fix(query): don't modify events that other query variables still hold

[ErikBjare]

Found while measuring the query parity suite (ActivityWatch/activitywatch#1467) for the ActivityWatch/activitywatch#1466 changes. Query variables share Event objects, and several transforms modified their input events in place, so one variable's events changed when another was transformed. aw-server-rust has value semantics, so the servers disagreed on the canonical queries.

period_union cleared .data on the events it returned, and events it didn't merge were the input objects themselves. aw-client's and aw-webui's always_active_pattern query does

not_treat_as_afk = filter_keyvals_regex(events, "app", "Slack|Terminal");  // same objects as `events`
not_afk = period_union(not_afk, not_treat_as_afk);
events = filter_period_intersect(events, not_afk);

so Slack/Terminal window events lost app and title and were categorized Uncategorized. It now returns new events.

categorize, tag, split_url_events set keys in event.data in place. In aw-webui's multideviceQuery (not_afk = events; events = categorize(events, ...)), the Android events in not_afk got $category too. The query wrappers now pass copies: a new Event and data dict per event, which is all these transforms modify. The aw_transform functions themselves are unchanged.

I checked the other query transforms for input mutation (flood, merge, chunk, sort, limit, filters, union_no_overlap, concat, simplify): they don't modify their inputs.

Tests: test_period_union_does_not_modify_inputs, and a query-level test covering all three transforms plus the always_active pattern on every datastore. Both fail before the fix. 308 passed locally.

Parity: on top of the #1466 PR (#173), this fixes 64 more canonical-query cases: aw-client fullDesktopQuery+always_active/+browser and webui multidevice (from 451 known failures to 166 with both PRs). The remaining multidevice-webui-android-merge cases come from aw-server-rust's merge_events_by_keys returning HashMap order (ActivityWatch/aw-server-rust#757).

[ErikBjare]

@greptileai review

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-09-26T21:44:17.366938Z	05719b8	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[greptile-apps]

Confidence Score: 5/5

[Medium risk] Fixes event mutation bug in query transforms.

The PR appears safe to merge; no actionable regression was identified.

Summary

The PR prevents query transforms from changing Event objects still held by other variables.

• Copies events before categorization, tagging, and URL splitting.
• Makes period_union return new, data-free events rather than clearing input data.
• Adds transform-level and query-level regression tests.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart LR
    A[Shared input events] --> B[Copy event and data dictionary]
    B --> C[Categorize, tag, or split URL]
    A --> D[Other query variable remains unchanged]
    A --> E[period_union]
    E --> F[New events with empty data]

Loading

_{Reviews (1) · Last reviewed commit: "fix(query): don't let categorize, tag an..."}