff --git a/src/utils/__tests__/safeWriteJson.test.ts b/src/utils/__tests__/safeWriteJson.test.ts index 79d08678a0..a52cfef8b3 100644 --- a/src/utils/__tests__/safeWriteJson.test.ts +++ b/src/utils/__tests__/safeWriteJson.test.ts @@ -4,6 +4,7 @@ import * as path from "path" import * as os from "os" import { safeWriteJson } from "../safeWriteJson" +import * as lockfile from "proper-lockfile" // Capture actual implementations before the vi.mock factory runs, // so they are never wrapped by vi.fn() — avoids infinite recursion when @@ -312,9 +313,8 @@ describe("safeWriteJson", () => { expect(content).toEqual(newData) }) - // Test for console error suppression during backup deletion - test("should suppress console.error when backup deletion fails", async () => { - const consoleErrorSpy = vi.spyOn(console, "error").mockImplementation(() => {}) // Suppress console.error + // Test for best-effort backup deletion (the backup lifecycle now lives in safeWriteText) + test("does not fail the write when backup deletion fails (orphaned backup is acceptable)", async () => { const initialData = { message: "Initial" } const newData = { message: "New" } @@ -322,18 +322,23 @@ describe("safeWriteJson", () => { // fs.unlink is already vi.fn() — use vi.mocked to avoid double-wrapping via vi.spyOn vi.mocked(fs.unlink).mockImplementation(async (filePath: any) => { - if (filePath.toString().includes(".bak_")) { + if (filePath.toString().includes("safeWriteText.bak_")) { throw new Error("Backup deletion failed") } return fsPromisesActuals.unlink!(filePath) }) + // The write must still succeed: backup cleanup is best-effort inside + // safeWriteText and never masks the committed content. await safeWriteJson(currentTestFilePath, newData) - // Verify console.error was called with the expected message - expect(consoleErrorSpy).toHaveBeenCalledWith(expect.stringContaining("Successfully wrote"), expect.any(Error)) + const content = await readFileContent(currentTestFilePath) + expect(content).toEqual(newData) + + // The orphaned backup is still on disk because its deletion failed. + const entries = await fs.readdir(tempDir) + expect(entries.some((entry) => entry.includes("safeWriteText.bak_"))).toBe(true) - consoleErrorSpy.mockRestore() vi.mocked(fs.unlink).mockRestore() }) @@ -385,7 +390,10 @@ describe("safeWriteJson", () => { // Clean up await fs.unlink(lockTestFilePath).catch(() => {}) // Ignore errors if file doesn't exist - vi.unmock("proper-lockfile") // Ensure the mock is removed after this test + // A hoisted vi.unmock runs before this test's runtime vi.doMock, so it + // cannot remove it; doUnmock + resetModules clear the registry entry. + vi.doUnmock("proper-lockfile") + vi.resetModules() }) test("should release lock even if an error occurs mid-operation", async () => { const data = { message: "test lock release on error" } @@ -434,9 +442,9 @@ describe("safeWriteJson", () => { expect(vi.mocked(fs.access)).toHaveBeenCalled() }) - // Test for rollback failure scenario - test("should log error and re-throw original if rollback fails", async () => { - const initialData = { message: "Initial, should be lost if rollback fails" } + // Test for rollback failure scenario (the rollback rename now lives in safeWriteText) + test("re-throws the original error when the rollback rename fails, leaving an orphaned backup", async () => { + const initialData = { message: "Initial, orphaned when rollback fails" } const newData = { message: "New content" } await fsPromisesActuals.writeFile!(currentTestFilePath, JSON.stringify(initialData)) @@ -451,20 +459,20 @@ describe("safeWriteJson", () => { // Second call: tempNewFilePath -> filePath (fail) throw new Error("Primary rename failed") } else if (renameCallCount === 3) { - // Third call: tempBackupFilePath -> filePath (rollback, also fail) + // Third call: backup -> filePath (rollback, also fail) throw new Error("Rollback rename failed") } return fsPromisesActuals.rename!(oldPath, newPath) }) - // Should throw the original error, not the rollback error + // The original error must propagate, not the rollback error await expect(safeWriteJson(currentTestFilePath, newData)).rejects.toThrow("Primary rename failed") - // Verify console.error was called for the rollback failure - expect(consoleErrorSpy).toHaveBeenCalledWith( - expect.stringContaining("Failed to restore backup"), - expect.objectContaining({ message: "Rollback rename failed" }), - ) + // The rollback failed inside safeWriteText, so the target is gone and + // the backup is orphaned on disk. + expect(await fileExists(currentTestFilePath)).toBe(false) + const entries = await fs.readdir(tempDir) + expect(entries.some((entry) => entry.includes("safeWriteText.bak_"))).toBe(true) consoleErrorSpy.mockRestore() }) @@ -542,4 +550,133 @@ describe("safeWriteJson", () => { const content = await readFileContent(currentTestFilePath) expect(content).toEqual({ c: 3 }) }) + + // The commit rename targets the symlink referent. The staged temp file must + // therefore be created beside the RESOLVED target — staging beside the link + // would make the commit rename fail with EXDEV when the referent is on + // another filesystem. (Real symlinks are unavailable in this CI lane, so the + // resolution is simulated by mocking fs.realpath the same way.) + test("stages the temp file beside the symlink referent and commits onto it", async () => { + const referentDir = path.join(tempDir, "referent") + const linkDir = path.join(tempDir, "link") + await fs.mkdir(referentDir, { recursive: true }) + await fs.mkdir(linkDir, { recursive: true }) + // caller-visible path (the link) vs the resolved referent path + const callerPath = path.join(linkDir, "test-file.json") + const referentPath = path.join(referentDir, "test-file.json") + // Seed the RESOLVED referent with real content (via the actual fs) so the + // write exercises replacement of an EXISTING referent: the lock is + // acquired on the caller path (realpath:false, which may be absent) while + // the backup + commit happen on the referent. + await fsPromisesActuals.writeFile!