feat(docs): add multi-artist spec

Part-of: <https://dev.funkwhale.audio/funkwhale/funkwhale/-/merge_requests/2593>
environments/review-docs-multi-4wjalu/deployments/18619
Ciarán Ainsworth 2023-10-09 13:05:14 +02:00 zatwierdzone przez Marge
rodzic a54522eac2
commit e120fc6815
3 zmienionych plików z 464 dodań i 0 usunięć

Wyświetl plik

@ -0,0 +1 @@
Added multi-artist support spec.

Wyświetl plik

@ -105,6 +105,7 @@ specs/collections/index
specs/nodeinfo21/index
specs/offline-mode/index
specs/quality-filter/index
specs/multi-artist/index
```

Wyświetl plik

@ -0,0 +1,462 @@
# Multi-artist support
## Terminology
The following terminology is used throughout this spec:
Join phrase
: The token used to split a list of artists and represent their role in content
Release
: An **album** or **single**. In Funkwhale a release is represented by the `Album` model
Recording
: A single track as part of an **album** or **single**. In Funkwhale, a recording is represented by the `Track` model
## The issue
Funkwhale requires all releases and recordings to be associated to an artist. In Funkwhale's current structure, only one artist can be associated with a release or recording. However, many releases and recordings are credited to multiple artists.
Funkwhale supports adding releases with multiple contributors by assigning a generic "Various Artists" artist to collaborative releases. This approach doesn't give proper credit to the contributors, and leads to lots of content being improperly catalogued.
When a user uploads content created by multiple collaborative artists, they expect the following behavior:
1. Releases to which the artist has contributed should be present on the artist's page
2. Releases to which the artist has contributed should be present in search results for the artist
3. Releases and recordings should contain links to all contributing artists so that users can discover their other work
Currently, Funkwhale has no facility to parse multi-artist releases. The metadata fields that hold this information (`ALBUMARTIST` and `ARTIST`) are challenging to parse as they aren't formatted in a consistent way. When a release has multiple contributing artists, Funkwhale needs to be able to parse the following information about each artist to ensure the content is tagged with each artist as a user would expect:
1. Any aliases the contributing artist might use
2. The **join phrase** used to separate artists in the tag
## Proposed solution
To support multi-artist content, a new `ArtistCredit` model should be added to house contributing artist information, including links to `Artist` models and details about credited names and join phrases. The `ArtistCredit` model should link to `Album` and `Track` entries using a `ForeignKey` relationship to allow multiple `artist_credit` entries on each `Album` and `Track`. The existing `artist` field in `Album` and `Track` models is deprecated in favor of the new `artist_credit` field.
To account for content tagged in MusicBrainz and content tagged in other systems, two strategies need to be used to populate the `ArtistCredit` model:
1. `parse_credit`: a strategy to parse artist credit from a file's metadata using join phrase parsing.
2. `fetch_credit`: a strategy to fetch artist credit directly from MusicBrainz's entry for releases and recordings
## Feature behavior
The workflow for `ArtistCredit` population goes as follows:
```{mermaid}
flowchart TD
upload([The user uploads content to Funkwhale])
upload --> read(Funkwhale reads the file tags of each release/recording)
read --> mb-tagged{Does the content have a MusicBrainz ID?}
mb-tagged -->|no| parse(Funkwhale parses the artists tag\nand splits the artist at any tokens\n in the join phrase list)
parse --> artist(Funkwhale creates Artist entries for any\nartists that aren't present in the database)
mb-tagged -->|yes| query(Funkwhale queries the MusicBrainz API to\nretrieve artist information for each\nunique mbid)
query --> create(Funkwhale fetches artist information\nand creates Artist entries for any\nartists not present in the database)
create & artist --> model(Funkwhale stores the artist_credit information\nfor each release/recording)
model --> display([Funkwhale displays the credit for each release/recording\nin the web app/API representation])
```
### Backend
The backend is responsible for
1. Tokenizing artist tags when a MusicBrainz ID **isn't** present in the file metadata and creating `Artist` entries for new artists
2. Querying the MusicBrainz API when a MusicBrainz ID **is** present and creating `Artist` entries for new artists
3. Creating `ArtistCredit` entries for each **unique** combination of `artist_id`, `credit`, and `joinphrase`
4. Adding the `ArtistCredit` links to `Track` and `Album` entries
The `artist_credit` entries must be linked sequentially to preserve ordering. For each entry, Funkwhale should query for an existing `ArtistCredit` entry and return it if found, or create a new one and return the new entry. This ensures that artist credit information is presented in the same order as in the tags.
