Pause only stopped the stdout reader loop; yt-dlp kept transferring at
full rate until the pipe buffer filled, consuming bandwidth while the
UI claimed the download was paused. Send SIGSTOP/SIGCONT to the whole
process group (yt-dlp and its ffmpeg children) on Unix. Windows has no
equivalent signal; the limitation is documented in the helper.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
cleanup_partial_files() deleted every *.part and *.fNNN.* file in the
whole download directory, and cleanup_subtitle_files() deleted any new
.vtt/.srt under it recursively - including files belonging to other
applications (e.g. a browser's own .part downloads in ~/Downloads).
Track every destination path yt-dlp reports for this download and
restrict both cleanup passes to those files and their .part/.ytdl
siblings.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
For playlist downloads, treat an item as completed if a file was created even when the downloader returned a non-zero code. Emit 100% progress for such cases and update the status to "download.completed (with some errors)" when return_code != 0, otherwise keep the normal completed status. This avoids reporting false failures for playlist items that produced output despite errors.
When downloading playlists, add --ignore-errors and --no-abort-on-error so a single failure won't stop the entire playlist download. Also normalize values stored in the version cache: coerce version_info to a string (or empty string) and path to a string or None to ensure consistent types when reading/writing cached tool metadata.
Downloader: add detection for "Auto-generated" subtitle selection and only append --write-auto-subs when an auto-generated subtitle is explicitly chosen (prevents always enabling the flag when any subtitles are requested). GUI: update subtitle label to display the number of selected subtitles when >0, falling back to the zero-selected message otherwise. Small UX and behavior fixes in DownloadThread and AnalysisMixin.
When building the output template for single-file downloads, remove any playlist-specific preamble (e.g. '%(playlist_index)s - ') from filename_part. Adds a regex-based substitution to clean up leftover playlist_index placeholders so single downloads don't get playlist-style prefixes in their filenames.
When downloading playlists, avoid reusing a specific format_id from the first item (which can fail for subsequent videos). For audio-only playlists, use yt-dlp's bestaudio fallback; for video playlists, respect a provided resolution by extracting the height and applying -S res:<height>, otherwise fall back to bestvideo+bestaudio/best. Invalid resolution strings are caught and logged, and revert to the dynamic best-quality fallback.
Add video id to the default output filename template across the downloader and settings UI. The template was changed from "%(title)s_%(resolution)s.%(ext)s" to "%(title)s_%(resolution)s_[%(id)s].%(ext)s" in ytsage_downloader.py and ytsage_dialogs_settings.py (including placeholder and reset behavior). This helps avoid filename collisions and makes downloaded files easier to identify.
Add a concurrent_fragments parameter to DownloadThread (default 1) and store it as an instance attribute. When building the yt-dlp command, include a -N <concurrent_fragments> option and log the concurrent connections. Also emit finished_signal upon cancellation and on error paths to ensure the UI is notified of thread completion. Minor comment/logging clarifications added around command construction.
Introduce an "audio_normalization" option across the app and implement normalization for audio-only downloads. Updates include: add config default (ConfigManager), new language strings, settings UI checkbox + help text and logic to auto-enable force-audio-format when normalization is enabled, and persist the setting. Pass the setting from main app to DownloadThread, and in the downloader force re-encoding to mp3 when necessary and append --postprocessor-args ExtractAudio:-af loudnorm=... to apply EBU R128 normalization. This ensures normalization works reliably (avoids ffmpeg stream-copy errors) and keeps the behavior scoped to ExtractAudio postprocessor.
Introduce a new filename_format setting and UI to control yt-dlp output templates. ConfigManager now includes a default filename_format (%(title)s_%(resolution)s.%(ext)s). The DownloadSettingsDialog exposes a text input and help text for the format and saves the value to ConfigManager. DownloadThread now accepts a filename_format argument and uses it when building output templates for single videos and playlists. YTSageApp reads the config and passes the filename format into the download thread. Added corresponding English language strings.
Initialize an error buffer and collect yt-dlp "ERROR:" output lines during direct command runs. When the process exits non‑zero, emit a more informative error using the last two captured error lines (falling back to the generic return-code message if none were captured). Also ensure the buffer is present before appending and keep the existing delay/cleanup flow.
Renamed 'src/core' to 'ytsage/core' and updated all internal imports to use relative paths. This improves package structure and ensures correct module resolution after the directory move.
Jaroslav Beneš