[zfleeman]

Closes #113. Closes #121.

Changes

1. --astreams numbers must match an audio track (Reject --astreams numbers that don't match an audio track #113). TwoPass.run() now calls a new _check_audio_tracks() right after _check_video_size(), before ffmpeg runs. Any --astreams number past the last audio track raises a ValueError, e.g. "Audio track 7 doesn't exist. This file has 2 audio tracks, numbered 0 to 1." A file with no audio says "This file has no audio tracks." Before this, bad numbers were silently dropped, so --astreams 7 gave a video with no audio but still spent part of the bitrate budget on audio.

   • _apply_audio_filters() no longer filters the list. It uses settings.astreams as given, because run() has already checked it.
   • With --no-audio, the check is skipped. No audio goes into the output anyway, so a leftover astreams value in a config file shouldn't stop the encode.
   • Negative numbers never get this far. _parse_astreams() in arguments.py already drops them with a warning.

2. Setting errors on the CLI without a traceback (Show TwoPass setting errors on the CLI without a traceback #121). A new context manager in __main__.py, exit_on_setting_error(), catches a ValueError, logs only its message, and exits with status 1. main() uses it in two places:

   • Around the --dry-run call to twopass.run() and the normal twopass_loop() call. This covers every ValueError that TwoPass raises: the new audio track check, the crop and even-size checks, the target-size check, bad --from/--to times, and the "no bitrate left for video" error on a retry.
   • Around the twopass.clip_times() check that runs before the web server starts, so bad --from/--to times with --web don't show a traceback either.

   Errors from the web page's own form were already shown on the page, and they're unchanged.

Things to look at closely

• raise SystemExit(1) from None in exit_on_setting_error() hides the ValueError behind it. Without from None, Python would still print the original traceback above the exit.
• The "Time Paradox?" message from clip_times() now prints without a traceback, but the message itself still has its old color codes and extra indentation. That formatting was already there before this PR, so I didn't change it.
• Existing tests changed to use real tracks. test_apply_audio_filters_without_mixing_keeps_one_track dropped its [5, 1] case, and test_apply_audio_filters_mixes_selected_tracks now mixes [0, 2] instead of [0, 2, 5]. Both were testing the old drop-bad-numbers behavior. In test_main.py, test_encode_audio_stream_selection posts track 0 instead of 0 and 2, because the web fixture's file has one audio track and the web UI's dry run now rejects track 2 with a 400.
• run_main fixture. It now replaces twopass_loop when the fixture is set up, not inside _run(). That lets the new test swap in a twopass_loop that raises before calling run_main(). The other tests get the same mock as before.

Testing

• uv run pytest --cov: 285 passed, 99% coverage. ruff format and ruff check are clean.
• New tests in test_twopass.py: a track number past the end raises (alone, and mixed with a valid one), --astreams 0 on a file with no audio raises, and --no-audio with a bad number still encodes.
• New tests in test_main.py: a ValueError from twopass.run() (with --dry-run) or twopass_loop() (without it) exits with status 1 and logs only the message. A bad --from/--to with --web does the same, without starting Flask.
• With the CLI on example.mp4 (2 audio tracks), trimmed to 10 seconds:
  • --astreams 7 --dry-run prints ERROR:root:Audio track 7 doesn't exist. This file has 2 audio tracks, numbered 0 to 1. and exits 1.
  • -x 0x0x9999x9999 (normal run) prints the crop error on one line and exits 1.
  • --from 00:00:20 --to 00:00:10 --web prints the "Time Paradox?" message and exits 1 without starting the server.
  • --astreams 0,1 --amix -s 5 encodes with [0:a:0][0:a:1]amix=inputs=2 and lands at 4.69 MB.