#### Data model
A new `ArtistCredit` model needs to hold the contributing artist information. This model contains the following fields:
- The Funkwhale `artist_id` of the artist associated with the release or recording
- The `credit` representing the credited name of the artist as given for the release or recording
- The `joinphrase` associated with the artist (for example: `" feat. "`, `" & "`)
```{mermaid}
classDiagram
direction LR
class Album {
String title
ArtistCredit artist_credit
Date release_date
String release_group_id
Attachment attachment_cover
String type
Actor attributed_to
TaggedItem tagged_items
Fetch fetches
Content description
List api_includes
String api
String federation_namespace
MusicBrainz musicbrainz_model
String musicbrainz_mapping
}
class Track {
UUID mbid
String title
ArtistCredit artist_credit
Integer disc_number
Integer position
Album album
License license
Actor attributed_to
String copyright
Attachment attachment_cover
Integer downloads_count
String federation_namespace
String musicbrainz_model
String api
List api_includes
MusicBrainz musicbrainz_mapping
String import_hooks
QuerySet objects
Fetch fetches
}
class Artist {
String name
String federation_namespace
String musicbrainz_model
Object musicbrainz_mapping
Actor attributed_to
TaggedItem tagged_items
Fetch fetches
Content description
Attachment attachment_cover
String content_category
Date modification_date
String api
QuerySet objects
}
class ArtistCredit {
Artist artist_id
String credit
String joinphrase
}
Track "1" --> Album : album
Track "1..*" --> ArtistCredit : artist_credit
Album "1..*" --> ArtistCredit : artist_credit
ArtistCredit "1" --> Artist : Artist
```
#### Workflow
On the backend, the workflow for fetching `ArtistCredit` information is as follows:
```{mermaid}
sequenceDiagram
ArtistCredit ->> fetch_credit: Fetch credit from MusicBrainz
fetch_credit -->> ArtistCredit: Return credit information from MusicBrainz
ArtistCredit ->> parse_credit: Parse credit from tags
parse_credit -->> ArtistCredit: Return credit information
```
If MusicBrainz tags are present in the file metadata, the `fetch_credit` strategy should always be preferred to ensure that content aligns with MusicBrainz's tagging.
#### parse_credit strategy
When content doesn't contain MusicBrainz tags, Funkwhale should provide the following to enable artist parsing:
1. A sensible list of default **join phrases** that are commonly found in artist tags
2. A setting to enable admins to override or add to the list of join phrases
Funkwhale should tokenize the `ALBUMARTIST` and `ARTIST` fields using a list of **join phrases**. For each token in the list, Funkwhale should do the following:
1. Store content split at the **join phrase** token as the `credit`
2. Search for any artist with the same name as the `credit` and store the id as the `artist_id`
- If no existing artist is found, create a new artist and return the ID as `artist_id`
3. Store the join phrase token that it split at as the `joinphrase`
4. Create an `ArtistCredit` object for each **unique** combination of `artist_id`, `credit`, and `joinphrase`
5. If a matching `ArtistCredit` entry is found, Funkwhale should link to this
```{mermaid}
sequenceDiagram
Upload ->> parse_content: Read file metadata
loop For each artist name
parse_content ->> Artist: Search Artist entries
Artist -->> parse_content: Return entry (if exists)
parse_content ->> Artist: Create Artist entry (if none found)
end
loop For each unique artist credit combination
parse_content ->> ArtistCredit: Search ArtistCredit entries
ArtistCredit -->> parse_content: Return entry (if exists)
parse_content ->> ArtistCredit: Create ArtistCredit entry (if none found)
parse_content ->> Track: Create Track with artist_credit entries
parse_content ->> Album: Create Album with artist_credit entries
end
```
Funkwhale should do the following automatically:
- Trim whitespace from the artist `credit`
- Preserve whitespace around **both sides** of the `joinphrase` for readability
- Use an **empty** string (`""`) as the `joinphrase` for the last entry in the list
For example, given a list of join phrases like this:
- `$`, `|`, `&`, `/`, `feat.`
And the following tags:
| Tag | Values |
| ----------- | ------------------------------------------------------- |
| ALBUMARTIST | Tommy J. & Bobby Forth |
| ARTIST | Tommy J. feat. Robin Devil, Jerry Sabbath & Sammy Burns |
Funkwhale would create the following `ArtistCredit` entries for the `Album`:
| id | artist_id | credit | joinphrase |
| --- | --------- | ----------- | ---------- |
| 1 | 1 | Tommy J. | & |
| 2 | 2 | Bobby Forth | |
And the following for the `Track`:
| id | artist_id | credit | joinphrase |
| --- | --------- | ------------- | ---------- |
| 3 | 1 | Tommy J. | " feat. " |
| 4 | 3 | Robin Devil | ", " |