(referentPath, JSON.stringify({ seed: true })) + + vi.spyOn(fs, "realpath").mockResolvedValue(referentPath) + + await safeWriteJson(callerPath, { after: true }) + + // the temp file was created next to the resolved referent, NOT beside the link + const tempPaths = vi.mocked(fsSyncActual.createWriteStream).mock.calls.map((call) => String(call[0])) + expect(tempPaths.some((p) => p.startsWith(referentDir + path.sep) && p.includes(".new_"))).toBe(true) + expect(tempPaths.some((p) => p.startsWith(linkDir + path.sep))).toBe(false) + + // the content was committed onto the referent + expect(await readFileContent(referentPath)).toEqual({ after: true }) + }) + + // proper-lockfile with realpath:false keys the lock by the given path, so a + // symlink alias and its referent must coordinate through ONE lock on the + // resolved referent — otherwise a concurrent merge through both aliases + // reads the same JSON and overwrites one update. (Real symlinks are + // unavailable in this CI lane, so the resolution is simulated by mocking + // fs.realpath, the same way as the staging test above.) + test("acquires the lock on the resolved referent, not the caller alias", async () => { + vi.resetModules() // fresh module instances so the doMock below is picked up + + const referentDir = path.join(tempDir, "lock-referent") + const linkDir = path.join(tempDir, "lock-link") + await fs.mkdir(referentDir, { recursive: true }) + await fs.mkdir(linkDir, { recursive: true }) + // caller-visible path (the link) vs the resolved referent path + const callerPath = path.join(linkDir, "locked.json") + const referentPath = path.join(referentDir, "locked.json") + await fsPromisesActuals.writeFile!(referentPath, JSON.stringify({ seed: 1 })) + + vi.spyOn(fs, "realpath").mockResolvedValue(referentPath) + + // Wrap the real lock in a capturing mock, and drive the two rare error paths + // (the onCompromised callback and a failing release) so they stay covered + // without real lockfile staleness. The callback rethrows by design, so + // the mock swallows that throw and lets the real lock proceed. + const realLockfile = await vi.importActual("proper-lockfile") + const lockMockFn = vi.fn( + async ( + file: Parameters[0], + options?: Parameters[1], + ) => { + try { + options?.onCompromised?.(new Error("lock compromised (test)")) + } catch { + // onCompromised rethrows by design; swallow so the real lock proceeds. + } + const release = await realLockfile.lock(file, options) + return async () => { + await release() + throw new Error("release failed (test)") + } + }, + ) + const lockMock = lockMockFn as unknown as typeof realLockfile.lock + vi.doMock("proper-lockfile", () => ({ + ...realLockfile, + lock: lockMock, + })) + + // Re-import safeWriteJson so it picks up the mocked proper-lockfile. + const { safeWriteJson: mockedSafeWriteJson } = await import("../safeWriteJson") + + const mergeFn = vi.fn((existing: unknown, incoming: unknown) => ({ + ...(existing as Record), + ...(incoming as Record), + })) + + // Capture the compromise + release-failure logs. + const consoleErrorSpy = vi.spyOn(console, "error") + await mockedSafeWriteJson(callerPath, { added: true }, { merge: mergeFn }) + + // The lock was keyed by the resolved referent — every alias shares it. + expect(lockMock).toHaveBeenCalledTimes(1) + expect(String(lockMockFn.mock.calls[0][0])).toBe(referentPath) + // The merge read the referent's content through that single lock. + expect(mergeFn).toHaveBeenCalledWith({ seed: 1 }, { added: true }) + expect(await readFileContent(referentPath)).toEqual({ seed: 1, added: true }) + // The compromise callback and the failed release were logged, not thrown. + expect(consoleErrorSpy).toHaveBeenCalledWith(expect.stringContaining("was compromised"), expect.any(Error)) + expect(consoleErrorSpy).toHaveBeenCalledWith( + expect.stringContaining("Failed to release lock"), + expect.any(Error), + ) + + // The hoisted vi.unmock runs before this test's runtime vi.doMock, so it + // cannot remove it; doUnmock + resetModules clear the registry entry so + // later test files import the real proper-lockfile. + vi.doUnmock("proper-lockfile") + vi.resetModules() + }) + + // CWE-732 regression: safeWriteJson stages the temp itself and passes it + // via tempPath, so safeWriteText must apply the existing target's mode to + // the staged temp before the atomic rename — otherwise a 0o600 target is + // published as 0o644. POSIX-only assertion (Windows ignores POSIX modes). + test.skipIf(process.platform === "win32")( + "preserves a restrictive 0o600 target mode through the atomic publish", + async () => { + await fsPromisesActuals.writeFile!(currentTestFilePath, JSON.stringify({ before: true })) + fsSyncActual.chmodSync(currentTestFilePath, 0o600) + + await safeWriteJson(currentTestFilePath, { after: true }) + + expect(fsSyncActual.statSync(currentTestFilePath).mode & 0o777).toBe(0o600) + expect(await readFileContent(currentTestFilePath)).toEqual({ after: true }) + }, + ) }) diff --git a/src/utils/__tests__/versionToken.spec.ts b/src/utils/__tests__/versionToken.spec.ts new file mode 100644 index 0000000000..3e2ca26b5c --- /dev/null +++ b/src/utils/__tests__/versionToken.spec.ts @@ -0,0 +1,105 @@ +import * as fs from "fs/promises" +import * as os from "os" +import * as path from "path" +import type { BigIntStats } from "fs" +import { afterEach, beforeEach, describe, expect, it } from "vitest" + +import { computeVersionToken, versionTokenOfStat } from "../versionToken" + +// BigIntStats is a class-backed interface without a public constructor, so a +// plain-object test double is the only practical way to pin the token format +// without real files. Last-resort double assertion (test-local, per AGENTS.md). +function makeStats(overrides: Partial = {}): BigIntStats { + // This repo's @types/node models every StatsBase field (including the *Ms + // fields) as the parameter type T, so all values here are bigint literals; + // the token only reads the *Ns fields. Single-step downcast from Partial to + // the full type (BigIntStats has no public constructor). + const base: Partial = { + dev: 7n, + ino: 4242n, + size: 1234n, + atimeMs: 1_700_000_000_000n, + mtimeMs: 1_700_000_000_123n, + ctimeMs: 1_700_000_000_789n, + birthtimeMs: 1_700_000_000_000n, + atimeNs: 1_700_000_000_000_000_000n, + mtimeNs: 1_700_000_000_123_456_789n, + ctimeNs: 1_700_000_000_789_999_999n, + birthtimeNs: 1_700_000_000_000_000_000n, + } + return { ...base, ...overrides } as BigIntStats +} + +describe("versionTokenOfStat (A1, epic #1375)", () => { + it("is deterministic for an identical stat", () => { + expect(versionTokenOfStat(makeStats())).toBe(versionTokenOfStat(makeStats())) + }) + + it("matches the documented dev:ino:size:mtimeNs:ctimeNs format with exact decimal fields", () => { + expect(versionTokenOfStat(makeStats())).toBe("7:4242:1234:1700000000123456789:1700000000789999999") + }) + + it("distinguishes size changes at identical timestamps", () => { + expect(versionTokenOfStat(makeStats({ size: 1235n }))).not.toBe(versionTokenOfStat(makeStats())) + }) + + it("distinguishes a one-nanosecond mtime change", () => { + expect(versionTokenOfStat(makeStats({ mtimeNs: 1_700_000_000_123_456_790n }))).not.toBe( + versionTokenOfStat(makeStats()), + ) + }) + + it("distinguishes