Regression v2.6.0+: MusicBrainz plugin not invoked for Various Artists albums

[djdembeck]

Bug Description

In beets v2.6.0 and v2.6.1, when importing albums detected as Various Artists, the MusicBrainz plugin is never invoked, resulting in immediate "No matching release found" with zero candidates evaluated.

Environment

• beets version: 2.6.1 (also affects 2.6.0)
• Working version: 2.5.1
• Python version: 3.12
• Installation method: Docker (linuxserver/beets:nightly)

Steps to Reproduce

1. Import an album that triggers Various Artists detection:

   beet import "/downloads/new/Summer Walker/Over It (Complete Set Video Version)/"

2. Observe the output:

   • Album might be VA: True
   • chroma: acoustid album candidates: 0
   • Evaluating 0 candidates. (immediate, no MusicBrainz search)
   • No matching release found for 19 tracks.

Expected vs Actual Behavior

v2.5.1 (Working):

Album might be VA: True
chroma: acoustid album candidates: 0
musicbrainz: Searching for MusicBrainz releases with: {'release': 'over it...', 'alias': 'over it...', 'tracks': '19', 'arid': '89ad4ac3-...'}
musicbrainz: Requesting MusicBrainz release e6076fa2-...
Evaluating 2 candidates.

v2.6.1 (Broken):

Album might be VA: True
chroma: acoustid album candidates: 0
Evaluating 0 candidates.
No matching release found for 19 tracks.

Critical difference: No musicbrainz: log lines appear - the plugin is never invoked.

Regression Range

• Working: v2.5.1
• Broken: v2.6.0 and v2.6.1

Likely Cause

The refactoring of the MusicBrainz plugin in v2.6.0 (replacing python-musicbrainzngs with custom MusicBrainzAPI class) appears to have broken the plugin invocation path when va_likely=True.

Workaround

Stay on v2.5.1 or manually search using MusicBrainz ID (option I during import).

[arsaboo]

Do you have musicbrainz plugin enabled in the config?

[djdembeck]

Yes, here are the relevent sections:

plugins: chroma duplicates embedart fetchart lastgenre lyrics musicbrainz mbsync missing parentwork unimported

match:
  max_rec:
    year: medium
    label: strong
    country: low
  preferred:
    countries: ["US", "XW|XE", "GB|UK"]
    media: ["Digital Media", '12" Vinyl', "Vinyl"]
    original_year: yes
  ignored: unmatched_tracks

musicbrainz:
  host: musicbrainz.org
  https: yes
  ratelimit: 1
  ratelimit_interval: 1.0
  extra_tags: []
  genres: no
  genres_tag: genre
  external_ids:
    discogs: no
    bandcamp: no
    spotify: yes
    deezer: no
    beatport: no
    tidal: yes
  data_source_mismatch_penalty: 0.5
  search_limit: 7

[snejus]

Can confirm I can reproduce it. I'm on it

[djdembeck]

Glad you can repro! It also seems the logic is a little to eager to classify things as VA. Almost all my artists are classified as VA (which is how I found this issue)

[snejus]

Found the issue, will soon submit a PR to address it.

Regarding the VA classification logic, see below:

VA_ARTISTS = ("", "various artists", "various", "va", "unknown")
...
        va_likely = (
            (not consensus["artist"])
            or (search_artist.lower() in VA_ARTISTS)
            or any(item.comp for item in items)
        )

Note: consensus["artist"] is a boolean saying every track has the same artist.