Harden media verification, recovery, and watch-mode safety

.github/workflows/tests.yml

@@ -26,10 +26,6 @@ jobs:
         run: |
           pip install pytest
 
-      # mkvmerge + mediainfo are AudioCleaner's own runtime dependencies
-      # (bundled with the Windows build). Installing them here lets the
-      # integration suite (tests/test_integration_fixtures.py) exercise
-      # the real tools instead of only mocks.
       - name: Install mkvtoolnix & mediainfo
         run: |
           sudo apt-get update
@@ -48,12 +44,8 @@ jobs:
           pytest tests/ -v
 
   windows-build:
-    name: PyInstaller build test (Windows)
+    name: PyInstaller build & launch smoke test (Windows)
     runs-on: windows-latest
-    # Build-only smoke test: confirms the app still packages into a
-    # working .exe with PySide6 + the bundled MediaInfo/libcurl binaries.
-    # It does not run the packaged exe (no display/media tools available
-    # on the runner) -- that's still a manual release-checklist step.
     steps:
       - uses: actions/checkout@v4
 
@@ -79,3 +71,22 @@ jobs:
             exit 1
           }
           Write-Host "Build OK: $((Get-Item dist/AudioCleaner.exe).Length) bytes"
+
+      - name: Launch executable smoke test
+        shell: pwsh
+        run: |
+          $exe = (Resolve-Path "dist/AudioCleaner.exe").Path
+          $proc = Start-Process -FilePath $exe -PassThru
+          try {
+            Start-Sleep -Seconds 8
+            if ($proc.HasExited) {
+              Write-Error "AudioCleaner.exe exited during the launch smoke test with code $($proc.ExitCode)"
+              exit 1
+            }
+            Write-Host "Launch OK: process remained running for 8 seconds"
+          }
+          finally {
+            if (-not $proc.HasExited) {
+              Stop-Process -Id $proc.Id -Force -ErrorAction SilentlyContinue
+            }
+          }

audiocleaner/config.py

@@ -7,10 +7,11 @@
 # --- Cache / rules versioning ---
 # Bump SCANNER_VERSION whenever probe.py's extraction logic changes in a way
 # that could change what a cached FileProbeResult contains (new fields, new
-# matching logic, etc). Bump RULES_VERSION whenever track-selection logic
-# changes (codec_rank.py) in a way that could change *decisions* made from
-# an unchanged probe result. Either bump invalidates old cache entries, so a
-# stale cache never silently keeps applying pre-change decisions.
+# matching logic, etc). This is the version ProbeCache uses to invalidate
+# stale probe data. RULES_VERSION identifies track-selection rules and can be
+# recorded by callers/history when decision semantics change; changing it does
+# not invalidate probe data because codec-ranking changes do not alter the
+# underlying MediaInfo/mkvmerge facts stored in the cache.
 SCANNER_VERSION = "2"
 RULES_VERSION = "2"
 
@@ -30,22 +31,21 @@
 # Ordered best -> worst. Keys are internal codec identifiers produced by
 # codec_rank.classify_audio_track(). Lower index = higher priority.
 CODEC_PRIORITY = [
-    "truehd_atmos",   # Dolby TrueHD + Atmos
-    "dtsx",            # DTS:X
-    "truehd",          # Dolby TrueHD
-    "dts_hd_ma",       # DTS-HD Master Audio
-    "lpcm",            # LPCM
-    "flac",            # FLAC
-    "eac3",            # Dolby Digital Plus (E-AC3)
-    "dts",             # DTS (core)
-    "ac3",             # Dolby Digital (AC3)
-    "aac",             # AAC
-    "mp3",             # MP3
+    "truehd_atmos",
+    "dtsx",
+    "truehd",
+    "dts_hd_ma",
+    "lpcm",
+    "flac",
+    "eac3",
+    "dts",
+    "ac3",
+    "aac",
+    "mp3",
 ]
 
 CODEC_RANK = {codec: i for i, codec in enumerate(CODEC_PRIORITY)}
 
-# Human-readable labels for logs / UI.
 CODEC_LABELS = {
     "truehd_atmos": "Dolby TrueHD + Atmos",
     "dtsx": "DTS:X",
@@ -61,41 +61,15 @@
     "unknown": "Unknown",
 }
 
-# English language tags as they appear in MKV/mkvmerge/mediainfo output.
 ENGLISH_LANG_CODES = {"eng", "en", "en-us", "en-gb"}
 
-# --- Commentary tracks ---
-# Off by default: commentary tracks are treated like any other extra audio
-# track and stripped. When on, the single best non-commentary track is
-# still chosen as "the" kept track, but commentary track(s) are kept
-# alongside it rather than being removed.
 DEFAULT_KEEP_COMMENTARY = False
-
-# --- Subtitle language filtering ---
-# Off by default, so existing behaviour (subtitles are never touched) is
-# unchanged unless a user explicitly opts in. When on, only subtitle
-# tracks matching a language in DEFAULT_SUBTITLE_LANGUAGES are kept,
-# *plus* any track flagged Forced regardless of language (these are
-# usually foreign-dialogue captions within an otherwise-English film that
-# people expect kept even when everything else in that language is
-# stripped).
 DEFAULT_SUBTITLE_FILTER_ENABLED = False
 DEFAULT_SUBTITLE_LANGUAGES = {"eng"}
 
-# Cache file name written into the target root folder.
 CACHE_FILENAME = ".audiocleaner_cache.json"
-
-# Log file name written into the target root folder.
 LOG_FILENAME = "audiocleaner_log.txt"
-
 MKV_EXTENSIONS = {".mkv"}
 
-# --- Watch mode ---
-# How often the watcher re-scans the folder tree for new/changed files.
 WATCH_POLL_INTERVAL_SECONDS = 15
-
-# How long a file's size must stay unchanged before it's considered fully
-# copied and safe to process. User-adjustable in the GUI; this is just the
-# default (matches the "wait a minute or two" guidance for Radarr/Sonarr
-# moving files into the library folder).
 WATCH_DEFAULT_SETTLE_SECONDS = 120