a replaced file (dev/ino change) with identical content state", () => { + const replaced = makeStats({ dev: 8n, ino: 999n }) + expect(versionTokenOfStat(replaced)).not.toBe(versionTokenOfStat(makeStats())) + }) + + it("renders nanosecond resolution exactly (no float quantization)", () => { + const base = versionTokenOfStat(makeStats()) + const plusOneMicrosecond = versionTokenOfStat(makeStats({ mtimeNs: 1_700_000_000_123_457_789n })) + // 1_000 ns apart — the BigInt derivation must keep the delta exact. + const baseNs = BigInt(base.split(":")[3]) + const microNs = BigInt(plusOneMicrosecond.split(":")[3]) + expect(microNs - baseNs).toBe(1_000n) + }) + + it("handles sizes beyond Number.MAX_SAFE_INTEGER without precision loss", () => { + const size = 10_000_000_000_000_001n // 10^16 + 1 > 2^53 + const token = versionTokenOfStat(makeStats({ size })) + expect(token).toBe(`7:4242:${size}:1700000000123456789:1700000000789999999`) + }) +}) + +describe("computeVersionToken (A1, epic #1375)", () => { + let tmpDir: string + let file: string + + beforeEach(async () => { + tmpDir = await fs.mkdtemp(path.join(os.tmpdir(), "version-token-")) + file = path.join(tmpDir, "seed.txt") + await fs.writeFile(file, "seed content", "utf8") + }) + + afterEach(async () => { + await fs.rm(tmpDir, { recursive: true, force: true }) + }) + + it("derives the token from the on-disk state (single bigint stat)", async () => { + const token = await computeVersionToken(file) + expect(token).toBe(versionTokenOfStat(await fs.stat(file, { bigint: true }))) + }) + + it("changes when the file content changes", async () => { + const before = await computeVersionToken(file) + // Different size + a new mtime — both must move the token. + await fs.writeFile(file, "seed content, extended", "utf8") + await new Promise((resolve) => setTimeout(resolve, 5)) + expect(await computeVersionToken(file)).not.toBe(before) + }) + + it("rejects with ENOENT for an absent file", async () => { + await expect(computeVersionToken(path.join(tmpDir, "absent.txt"))).rejects.toMatchObject({ + code: "ENOENT", + }) + }) +}) diff --git a/src/utils/safeWriteJson.ts b/src/utils/safeWriteJson.ts index 957a0bb20f..a9f837fc50 100644 --- a/src/utils/safeWriteJson.ts +++ b/src/utils/safeWriteJson.ts @@ -4,6 +4,8 @@ import * as path from "path" import * as lockfile from "proper-lockfile" import { JsonStreamStringify } from "json-stream-stringify" +import { resolvePublishTarget, safeWriteText, type SafeWriteTextOptions } from "../services/file-safety/safeWriteText" + /** * Options for safeWriteJson function */ @@ -31,7 +33,7 @@ export interface SafeWriteJsonOptions { * Safely writes JSON data to a file. * - Creates parent directories if they don't exist * - Uses 'proper-lockfile' for inter-process advisory locking to prevent concurrent writes to the same path. - * - Writes to a temporary file first. + * - Writes to a temporary file first via JsonStreamStringify streaming. * - If the target file exists, it's backed up before being replaced. * - Attempts to roll back and clean up in case of errors. * - Supports pretty-printing with indentation while maintaining streaming efficiency. @@ -41,7 +43,6 @@ export interface SafeWriteJsonOptions { * @param {SafeWriteJsonOptions} options - Optional configuration for JSON formatting. * @returns {Promise} */ - async function safeWriteJson(filePath: string, data: any, options?: SafeWriteJsonOptions): Promise { const absoluteFilePath = path.resolve(filePath) let releaseLock = async () => {} // Initialized to a no-op @@ -51,22 +52,29 @@ async function safeWriteJson(filePath: string, data: any, options?: SafeWriteJso // Ensure directory structure exists with improved reliability try { - // Create directory with recursive option await fs.mkdir(dirPath, { recursive: true }) - - // Verify directory exists after creation attempt await fs.access(dirPath) } catch (dirError: any) { console.error(`Failed to create or access directory for ${absoluteFilePath}:`, dirError) throw dirError } + // Resolve the publish target BEFORE acquiring the lock: proper-lockfile keys + // the lock by the given path (realpath is false below because the file may + // not exist yet), so a symlink alias and its referent would otherwise take + // two distinct locks for one underlying file — a concurrent merge through + // both aliases could then read the same JSON and overwrite one update. + // Locking the resolved referent coordinates every alias through one lock. + // resolvePublishTarget tolerates a not-yet-existing file (it returns the + // given path on ENOENT), preserving the previous create-from-absent flow. + const resolvedTargetPath = await resolvePublishTarget(absoluteFilePath) + // Acquire the lock before any file operations try { - releaseLock = await lockfile.lock(absoluteFilePath, { + releaseLock = await lockfile.lock(resolvedTargetPath, { stale: LOCK_STALE_MS, update: 10000, // Update mtime every 10 seconds to prevent staleness if operation is long - realpath: false, // the file may not exist yet, which is acceptable + realpath: false, // resolvedTargetPath is already the referent; the file may still not exist yet, which is acceptable retries: { // Configuration for retrying lock acquisition retries: 5, // Number of retries after the initial attempt @@ -75,7 +83,7 @@ async function safeWriteJson(filePath: string, data: any, options?: SafeWriteJso maxTimeout: 1000, // Maximum time to wait for any single retry (in ms) }, onCompromised: (err) => { - console.error(`Lock at ${absoluteFilePath} was compromised:`, err) + console.error(`Lock at ${resolvedTargetPath} was compromised:`, err) throw err }, }) @@ -83,14 +91,12 @@ async function safeWriteJson(filePath: string, data: any, options?: SafeWriteJso // If lock acquisition fails, we throw immediately. // The releaseLock remains a no-op, so the finally block in the main file operations // try-catch-finally won't try to release an unacquired lock if this path is taken. - console.error(`Failed to acquire lock for ${absoluteFilePath}:`, lockError) - // Propagate the lock acquisition error + console.error(`Failed to acquire lock for ${resolvedTargetPath}:`, lockError) throw lockError } - // Variables to hold the actual paths of temp files if they are created. + // Variables to hold the actual path of the temp file if it is created. let actualTempNewFilePath: string | null = null - let actualTempBackupFilePath: string | null = null try { // If a merge callback was provided, read the current file under the lock @@ -99,7 +105,7 @@ async function safeWriteJson(filePath: string, data: any, options?: SafeWriteJso if (options?.merge) { let existing: unknown = null try { - existing = JSON.parse(await fs.readFile(absoluteFilePath, "utf8")) + existing = JSON.parse(await fs.readFile(resolvedTargetPath, "utf8")) } catch (error: unknown) { const code = error && typeof error === "object" && "code" in error ? (error as { code: string }).code : undefined @@ -110,79 +116,42 @@ async function safeWriteJson(filePath: string, data: any, options?: SafeWriteJso data = options.merge(existing, data) } - // Step 1: Write data to a new temporary file. + // Step 1: Write data to a new temporary file via JSON streaming. + // Stage it beside the *resolved* target (the symlink referent when the path is + // a symlink; resolvedTargetPath above): safeWriteText commits by renaming + // onto that referent, and a rename across filesystems would fail with EXDEV. actualTempNewFilePath = path.join( - path.dirname(absoluteFilePath), - `.