feat: v3.0.3 installer refactor and shared env#577
Open
doomsday616 wants to merge 2 commits into
Open
Conversation
doomsday616
force-pushed
the
feat/v3.0.3-installer-refactor-pr
branch
from
08de1ed to
3c80fc3
Compare
Bumps version to 3.0.2.
版本号升至 3.0.2。
主要内容 / Major themes:
i18n
- Browser language auto-detection (Accept-Language) on first load,
with a top-right language selector overriding the choice per session.
首次加载根据浏览器 Accept-Language 自动识别语言；右上角语言选择器可按会话覆盖。
- Sidebar duplicate language selector removed.
移除侧边栏重复的语言选择器。
- Routes display_language through query params + session_state, with
config.yaml as a fallback only.
display_language 改为优先走 query params + session_state，config.yaml 仅作兜底。
- Adds normalize_language_code() to map zh / zh-CN / zh-HK / zh-Hant /
variants to the supported set.
新增 normalize_language_code()，把 zh / zh-CN / zh-HK / zh-Hant 等变体统一映射到受支持的语言。
- Translates previously hard-coded UI strings: WhisperX runtime, TTS
engine names, Voice / 302ai API / ElevenLabs API labels, "Star on
GitHub" button, YouTube resolution "Best".
翻译之前硬编码的 UI 文案：WhisperX runtime、TTS 引擎名、Voice / 302ai API / ElevenLabs API 标签、"Star on GitHub" 按钮、YouTube 分辨率 "Best"。
- Fixes 'here' link text leaking English in zh-CN / zh-HK welcome
string.
修复欢迎语在简中/繁中里 "here" 链接文字仍为英文的问题。
- Adds CSS overlay for the file_uploader internals (Streamlit has no
official i18n for these) covering "Drag and drop file here",
"Limit ... per file" and "Browse files" labels.
通过 CSS 覆盖 file_uploader 内部文案（Streamlit 官方未提供 i18n），包括 "Drag and drop file here"、"Limit ... per file"、"Browse files"。
- Hides Streamlit developer toolbar (client.toolbarMode = "viewer")
and disables the file watcher (server.fileWatcherType = "none") so
"File change / Rerun / Always rerun" prompts no longer appear.
隐藏 Streamlit 开发者工具栏（client.toolbarMode = "viewer"），关闭文件监听（server.fileWatcherType = "none"），避免出现 "File change / Rerun / Always rerun" 英文提示。
- Fills missing translation keys across en / zh-CN / zh-HK / es / fr /
ja / ru.
补全 en / zh-CN / zh-HK / es / fr / ja / ru 七种语言中缺失的翻译键。
Audio-only input flow / 纯音频输入流程
- Adds output/input_manifest.json written by the upload / YouTube
download path, recording the original media type. find_media_file()
now reads the manifest first, so generated artefacts (dub.mp3,
normalized_dub.wav) no longer poison detection.
上传 / YouTube 下载后写入 output/input_manifest.json 记录原始媒体类型。find_media_file() 优先读 manifest，避免生成产物（dub.mp3、normalized_dub.wav）污染识别。
- find_audio_files() now skips generated audio names.
find_audio_files() 自动跳过生成产物文件名。
- find_media_file() distinguishes "no media" vs "multiple media"
errors instead of silently falling back.
find_media_file() 区分"无媒体"和"多个媒体"两种错误，不再静默 fallback。
- Sidebar no longer persistently writes burn_subtitles = false when
the input is audio; the toggle is only disabled in the UI.
音频输入时不再持久化写入 burn_subtitles = false，只在 UI 层禁用开关。
- Main pipeline now flows download -> subtitles -> (optional) dubbing,
only showing the dubbing section after subtitles are done AND the
input is not audio.
主流程改为：下载 → 字幕 →（可选）配音；只有字幕完成且输入不是音频时才显示配音段。
- Adds prepare_audio_for_asr() so audio-only inputs are normalized to
16k / mono / mp3 without going through video conversion.
新增 prepare_audio_for_asr()，纯音频输入直接归一化为 16k / 单声道 / mp3，不再经过视频转换。
- Removes the obsolete convert_audio_to_video() placeholder path.
移除已废弃的 convert_audio_to_video() 占位路径。
Task control (pause / resume / stop) / 任务控制（暂停 / 继续 / 停止）
- TaskRunner gains a class-level _current pointer and TaskRunner.check_cancel()
so long-running core loops can cooperatively cancel.
TaskRunner 新增类级 _current 指针和 TaskRunner.check_cancel()，让长循环可以协作式取消。
- Adds core.utils.check_cancel() wrapper, imported via the existing
`from core.utils import *` pattern.
新增 core.utils.check_cancel() 包装，沿用现有 `from core.utils import *` 导入方式。
- Inserts check_cancel() into the hot loops: ASR segment loop,
translate parallel loop, TTS warmup / parallel collection / chunk
merge, audio segment merge, translate_lines entry. Parallel loops
also cancel pending futures on stop.
在热点循环里插入 check_cancel()：ASR 分段循环、翻译并行循环、TTS warmup / 并行收集 / chunk 合并、音频段合并、translate_lines 入口。并行循环在 stop 时主动 cancel 未启动的 futures。
Done markers and completion detection / 完成标记与状态判断
- Adds output/.subtitle_done and output/.dubbing_done markers written
by the runner as the last step of each stage.
