[doqhu]

Problem

URL imports had no local file until FileTranscriber.run() downloaded the audio. Speech extraction and the before_transcription / check_skip plugins run before that, so they got file_path=None. URL imports with Extract speech enabled could not work, and plugins never saw the audio.

Fix

The download is now its own phase. For URL tasks the queue worker runs FileTranscriber.download() on a worker thread first. When the audio is on disk, it continues with speech extraction, plugins and transcription, as for local files.

Moving the download earlier exposed a few issues, fixed in separate commits:

• Skip Already Transcribed skipped unrelated URL imports. Plugins now see the download under its fixed temp name audio.wav. The result file check then matched any audio*.txt/srt/vtt and marked new imports as skipped. URL imports are never skipped now, which was the behaviour before.
• Races between cancel and the download.
  • A late completed signal from a canceled download could start transcription for whichever task was current by then. Download signals now carry their own task and activity token and are ignored once that task is no longer current.
  • A cancel between the download and the transcriber thread had nothing to stop, so the task was transcribed anyway. The worker now checks for a cancel before and after speech extraction and after plugins.
• Canceling a download froze the UI. FileTranscriber.stop() did nothing, so cancel_task blocked the GUI thread for up to 5 s and then terminate()d the thread.
  • stop() now aborts yt-dlp from its progress hook (DownloadCancelled) and kills the ffmpeg conversion, which now runs with Popen.
  • yt-dlp can't be interrupted while it resolves the URL, so cancel_task releases the queue right away instead of waiting for a download thread.
• Cleanups:
  • task_started is sent once per URL task. A second send reset time_started.
  • Downloading happens only in download().
  • original_file_path is reset after a download, so it can't point at a previous run's file.
  • The download phase has its own _setup_download_thread.

The commits can be reviewed one by one. Every commit passes the affected tests.

Testing

• Full suite on Linux (Fedora 44, whisper.cpp built with Vulkan): 1382 passed, 15 skipped, 3 failed (measured with Fix CUDA prompt firing on a deleted main window #1668 applied as well, which adds 2 of those tests). The 3 failures are in tests/widgets/speaker_count_test.py (MSDDDiarizer.diarize() got an unexpected keyword argument 'num_speakers'). They fail the same way on main, and this PR doesn't touch speaker code. On machines with an NVIDIA GPU, test_should_run_file_transcription_task can also fail intermittently; that is fixed separately in Fix CUDA prompt firing on a deleted main window #1668.
• test_transcribe_from_url calls run() directly, so it now calls download() first, as the queue worker does.
• New tests:
  • the download runs before speech extraction, plugins and transcription, and each step gets the downloaded (or extracted) file;
  • late or stale download signals are ignored;
  • a cancel during extraction or plugins stops the task;
  • stop() aborts yt-dlp and kills ffmpeg;
  • the skip plugin ignores URL imports.
• ruff check reports nothing new. Only the lines this PR adds are ruff-formatted; the rest of each file is unchanged.

🤖 Generated with Claude Code

https://claude.ai/code/session_012ozoVtp2JvfpDkZ1SpBgvW

[codecov]

Codecov Report

❌ Patch coverage is 97.52066% with 3 lines in your changes missing coverage. Please review.
✅ Project coverage is 84.14%. Comparing base (6186d09) to head (8c84bb3).

Files with missing lines	Patch %	Lines
buzz/file_transcriber_queue_worker.py	96.05%	3 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1667      +/-   ##
==========================================
+ Coverage   83.94%   84.14%   +0.20%     
==========================================
  Files         115      115              
  Lines       13835    13928      +93     
==========================================
+ Hits        11614    11720     +106     
+ Misses       2221     2208      -13     

Flag	Coverage Δ
Linux	83.42% <97.52%> (+0.24%)	⬆️
macOS	83.63% <97.52%> (+0.18%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.