${path.basename(absoluteFilePath)}.new_${Date.now()}_${Math.random().toString(36).substring(2)}.tmp`, + path.dirname(resolvedTargetPath), + ".new_" + Date.now() + "_" + Math.random().toString(36).substring(2) + ".tmp", ) await _streamDataToFile(actualTempNewFilePath, data, options?.prettyPrint) - // Step 2: Check if the target file exists. If so, rename it to a backup path. - try { - // Check for target file existence - await fs.access(absoluteFilePath) - // Target exists, create a backup path and rename. - actualTempBackupFilePath = path.join( - path.dirname(absoluteFilePath), - `.${path.basename(absoluteFilePath)}.bak_${Date.now()}_${Math.random().toString(36).substring(2)}.tmp`, - ) - await fs.rename(absoluteFilePath, actualTempBackupFilePath) - } catch (accessError: any) { - // Explicitly type accessError - if (accessError.code !== "ENOENT") { - // An error other than "file not found" occurred during access check. - throw accessError - } - // Target file does not exist, so no backup is made. actualTempBackupFilePath remains null. + // Step 2: Delegate backup + commit + rollback to safeWriteText with the + // pre-written temp path. backup:true keeps the old safeWriteJson + // semantics (target -> backup before commit, rollback on failure) and + // keeps the target in place until safeWriteText captures its Windows + // DACL (safeWriteText dumps the DACL before its own backup rename and + // restores it onto the directory after the commit rename). + const textOptions: SafeWriteTextOptions = { + tempPath: actualTempNewFilePath, + backup: true, } - // Step 3: Rename the new temporary file to the target file path. - // This is the main "commit" step. - await fs.rename(actualTempNewFilePath, absoluteFilePath) + await safeWriteText(resolvedTargetPath, "", textOptions) - // If we reach here, the new file is successfully in place. - // The original actualTempNewFilePath is now the main file, so we shouldn't try to clean it up as "temp". - // Mark as "used" or "committed" + // If we reach here, the new file is successfully in place and any + // backup has already been handled by safeWriteText. actualTempNewFilePath = null - - // Step 4: If a backup was created, attempt to delete it. - if (actualTempBackupFilePath) { - try { - await fs.unlink(actualTempBackupFilePath) - // Mark backup as handled - actualTempBackupFilePath = null - } catch (unlinkBackupError) { - // Log this error, but do not re-throw. The main operation was successful. - // actualTempBackupFilePath remains set, indicating an orphaned backup. - console.error( - `Successfully wrote ${absoluteFilePath}, but failed to clean up backup ${actualTempBackupFilePath}:`, - unlinkBackupError, - ) - } - } } catch (originalError) { - console.error(`Operation failed for ${absoluteFilePath}: [Original Error Caught]`, originalError) + console.error(`Operation failed for ${resolvedTargetPath}: [Original Error Caught]`, originalError) const newFileToCleanupWithinCatch = actualTempNewFilePath - const backupFileToRollbackOrCleanupWithinCatch = actualTempBackupFilePath - // Attempt rollback if a backup was made - if (backupFileToRollbackOrCleanupWithinCatch) { - try { - await fs.rename(backupFileToRollbackOrCleanupWithinCatch, absoluteFilePath) - // Mark as handled, prevent later unlink of this path - actualTempBackupFilePath = null - } catch (rollbackError) { - // actualTempBackupFilePath (outer scope) remains pointing to backupFileToRollbackOrCleanupWithinCatch - console.error( - `[Catch] Failed to restore backup ${backupFileToRollbackOrCleanupWithinCatch} to ${absoluteFilePath}:`, - rollbackError, - ) - } - } - - // Cleanup the .new file if it exists + // A failed safeWriteText already rolled the backup (if any) back to + // the target path. Clean up the .new file if it still exists + // (safeWriteText also cleans up its tempPath on failure; this is a + // safety net in case its cleanup missed it). if (newFileToCleanupWithinCatch) { try { await fs.unlink(newFileToCleanupWithinCatch) @@ -194,27 +163,13 @@ async function safeWriteJson(filePath: string, data: any, options?: SafeWriteJso } } - // Cleanup the .bak file if it still needs to be (i.e., wasn't successfully restored) - if (actualTempBackupFilePath) { - try { - await fs.unlink(actualTempBackupFilePath) - } catch (cleanupError) { - console.error( - `[Catch] Failed to clean up temporary backup file ${actualTempBackupFilePath}:`, - cleanupError, - ) - } - } throw originalError // This MUST be the error that rejects the promise. } finally { // Release the lock in the main finally block. try { - // releaseLock will be the actual unlock function if lock was acquired, - // or the initial no-op if acquisition failed. await releaseLock() } catch (unlockError) { - // Do not re-throw here, as the originalError from the try/catch (if any) is more important. - console.error(`Failed to release lock for ${absoluteFilePath}:`, unlockError) + console.error(`Failed to release lock for ${resolvedTargetPath}:`, unlockError) } } } diff --git a/src/utils/versionToken.ts b/src/utils/versionToken.ts new file mode 100644 index 0000000000..1738280087 --- /dev/null +++ b/src/utils/versionToken.ts @@ -0,0 +1,48 @@ +import { stat } from "fs/promises" +import type { BigIntStats } from "fs" + +/** + * Version token for the compare-and-swap write guard (upstream epic #1375, phase A1). + * + * A token is a pure function of a file's on-disk state, derived from a single + * `fs.stat(path, { bigint: true })`, so every process that observes the same file + * state (a second VS Code window, the CLI, the user's own editor tooling) computes + * the same token. The downstream guard phases (A2/A3) compare the token observed at + * read time with the token recomputed just before a write to detect "the file + * changed since the read" (stale) or "the file was replaced by a different file" + * (dev/ino change). + * + * Format: `dev:ino:size:mtimeNs:ctimeNs` + * + * Precision: the stat is fetched in `bigint` mode, so all five fields are exact + * `BigInt` values rendered as decimal strings — no float is involved anywhere. + * There is therefore no precision loss for large sizes or inodes (a Windows file ID + * exceeds 2^53 and is still exact), and the ns timestamps are the kernel's exact + * nanosecond values rather than a ms→ns derivation (no ~256 ns double-precision + * quantum). Guarantee: same disk state → same token, deterministic across + * processes; any change to size, file identity, or mtime/ctime → a different token. + * + * Platform note: on POSIX `ctime` is the last file-status change; on Windows it is + * the file creation time. The token only requires it to move when the file's + * metadata is replaced, which holds on both. + */ + +/** + * Build the version token from an already-fetched `BigIntStats` — no I/O. + * + * Exported separately from {@link computeVersionToken} so tests can pin the exact + * format against synthetic stats. + */ +export function versionTokenOfStat(stats: BigIntStats): string { + return [stats.dev, stats.ino, stats.size, stats.mtimeNs, stats.ctimeNs].map((value) => value.toString()).join(":") +} + +/** + * Compute the version token for a file (one `fs.stat` in bigint mode). + * + * Rejects with the underlying ENOENT (or equivalent) error when the file is absent; + * how an unobservable target is treated is decided by the guard layer (A3). + */ +export async function computeVersionToken(filePath: string): Promise { + return versionTokenOfStat(await stat(filePath, { bigint: true })) +} diff --git a/webview-ui/src/components/chat/ChangeCard.tsx b/webview-ui/src/components/chat/ChangeCard.tsx new file mode 100644 index 0000000000..14986621bc --- /dev/null +++ b/webview-ui/src/components/chat/ChangeCard.tsx @@ -0,0 +1,445 @@ +import { useEffect, useMemo, useState } from "react" +import { useTranslation } from "react-i18next" +import { VSCodeProgressRing } from "@vscode/webview-ui-toolkit/react" +import { Check, FileDiff, History, RotateCcw, X } from "lucide-react" +import { safeJsonParse } from "@roo/core" + +import { changeCardSchema, type ClineMessage, type ExtensionMessage } from "@roo-code/types" + +import { Button, StandardTooltip } from "@/components/ui" +import { vscode } from "@src/utils/vscode" +import { formatPathTooltip } from "@src/utils/formatPathTooltip" + +import CodeAccordion from "../common/CodeAccordion" + +type RollbackStatus = "idle" | "confirming" | "pending" | "success" | "error" + +type RollbackState = { + status: RollbackStatus + error?: string +} + +const IDLE: RollbackState = { status: "idle" } + +const successState: RollbackState = { status: "success" } + +/** + * Per-step change card (B3a payload, B3b UI): header with the file count, a + * per-file list with +/− diff badges, and per-file / per-step restore + * controls wired to the extension host through the + * `checkpointRollbackFile` / `checkpointRollbackStep` / + * `checkpointRestoreLatestFile` messages. A rollback restores a file to the + * state it had BEFORE the step (undoing the step's write); a restore-latest + * brings the file back to its most recent recorded version (the forward + * direction, available from every card of the task). Diffs come from the + * payload's per-file `diff` field: `full` cards expand by default, + * `summary` cards expand lazily on toggle, compact cards carry no diff. + */ +export const ChangeCard = ({ message }: { message: ClineMessage }) => { + const { t } = useTranslation() + + const card = useMemo(() => { + // Validate the shape, not just the JSON-ness: this text is persisted + // task history, so a truncated or pre-series record must not throw + // during render. Records that fail to parse (safeJsonParse returns + // `undefined` without a default) or fail shape validation fall through + // to the null path (an inert card row). + const parsed = safeJsonParse(message.text) + if (parsed === undefined) { + return null + } + const validated = changeCardSchema.safeParse(parsed) + return validated.success ? validated.data : null + }, [message.text]) + + // Files whose diff is currently expanded. "full" cards expand inline by + // default; "summary" cards keep the diff collapsed until toggled. + const [expandedFiles, setExpandedFiles] = useState>(() => { + if (!card || card.detail !== "full") { + return new Set() + } + return new Set(card.files.filter((file) => file.diff != null).map((file) => file.path)) + }) + + const [fileRollbacks, setFileRollbacks] = useState>({}) + const [fileRestores, setFileRestores] = useState>({}) + const [stepRollback, setStepRollback] = useState(IDLE) + + const checkpointId = card?.checkpointIds[0] + + // Correlate extension restore results with this card by message ts. + useEffect(() => { + const handler = (event: MessageEvent) => { + const data = event.data as ExtensionMessage | undefined + const result = data?.checkpointRollbackResult + if (data?.type !== "checkpointRollbackResult" || !result) { + return + } + if (result.cardTs !== message.ts) { + return + } + const filePath = result.filePath + if (filePath !== undefined) { + // Per-file results route to the control that requested them: + // restore-latest results update the restore control; everything + // else (including results posted before `kind` existed) the + // rollback control. + const state: RollbackState = result.success ? successState : { status: "error", error: result.error } + if (result.kind === "restore-latest") { + setFileRestores((prev) => ({ ...prev, [filePath]: state })) + } else { + setFileRollbacks((prev) => ({ ...prev, [filePath]: state })) + } + } + if (result.files) { + const fileUpdates: Record = {} + for (const file of result.files) { + fileUpdates[file.filePath] = file.success ? successState : { status: "error", error: file.error } + } + setFileRollbacks((prev) => ({ ...prev, ...fileUpdates })) + const firstFailure = result.files.find((file) => !file.success) + setStepRollback( + result.success ? successState : { status: "error", error: firstFailure?.error ?? result.error }, + ) + } else if (filePath === undefined) { + // Step-level result with no per-file payload (for example the + // missing-task response: success: false, no files). Without this the + // step button would stay in the in-progress state forever. + setStepRollback(result.success ? successState : { status: "error", error: result.error }) + } + } + window.addEventListener("message", handler) + return () => window.removeEventListener("message", handler) + }, [message.ts]) + + if (!card || !checkpointId) { + return null + } + + const toggleFile = (path: string) => { + setExpandedFiles((prev) => { + const next = new Set(prev) + if (next.has(path)) { + next.delete(path) + } else { + next.add(path) + } + return next + }) + } + + const requestFileRollback = (path: string) => { + vscode.postMessage({ + type: "checkpointRollbackFile", + payload: { cardTs: message.ts, checkpointId, filePath: path }, + }) + setFileRollbacks((prev) => ({ ...prev, [path]: { status: "pending" } })) + } + + const requestFileRestoreLatest = (path: string) => { + vscode.postMessage({ + type: "checkpointRestoreLatestFile", + payload: { cardTs: message.ts, filePath: path }, + }) + setFileRestores((prev) => ({ ...prev, [path]: { status: "pending" } })) + } + + const requestStepRollback = () => { + vscode.postMessage({ + type: "checkpointRollbackStep", + payload: { cardTs: message.ts, checkpointId, filePaths: card.files.map((file) => file.path) }, + }) + setStepRollback({ status: "pending" }) + } + + // Open the changed file in the editor. The extension host resolves relative + // paths against the current cwd (webviewMessageHandler "openFile"), so + // normalize the "./" prefix the same way FileChangesPanel does. + const openFileInEditor = (path: string) => { + vscode.postMessage({ type: "openFile", text: path.startsWith("./") ? path : "./" + path }) + } + + const diffBadges = (additions: number, deletions: number) => + additions > 0 || deletions > 0 ? ( + + +{additions} + -{deletions} + + ) : null + + const fileRollbackControls = (path: string, index: number) => { + const state = fileRollbacks[path] ?? IDLE + const confirmTestId = `change-card-file-confirm-${index}` + const cancelTestId = `change-card-file-cancel-${index}` + + switch (state.status) { + case "confirming": + return ( + + + {t("chat:changeCard.rollbackFileWarning")} + + + + + ) + case "pending": + return ( + + + + ) + case "success": + return ( + + + {t("chat:changeCard.rolledBack")} + + ) + case "error": + return ( + + {/* Focusable status so the error detail is reachable by keyboard and + screen-reader users, not only via the hover tooltip. */} + + + {t("chat:changeCard.rollbackFailed")} + + + ) + default: + return ( + + + + ) + } + } + + // Per-file forward direction: restore the file to its most recent recorded + // version. Independent of the rollback control so a user can undo a + // rollback (or a manual edit) from any card of the task. + const fileRestoreLatestControls = (path: string, index: number) => { + const state = fileRestores[path] ?? IDLE + const confirmTestId = `change-card-file-restore-confirm-${index}` + const cancelTestId = `change-card-file-restore-cancel-${index}` + + switch (state.status) { + case "confirming": + return ( + + + {t("chat:changeCard.restoreLatestWarning")} + + + + + ) + case "pending": + return ( + + + + ) + case "success": + return ( + + + {t("chat:changeCard.restored")} + + ) + case "error": + return ( + + {/* Focusable status (see the file-rollback error span above). */} + + + {t("chat:changeCard.restoreFailed")} + + + ) + default: + return ( + + + + ) + } + } + + const stepRollbackControls = () => { + switch (stepRollback.status) { + case "confirming": + return ( + + + {t("chat:changeCard.rollbackWarning")} + + + + + ) + case "pending": + return ( + + + {t("chat:changeCard.rollingBack")} + + ) + case "success": + return ( + + + {t("chat:changeCard.stepRolledBack")} + + ) + case "error": + return ( + + {/* Focusable status (see the file-rollback error span above). */} + + + {t("chat:changeCard.rollbackFailed")} + + + ) + default: + return ( + + ) + } + } + + return ( +
+
+ + + {t("chat:changeCard.header", { count: card.totalFiles })} + + + {stepRollbackControls()} +
+
+ {card.files.map((file, index) => ( +
+
+ {file.diff != null ? ( + toggleFile(file.path)} + onJumpToFile={() => openFileInEditor(file.path)} + diffStats={{ added: file.additions, removed: file.deletions }} + /> + ) : ( +
+ + {formatPathTooltip(file.path)} + + + {diffBadges(file.additions, file.deletions)} + {/* Native button (not a span) so keyboard users can open the + file from a compact row: Enter/Space activate it for free. */} + +
+ )} +
+
+ {fileRollbackControls(file.path, index)} + {fileRestoreLatestControls(file.path, index)} +
+
+ ))} +
+
+ ) +} diff --git a/webview-ui/src/components/chat/ChatRow.tsx b/webview-ui/src/components/chat/ChatRow.tsx index 952322084f..6d13bae0a0 100644 --- a/webview-ui/src/components/chat/ChatRow.tsx +++ b/webview-ui/src/components/chat/ChatRow.tsx @@ -39,6 +39,7 @@ import McpResourceRow from "../mcp/McpResourceRow" import { Mention } from "./Mention" import { CheckpointSaved } from "./checkpoints/CheckpointSaved" +import { ChangeCard } from "./ChangeCard" import { FollowUpSuggest } from "./FollowUpSuggest" import { BatchFilePermission } from "./BatchFilePermission" import { BatchDiffApproval } from "./BatchDiffApproval" @@ -1374,6 +1375,8 @@ export const ChatRowContent = ({ onJumpToPreviousCheckpoint={onJumpToPreviousCheckpoint} /> ) + case "change_card": + return case "condense_context": // In-progress state if (message.partial) { diff --git a/webview-ui/src/components/chat/__tests__/ChangeCard.spec.tsx b/webview-ui/src/components/chat/__tests__/ChangeCard.spec.tsx new file mode 100644 index 0000000000..92bd18fc21 --- /dev/null +++ b/webview-ui/src/components/chat/__tests__/ChangeCard.spec.tsx @@ -0,0 +1,563 @@ +// npx vitest run src/components/chat/__tests__/ChangeCard.spec.tsx + +import React from "react" +import { fireEvent, renderWithExtensionState, screen } from "@/utils/test-utils" +import type { ChangeCardData, ClineMessage } from "@roo-code/types" + +const mockPostMessage = vi.fn() + +vi.mock("@src/utils/vscode", () => ({ + vscode: { + postMessage: (...args: unknown[]) => mockPostMessage(...args), + }, +})) + +// Mock i18n (same pattern as the other ChatRow specs) +vi.mock("react-i18next", () => ({ + useTranslation: () => ({ + t: (key: string, options?: { count?: number; path?: string }) => { + const map: Record = { + "chat:changeCard.header": `${options?.count ?? 0} file(s) changed this step`, + "chat:changeCard.rollbackFile": "Rollback this file", + "chat:changeCard.rollbackFileWarning": "Restores this file to the content it had before this step.", + "chat:changeCard.restoreLatest": "Restore latest version", + "chat:changeCard.restoreLatestWarning": "Restores this file to the latest recorded version.", + "chat:changeCard.restored": "Restored latest version", + "chat:changeCard.restoreFailed": "Restore failed", + "chat:changeCard.rollbackStep": "Rollback step", + "chat:changeCard.rollbackWarning": "Restores the previous content of this step's files.", + "chat:changeCard.confirm": "Confirm", + "chat:changeCard.cancel": "Cancel", + "chat:changeCard.rollingBack": "Rolling back...", + "chat:changeCard.rolledBack": "Rolled back", + "chat:changeCard.stepRolledBack": "Step rolled back", + "chat:changeCard.rollbackFailed": "Rollback failed", + "chat:changeCard.openFile": "Open file: {{path}}", + } + // {{path}} interpolation: the compact row labels embed the file path + // (aria-label / title), mirroring the CodeAccordion label format. + const value = map[key] || key + return value.replace(/{{path}}/g, String(options?.path ?? "")) + }, + }), + initReactI18next: { type: "3rdParty", init: () => {} }, +})) + +// Mock DiffView so the diff text is directly assertable (the real one runs a +// syntax highlighter, which is irrelevant to the lazy-expansion behavior). +vi.mock("@src/components/common/DiffView", () => ({ + default: ({ source }: { source: string }) =>