TaskRunner 在每个阶段最后一步写入 output/.subtitle_done 和 output/.dubbing_done 标记。
- text_done / audio_done now prefer the marker, falling back to a full
outputs-present check, so half-failed runs are no longer mistaken
for completion.
text_done / audio_done 优先读标记，没有则回落到"所有最终产物齐全"检查，避免半失败被误判为完成。
- Subtitle length tuning controls (max_split_length, subtitle.max_length)
surfaced in an expandable section above "Start Processing Subtitles",
with suggested ranges and a "Restore defaults" button.
在"开始处理字幕"上方加入折叠的字幕长度微调（max_split_length、subtitle.max_length），含建议范围和"恢复默认"按钮。
Robustness fixes / 健壮性修复
- ElevenLabs ASR: elev2whisper() now always emits word-level
timestamps; process_transcription() tolerates segments without
`words` by synthesizing one from the segment text.
ElevenLabs ASR：elev2whisper() 始终输出词级时间戳；process_transcription() 对没有 `words` 的 segment 用 segment 文本合成一个，避免 KeyError。
- download_video_section now surfaces detection errors (e.g. multiple
media files in output/) with a clear message and a "Clear output and
reselect" button, instead of silently falling back to the upload
view.
download_video_section 在媒体识别失败（例如 output/ 里有多个媒体）时显示明确错误和"清空输出并重新选择"按钮，不再静默回到上传界面。
- Re-upload of the same file is detected via session_state, avoiding
an infinite rerun loop.
通过 session_state 识别同一文件的重复上传，避免无限 rerun 循环。
- give_star_button rewritten to a plain string template; previous
f-string broke on the literal `{` in the embedded CSS.
give_star_button 改写为普通字符串模板；原 f-string 因内嵌 CSS 里的 `{` 报错。
Tooling and config / 工具与配置
- OneKeyStart.bat consolidated: auto-detects .venv (uv install) or
falls back to the legacy Conda env "videolingo"; OneKeyStart_uv.bat
removed.
合并 OneKeyStart.bat：自动检测 .venv（uv 安装）或回落到旧的 Conda 环境 "videolingo"；删除 OneKeyStart_uv.bat。
- Logs now go to logs/videolingo_<timestamp>.log instead of the
project root.
日志写入 logs/videolingo_<timestamp>.log，不再散落在项目根目录。
- .streamlit/config.toml: client.toolbarMode = "viewer",
server.fileWatcherType = "none", server.maxUploadSize preserved.
.streamlit/config.toml：client.toolbarMode = "viewer"，server.fileWatcherType = "none"，保留 server.maxUploadSize。
- .gitignore: ignores logs/, videolingo_*.log, AGENTS.md, pr-body.md.
.gitignore：忽略 logs/、videolingo_*.log、AGENTS.md、pr-body.md。
- setup.py + config.yaml header bumped to 3.0.2.
setup.py 与 config.yaml 顶部版本号统一升至 3.0.2。
No new dependencies. No CLI behavior changes.
未引入新依赖。CLI 行为无变更。
English: - Add stage-based installer.py with resumable dependency setup and health checks - Support shared uv venv via setup_env.py and OneKeyStart.bat - Make OneKeyStart.bat run pre-launch checks and repair incomplete environments - Relax spaCy patch pin and make Demucs installation more resilient - Fall back to global HuggingFace cache when project WhisperX cache is incomplete - Bump version metadata to v3.0.3 中文: - 新增分阶段 installer.py,支持可重复执行的依赖安装和环境健康检查 - setup_env.py 与 OneKeyStart.bat 支持共享 uv venv - OneKeyStart.bat 启动前自动检查环境,缺失或过期时自动修复 - 放宽 spaCy patch 版本约束,提升 Demucs 安装容错 - 项目内 WhisperX 缓存不完整时自动回退到全局 HuggingFace 缓存 - 版本元数据升级到 v3.0.3
doomsday616
force-pushed
the
feat/v3.0.3-installer-refactor-pr
branch
from
3c80fc3 to
e0f39bd
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
English:
This PR depends on and builds on top of #576 (
feat: v3.0.2 - i18n, audio-only flow, task control, UX polish). Please review/merge #576 first.Changes:
installer.py, a stage-based, resumable installer with health checks.setup_env.pyto create/reuse local or shared uv virtual environments and delegate installs toinstaller.py.OneKeyStart.batto prefer a shared venv, run pre-launch checks, repair incomplete/outdated environments, and show colored status output.install.pyas a backward-compatible wrapper._model_cacheis incomplete, use the global HuggingFace cache instead of failing on a half-downloaded cache.Verification:
python -m py_compile installer.py setup_env.py install.py core/asr_backend/whisperX_local.pypython installer.py --checkOneKeyStart.bat --check-onlysetup_env.py --path <temp-venv> --skip-install --yes中文:
这个 PR 依赖并基于 #576 (
feat: v3.0.2 - i18n, audio-only flow, task control, UX polish)。请先 review/merge #576。改动:
installer.py,并加入环境健康检查。setup_env.py,支持本地/共享 uv 虚拟环境,安装逻辑交给installer.py。OneKeyStart.bat,优先使用共享 venv,启动前检查环境,缺失或过期时自动修复,并加入彩色状态输出。install.py保留为兼容入口。_model_cache不完整时自动回退到全局 HuggingFace 缓存。