alex/ephyra
Watch
1
0
Fork
You've already forked ephyra
 0
Code Issues 2 Pull requests Projects Releases Packages 1 Wiki Activity Actions

A track's tags are split where upstream does not split them #27

New issue
Closed
opened 2026-08-02 18:22:10 +02:00 by Claude · 1 comment
Claude commented 2026-08-02 18:22:10 +02:00
Collaborator
Copy link

Upstream does not read a track's tags through the ffprobe normalizer. AudioFileProber
reads them with ATL and splits on ATL's own separator — one tag frame per value —
applying further delimiters only when a library sets UseCustomTagDelimiters, which
is off by default (AudioFileProber.cs:414-425, and the same shape for artists at
:378).

Ours runs every track through probe::tags, whose genre delimiters are / ; , and
whose artist handling replaces feat. before splitting. On the real library:

column tracks differing
Genres 1403 of 4889
Artists 13 of 4889 
upstream Genres  'Ambient;Downtempo;Electronic'      one genre
ours     Genres  'Ambient|Downtempo|Electronic'      three

upstream Artists 'Aes Dana feat. MikTek'             one artist
ours     Artists 'Aes Dana|MikTek'                   two

The delimiter tables themselves match upstream's ProbeResultNormalizer exactly —
_basicDelimiters, _nameDelimiters, _genreDelimiters and the split whitelist are
all faithful. The divergence is that a track never reaches that code upstream.

Found while verifying #9: the by-name values a track is listed under are a faithful
reflection of these columns, so 46 of 429 ItemValues rows differ for this reason
alone — Acid Techno;Electronic;Idm;Techno is one genre upstream and four here.

Upstream does not read a track's tags through the ffprobe normalizer. `AudioFileProber` reads them with ATL and splits on ATL's own separator — one tag frame per value — applying further delimiters only when a library sets `UseCustomTagDelimiters`, which is off by default (`AudioFileProber.cs:414-425`, and the same shape for artists at `:378`). Ours runs every track through `probe::tags`, whose genre delimiters are `/ ; ,` and whose artist handling replaces `feat.` before splitting. On the real library: | column | tracks differing | |---|---| | `Genres` | 1403 of 4889 | | `Artists` | 13 of 4889 | upstream Genres 'Ambient;Downtempo;Electronic' one genre ours Genres 'Ambient|Downtempo|Electronic' three upstream Artists 'Aes Dana feat. MikTek' one artist ours Artists 'Aes Dana|MikTek' two The delimiter tables themselves match upstream's `ProbeResultNormalizer` exactly — `_basicDelimiters`, `_nameDelimiters`, `_genreDelimiters` and the split whitelist are all faithful. The divergence is that a track never reaches that code upstream. Found while verifying #9: the by-name values a track is listed under are a faithful reflection of these columns, so 46 of 429 `ItemValues` rows differ for this reason alone — `Acid Techno;Electronic;Idm;Techno` is one genre upstream and four here.
Claude commented 2026-08-03 15:58:14 +02:00
Author
Collaborator
Copy link

Fixed in the merge of fix/27-tag-splitting.

The rule could not be fixed as a rule

The premise held — upstream reads a track with a tag library and splits only on
that library's own separator — but no delimiter table could reproduce it,
because ffprobe cannot say how many values a tag holds. Two files from this
library, one whose genres upstream splits and one whose it does not:

A Silver Mt. Zion … .flac    GENRE = 'Experimental;Post-Rock;Rock'
AES Dana … .mp3              genre = 'Ambient;Downtempo;Electronic'

Identical to the prober. Upstream tells them apart because ATL reads the file:
three Vorbis fields against one ID3 frame. Nor does the container predict it —
of the FLACs upstream split 1317 and left 1302 whole.

So the track is read twice

As upstream reads one. The prober answers what is in the container; the file's
own tags answer how many values a tag holds, and land in MediaInfo::written,
a value per frame. A track's artists, album artists, genres and composers are
read from there, and from what the prober made of the tag where no reader could
open the file. A video is untouched: upstream reads one through the normalizer
and never through a tag library, which is what the probe corpus records.

The columns and the credits are read from the same place, so a track cannot be
listed under two artists and billed for one.

Measured

/mnt/media/music, 4889 tracks, rows differing from upstream:

column before after
Genres 1403 59
AlbumArtists 50 50
Artists 13 0

and the albums summarised from them (#24) go from 141 differing Genres to 10.
ItemValues is 508 against upstream's 515, from 538.

What is left is where the two tag libraries collapse a repeated field
differently — a file naming ALBUMARTIST twice is two values to one and one to
the other — which is not reachable without ATL itself.

Left for later

  • #39 — the port bills 9328 people against upstream's 9151. That is who gets
    credited rather than how a tag is split.
Fixed in the merge of `fix/27-tag-splitting`. ## The rule could not be fixed as a rule The premise held — upstream reads a track with a tag library and splits only on that library's own separator — but no delimiter table could reproduce it, because ffprobe cannot say how many values a tag holds. Two files from this library, one whose genres upstream splits and one whose it does not: ``` A Silver Mt. Zion … .flac GENRE = 'Experimental;Post-Rock;Rock' AES Dana … .mp3 genre = 'Ambient;Downtempo;Electronic' ``` Identical to the prober. Upstream tells them apart because ATL reads the file: three Vorbis fields against one ID3 frame. Nor does the container predict it — of the FLACs upstream split 1317 and left 1302 whole. ## So the track is read twice As upstream reads one. The prober answers what is in the container; the file's own tags answer how many values a tag holds, and land in `MediaInfo::written`, a value per frame. A track's artists, album artists, genres and composers are read from there, and from what the prober made of the tag where no reader could open the file. A video is untouched: upstream reads one through the normalizer and never through a tag library, which is what the probe corpus records. The columns and the credits are read from the same place, so a track cannot be listed under two artists and billed for one. ## Measured `/mnt/media/music`, 4889 tracks, rows differing from upstream: | column | before | after | |---|---|---| | `Genres` | 1403 | 59 | | `AlbumArtists` | 50 | 50 | | `Artists` | 13 | 0 | and the albums summarised from them (#24) go from 141 differing `Genres` to 10. `ItemValues` is 508 against upstream's 515, from 538. What is left is where the two tag libraries collapse a repeated field differently — a file naming `ALBUMARTIST` twice is two values to one and one to the other — which is not reachable without ATL itself. ## Left for later - #39 — the port bills 9328 people against upstream's 9151. That is who gets credited rather than how a tag is split.
Sign in to join this conversation.
No Branch/Tag specified
Branches Tags
main
No results found.
Labels
Clear labels   
correctness
Diverges from upstream

  
encoding
Transcoding, ffmpeg, and the arg strings

  
http
Route behaviour

  
jobs
Scheduled tasks

  
operations
Installing a server and keeping one running

  
p1
Do this first

  
p2
After the p1s

  
p3
Real, and not urgent

  
performance
Speed or resource use

  
providers
Metadata from somewhere other than the file

  
scanner
The library scan

  
security
Who is allowed to see what

  
testing
Test coverage and harness

No labels
correctness
encoding
http
jobs
operations
p1
p2
p3
performance
providers
scanner
security
testing
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
alex/ephyra#27
No description provided.