mp3rgui now also speaks German and Indonesian. On the command line, the analysis and the gains that -r, -a, and -g apply are unchanged. What changed: an ID3v2 tag keeps its padding and is rewritten in place, writes hold fewer copies of the file in memory, a few output lines that reported the wrong thing are fixed, and so are two mp3rgui cases where a row or an Apply did not match what was on disk or on screen. The library gains two functions, which is why this is 3.15.0 rather than 3.14.1.
mp3rgui in German and Indonesian (#440)
| Language | File | Translated by |
|---|---|---|
| Deutsch | de.po |
@M-Igashi with Claude Code, reviewed by @Rosenheim23 (#462) |
| Bahasa Indonesia | id.po |
@saimskywalker, a first contribution (#480) |
@Rosenheim23 reviewed the German line by line as a native speaker: "Track-Gain" and "Trackanalyse" follow the established ReplayGain terms, "Titel" is kept where tracks are counted, and the instructions use the infinitive rather than "Sie" or "du". The Italian tooltips and table headers now use @Privato72's own wording from [#406] (#461), and the Japanese catalogue was read again against the UI (#460).
Want mp3rgui in your language? It is one text file and needs no programming: see [#440].
Behaviour changes
| What | 3.14.0 | 3.15.0 |
|---|---|---|
An ID3v2 write (-s i, the ReplayGain values under the default layout, --tags-only, -s d) |
Dropped the tag's padding and moved the whole audio stream whenever the tag changed size | Keeps the padding and rewrites the tag in place when it fits; a new or grown tag gets 2 KB of padding |
-s c, and mp3rgui's tag view, on an album range -a -s i stored in ID3v2 |
Not shown under the default layout | Shown |
-g -o tsv on a file the apply did not scan (an M4A, a dry run, a failure) |
Printed 255 and 0 as its global_gain range |
Prints - |
-g -o tsv under -k |
Printed the requested steps | Prints the steps applied |
The per-file line of -r and -a on an M4A |
Showed the track's measured gain (+3.1 dB where -55.7 dB of album gain was applied) |
Shows the gain applied |
-s d on an untagged M4A, including one whose metadata ends in an empty ilst |
Rewrote the file | Leaves it alone, without reading more than moov (#468) |
-u on an M4A that had no udta |
Left an empty udta behind |
Gives back the original structure |
A crafted MP3GAIN_UNDO value |
Could overflow the undo arithmetic (a panic in a debug build) | Saturates |
ID3v2 tags are rewritten in place (#467)
The id3 crate writes a tag in place only when the new tag is exactly as long as the region it replaces, and otherwise moves everything after the tag, which for an MP3 is the whole audio stream. mp3rgain set no padding, so the first write to a file whose tag had padding (most taggers leave some) shrank the tag and moved the file, and a tag that grew moved it again. The tag is now padded to fill its old region when it fits, and given 2 KB of room when it has to grow, so the next write fits.
The audio and every tag value are unchanged; the file keeps its padding instead of losing it. Over 40 tagged MP3s with cover art, -g 2 -s i became 22% faster, -a -s i 13%, and --tags-only 6%. The saving is I/O, so it grows on slower storage.
Fewer copies of the file in memory (#469)
An MP3 apply or undo that writes APEv2 (the default layout and -s a) now holds one copy of the file instead of two, and an M4A apply or undo with ReplayGain tags two instead of three. On a 115 MB MP3, peak memory went from 223 to 113 MB; on a 100 MB M4A, from 304 to 208 MB. That is per file in flight, so -j 8 multiplies it. The output is byte-identical.
mp3rgui
- Apply after a rescan uses the rows you clicked. With Use stored tags on, Apply Track Gain and Apply Album Gain first rescan the files without usable tags. The apply then read the selection again when the rescan ended, so pressing Esc during the rescan applied gain to every loaded file. It now applies to the rows selected at the click.
- One-channel applies and undos (#471). Apply Channel Gain shifted a row's Volume and Gain as if both channels had moved, and an undo of different left and right amounts reversed the left channel only. The next Apply Track Gain took its steps from those numbers and came out short. Such a row now shows no numbers until it is analyzed again.
- Status texts (#466). A capped dry run below zero reads "-1 steps (capped)" rather than "+-1", and Apply Manual Gain and Apply Channel Gain say so while they run instead of "Applying track gain...".
- Memory (#470). The CJK fallback font takes half the memory (8 instead of 15 MB for the Japanese face, 52 instead of 105 MB for the Korean one), and grouping albums by tags reads the files in parallel.
Library
write_album_minmax_with_optionswithAlbumMinmaxOptionskeeps each file's mtime itself, so a frontend no longer wraps the album range write to restore it (#463).write_album_minmax_with_layoutforwards to it.read_undo_channel_stepsreturns the left and right undo steps separately (#471).- Without the
aacfeature,apply_gain_db_auto,undo_gain_auto,delete_gain_tags_auto,read_gain_tags_auto, andread_undo_stepsrefuse an M4A withFeatureNotAvailableinstead of running the MP3 path over it (#149). Default builds are unaffected.
Internal
- A whole-codebase review after 3.14.0: shared helpers replace duplicated code, dead code is gone, and stale comments are fixed. A ReplayGain re-run that needs no change syncs each file once, and the album range reuses the apply's scan.
- CI: the GitHub Actions bumped (#479).
Full changelog: https://github.com/M-Igashi/mp3rgain/compare/v3.14.0...v3.15.0