{source}
, +})) + +import { ChangeCard } from "../ChangeCard" +import { ChatRowContent } from "../ChatRow" + +function makeCardMessage(overrides: Partial = {}, ts = 1000): ClineMessage { + const card: ChangeCardData = { + checkpointIds: ["abc123"], + files: [ + { path: "src/a.ts", additions: 12, deletions: 3 }, + { path: "src/b.ts", additions: 1, deletions: 1 }, + ], + totalFiles: 2, + detail: "summary", + ...overrides, + } + return { + type: "say", + say: "change_card", + ts, + partial: false, + text: JSON.stringify(card), + } +} + +const DIFF_A = "@@ -1,1 +1,2 @@\n-old\n+new-a\n+extra\n" +const DIFF_B = "@@ -1,1 +1,1 @@\n-old\n+new-b\n" + +function fireRollbackResult(data: Record) { + fireEvent(window, new MessageEvent("message", { data })) +} + +describe("ChangeCard", () => { + beforeEach(() => { + vi.clearAllMocks() + }) + + it("renders the header count and per-file list from a multi-file payload", () => { + renderWithExtensionState() + + expect(screen.getByTestId("change-card-header")).toHaveTextContent("2 file(s) changed this step") + expect(screen.getByText((text) => text.includes("src/a.ts"))).toBeInTheDocument() + expect(screen.getByText((text) => text.includes("src/b.ts"))).toBeInTheDocument() + expect(screen.getByText("+12")).toBeInTheDocument() + expect(screen.getByText("-3")).toBeInTheDocument() + expect(screen.getByText("+1")).toBeInTheDocument() + expect(screen.getByText("-1")).toBeInTheDocument() + }) + + it("keeps the diff hidden in summary cards until the file row is expanded", () => { + renderWithExtensionState( + , + ) + + // Collapsed by default: the diff text is not rendered. + expect(screen.queryByTestId("diff-view")).toBeNull() + expect(screen.queryByText((content) => content.includes("+new-a"))).toBeNull() + + // Expand the file row. + fireEvent.click(screen.getByText((text) => text.includes("src/a.ts"))) + + // The diff text comes from the payload and is rendered lazily on expand. + expect(screen.getByTestId("diff-view").textContent).toContain(DIFF_A.trim()) + + // Collapse again. + fireEvent.click(screen.getByText((text) => text.includes("src/a.ts"))) + expect(screen.queryByTestId("diff-view")).toBeNull() + }) + + it("renders the diff inline by default in full cards", () => { + renderWithExtensionState( + , + ) + + const [diffA, diffB] = screen.getAllByTestId("diff-view") + expect(diffA.textContent).toContain(DIFF_A.trim()) + expect(diffB.textContent).toContain(DIFF_B.trim()) + }) + + it("renders compact rows without a diff section when the payload carries no diffs", () => { + // Auto-approved steps are always emitted as summary cards without any + // per-file diff field; the card then renders file rows with stats only. + renderWithExtensionState() + + expect(screen.getByTestId("change-card-header")).toBeInTheDocument() + expect(screen.queryByTestId("diff-view")).toBeNull() + expect(screen.getByText((text) => text.includes("src/a.ts"))).toBeInTheDocument() + expect(screen.getByText((text) => text.includes("src/b.ts"))).toBeInTheDocument() + }) + + it("renders nothing for an unparseable card payload", () => { + const { container } = renderWithExtensionState( + , + ) + + expect(container.innerHTML).toBe("") + }) + + it("renders nothing for a schema-invalid card payload", () => { + // Syntactically valid JSON that fails changeCardSchema: the + // safeParse rejection path (not only the parse-failure path) must + // also yield an inert card row. + const { container } = renderWithExtensionState( + , + ) + + expect(container.innerHTML).toBe("") + }) + + it("rolls back one file through the checkpointRollbackFile message and shows pending + success", async () => { + renderWithExtensionState() + + // Open the confirm step for the first file. + fireEvent.click(screen.getByTestId("change-card-file-rollback-0")) + expect(screen.getByTestId("change-card-file-confirm-0")).toBeInTheDocument() + + // Confirm sends the webview->extension message and goes pending. + fireEvent.click(screen.getByText("Confirm")) + expect(mockPostMessage).toHaveBeenCalledWith({ + type: "checkpointRollbackFile", + payload: { cardTs: 1000, checkpointId: "abc123", filePath: "src/a.ts" }, + }) + expect(screen.getByTestId("change-card-file-pending-0")).toBeInTheDocument() + + // The extension ack resolves the pending state. + fireRollbackResult({ + type: "checkpointRollbackResult", + checkpointRollbackResult: { cardTs: 1000, filePath: "src/a.ts", success: true }, + }) + await screen.findByTestId("change-card-file-success-0") + expect(screen.getByTestId("change-card-file-success-0")).toHaveTextContent("Rolled back") + }) + + it("shows the file rollback error state on a failed ack", async () => { + renderWithExtensionState() + + fireEvent.click(screen.getByTestId("change-card-file-rollback-0")) + fireEvent.click(screen.getByText("Confirm")) + + fireRollbackResult({ + type: "checkpointRollbackResult", + checkpointRollbackResult: { + cardTs: 1000, + filePath: "src/a.ts", + success: false, + error: "checkpoint not found", + }, + }) + + expect(await screen.findByTestId("change-card-file-error-0")).toHaveTextContent("Rollback failed") + }) + + it("rolls back the whole step through the checkpointRollbackStep message and shows pending + success", async () => { + renderWithExtensionState() + + // Open the confirm step for the step-level rollback. + fireEvent.click(screen.getByTestId("change-card-step-rollback")) + expect(screen.getByTestId("change-card-step-confirm")).toBeInTheDocument() + expect(screen.getByText("Restores the previous content of this step's files.")).toBeInTheDocument() + + // Confirm sends the step message with the step's file list. + fireEvent.click(screen.getByText("Confirm")) + expect(mockPostMessage).toHaveBeenCalledWith({ + type: "checkpointRollbackStep", + payload: { cardTs: 1000, checkpointId: "abc123", filePaths: ["src/a.ts", "src/b.ts"] }, + }) + expect(screen.getByTestId("change-card-step-pending")).toBeInTheDocument() + + // The extension ack (per-step result carries the per-file outcomes). + fireRollbackResult({ + type: "checkpointRollbackResult", + checkpointRollbackResult: { + cardTs: 1000, + success: true, + files: [ + { filePath: "src/a.ts", success: true }, + { filePath: "src/b.ts", success: true }, + ], + }, + }) + expect(await screen.findByTestId("change-card-step-success")).toHaveTextContent("Step rolled back") + // Per-file rows resolve to success as well. + expect(screen.getByTestId("change-card-file-success-0")).toBeInTheDocument() + expect(screen.getByTestId("change-card-file-success-1")).toBeInTheDocument() + }) + + it("shows the step rollback error state with the first failing file's error", async () => { + renderWithExtensionState() + + fireEvent.click(screen.getByTestId("change-card-step-rollback")) + fireEvent.click(screen.getByText("Confirm")) + + fireRollbackResult({ + type: "checkpointRollbackResult", + checkpointRollbackResult: { + cardTs: 1000, + success: false, + files: [ + { filePath: "src/a.ts", success: true }, + { filePath: "src/b.ts", success: false, error: "boom" }, + ], + }, + }) + + expect(await screen.findByTestId("change-card-step-error")).toHaveTextContent("Rollback failed") + expect(screen.getByTestId("change-card-file-error-1")).toBeInTheDocument() + expect(screen.getByTestId("change-card-file-success-0")).toBeInTheDocument() + + // The error detail is exposed to assistive technology: each failed + // control is a focusable status element (not just hover tooltip text). + const fileError = screen.getByTestId("change-card-file-error-1") + expect(fileError).toHaveAttribute("role", "status") + expect(fileError).toHaveAttribute("tabindex", "0") + expect(fileError).toHaveAttribute("aria-label", "boom") + expect(screen.getByTestId("change-card-step-error")).toHaveAttribute("role", "status") + expect(screen.getByTestId("change-card-step-error")).toHaveAttribute("tabindex", "0") + }) + + it("resolves the step state from a failure result that carries no files", async () => { + // The missing-task response is a step-level result with success: false + // and no per-file payload (no files, no filePath). Without handling the + // empty shape the step button would stay in the pending state forever. + renderWithExtensionState() + + fireEvent.click(screen.getByTestId("change-card-step-rollback")) + fireEvent.click(screen.getByText("Confirm")) + expect(screen.getByTestId("change-card-step-pending")).toBeInTheDocument() + + fireRollbackResult({ + type: "checkpointRollbackResult", + checkpointRollbackResult: { + cardTs: 1000, + success: false, + error: "Checkpoints are not enabled for this task", + }, + }) + + // The error detail rides in the tooltip content; the visible state is + // the rollback-failed label. The assertion that matters here is that the + // step left the pending state at all (previously it would stay pending). + const stepError = await screen.findByTestId("change-card-step-error") + expect(stepError).toHaveTextContent("Rollback failed") + // Focusable status with the actual error as its accessible name. + expect(stepError).toHaveAttribute("role", "status") + expect(stepError).toHaveAttribute("tabindex", "0") + expect(stepError).toHaveAttribute("aria-label", "Checkpoints are not enabled for this task") + }) + + it("resolves the step state from a success result that carries no files", async () => { + renderWithExtensionState() + + fireEvent.click(screen.getByTestId("change-card-step-rollback")) + fireEvent.click(screen.getByText("Confirm")) + expect(screen.getByTestId("change-card-step-pending")).toBeInTheDocument() + + fireRollbackResult({ + type: "checkpointRollbackResult", + checkpointRollbackResult: { cardTs: 1000, success: true }, + }) + + expect(await screen.findByTestId("change-card-step-success")).toBeInTheDocument() + }) + + it("ignores rollback results for other change cards", () => { + renderWithExtensionState() + + // Unrelated extension messages are dropped by the card's listener. + fireEvent(window, new MessageEvent("message", { data: { type: "state", text: "x" } })) + + fireEvent.click(screen.getByTestId("change-card-step-rollback")) + fireEvent.click(screen.getByText("Confirm")) + expect(screen.getByTestId("change-card-step-pending")).toBeInTheDocument() + + // A result for a different card ts must not resolve this card. + fireRollbackResult({ + type: "checkpointRollbackResult", + checkpointRollbackResult: { cardTs: 999, success: true, files: [] }, + }) + + expect(screen.getByTestId("change-card-step-pending")).toBeInTheDocument() + }) + + it("cancels the file and step rollback confirmations without sending a message", () => { + renderWithExtensionState() + + // File-level cancel returns to idle without a rollback message. + fireEvent.click(screen.getByTestId("change-card-file-rollback-0")) + fireEvent.click(screen.getByTestId("change-card-file-cancel-0")) + expect(screen.getByTestId("change-card-file-rollback-0")).toBeInTheDocument() + expect(mockPostMessage).not.toHaveBeenCalledWith(expect.objectContaining({ type: "checkpointRollbackFile" })) + + // Step-level cancel returns to idle as well. + fireEvent.click(screen.getByTestId("change-card-step-rollback")) + fireEvent.click(screen.getByTestId("change-card-step-cancel")) + expect(screen.getByTestId("change-card-step-rollback")).toBeInTheDocument() + expect(mockPostMessage).not.toHaveBeenCalledWith(expect.objectContaining({ type: "checkpointRollbackStep" })) + }) + + it("posts an openFile message from both the diff-row jump icon and the no-diff-row button", () => { + renderWithExtensionState( + , + ) + + // Diff row: the CodeAccordion header jump icon (own aria-label). + fireEvent.click(screen.getByLabelText("Open file: src/a.ts")) + expect(mockPostMessage).toHaveBeenCalledWith({ type: "openFile", text: "./src/a.ts" }) + mockPostMessage.mockClear() + + // No-diff row: the open control on the plain path row. + fireEvent.click(screen.getByTestId("change-card-file-open-1")) + expect(mockPostMessage).toHaveBeenCalledWith({ type: "openFile", text: "./src/b.ts" }) + }) + + it("does not double-prefix paths that already carry the ./ marker", () => { + renderWithExtensionState( + , + ) + + fireEvent.click(screen.getByTestId("change-card-file-open-0")) + expect(mockPostMessage).toHaveBeenCalledWith({ type: "openFile", text: "./src/c.ts" }) + }) + + it("renders the no-diff row open control as a native button so keyboard users can activate it", () => { + renderWithExtensionState( + , + ) + + // A native