| 5 | 4 | Jerry Sabbath | " & " |
| 6 | 5 | Sammy Burns | "" |
#### fetch_credit strategy
To maintain compatibility with MusicBrainz and ensure that Funkwhale can parse multi-artist releases, the MusicBrainz API should be used to fetch contributor information relating to the release.
The [`/release` and `/recording` endpoints](https://musicbrainz.org/doc/MusicBrainz_API#Lookups) return `artist-credit` information when called with a `inc=artists` parameter. This information can be parsed to fetch relevant information:
- The `joinphrase`
- The MusicBrainz `id` of the `artist`
- The artist's `name`
```text
https://musicbrainz.org/ws/2/release/ef140c88-8bf1-4e50-9555-5c1d1ed5865c?fmt=json&inc=artists
```
```json
{
"asin": null,
"cover-art-archive": {
"darkened": false,
"artwork": true,
"front": true,
"back": false,
"count": 1
},
"release-events": [
{
"area": {
"id": "525d4e18-3d00-31b9-a58b-a146a916de8f",
"type": null,
"name": "[Worldwide]",
"disambiguation": "",
"sort-name": "[Worldwide]",
"iso-3166-1-codes": ["XW"],
"type-id": null
},
"date": "2012-11-15"
}
],
"quality": "normal",
"barcode": "",
"date": "2012-11-15",
"status-id": "4e304316-386d-3409-af2e-78857eec5cfe",
"status": "Official",
"title": "One Christmas at a Time",
"disambiguation": "",
"text-representation": {
"script": "Latn",
"language": "eng"
},
"packaging-id": "119eba76-b343-3e02-a292-f0f00644bb9b",
"packaging": "None",
"country": "XW",
"id": "ef140c88-8bf1-4e50-9555-5c1d1ed5865c",
"artist-credit": [
{
"joinphrase": " & ",
"name": "Jonathan Coulton",
"artist": {
"type-id": "b6e035f4-3ce9-331c-97df-83397230b0df",
"name": "Jonathan Coulton",
"id": "d8df7087-06d5-4545-9024-831bb8558ad1",
"sort-name": "Coulton, Jonathan",
"disambiguation": "",
"type": "Person"
}
},
{
"artist": {
"type": "Person",
"disambiguation": "",
"id": "7b5b87d3-f3ee-4b5d-b111-1f2e87f87124",
"sort-name": "Roderick, John",
"name": "John Roderick",
"type-id": "b6e035f4-3ce9-331c-97df-83397230b0df"
},
"name": "John Roderick",
"joinphrase": ""
}
]
}
```
To catalogue multi-artist content, Funkwhale needs to:
1. Fetch information for **each** contributing artist and create `Artist` objects for any that don't exist
2. Create `ArtistCredit` objects associated with each release and recording for each **unique** combination of `artist_id`, `credit`, and `joinphrase`
3. If a matching `ArtistCredit` entry is found, Funkwhale should link to this
```{mermaid}
sequenceDiagram
Funkwhale ->> MusicBrainz: /release/{mbid}?fmt=json&inc=artists
MusicBrainz -->> Funkwhale: Album and ArtistCredit information
Funkwhale ->> MusicBrainz: /recording/{mbid}?fmt=json&inc=artists
MusicBrainz -->> Funkwhale: Track and ArtistCredit information
loop For each unique artist mbid
Funkwhale ->> MusicBrainz: /artist/{mbid}?fmt=json&inc=aliases
MusicBrainz -->> Funkwhale: Artist information
end
```
Given the above example, Funkwhale would create the following `ArtistCredit` object:
| id | artist_id | credit | joinphrase |
| --- | --------- | ---------------- | ---------- |
| 1 | 1 | Jonathan Coulton | & |
| 2 | 2 | John Roderick | |
#### API behavior
The Funkwhale API needs to return artist credit information in a way that is easily consumed by a client.
Endpoints should include a `credited_artist` filter that allows a client to return results for which artists are credited. This filter should take a list of IDs.
To return any albums where the artist is listed in the `artist_credit` field, you can filter by the `artist_id` field using the `credited_artist` filter:
```text
https://open.audio/api/v2/albums?credited_artist=6451,6452
```
The `credit` field of the `artist_credit` object must also be searchable using a standard query:
```text
https://tanukitunes.com/api/v2/albums?q=jonathan+coulton
```
#### Migration
To ensure all content has `artist_credit` information, an initial migration should copy the current artist information into an `ArtistCredit` object for each Album and Track with the following mapping:
- `artist_id` = Artist `id`
- `credit` = Artist `name`
- `joinphrase` = `""`
Each album/track must have at least one `artist_credit` entry listed against it.
### Frontend
Data needs to be clearly presented in the web app to ensure that users can easily see which artists collaborated on a work and discover more of their work.
#### Representation in releases and tracks
The frontend should use the `artist_credit` field to populate artist links on releases and tracks. Credited names and join phrases must be preserved to line up with the artists' intent.
The frontend should format the results as follows for **each** artist:
```html
<span>
<a href="{funkwhale_url}/library/artists/{artist_id}">{credit}</a>{joinphrase}
</span>
```
For example, given the following `artist_credit` response:
```json
[
{
"artist_id": 25,
"credit": "Jonathan Coulton",
"joinphrase": " & "
},
{
"artist_id": 395,
"credit": "John Roderick",
"joinphrase": ""
}
]
```
The frontend would render the following:
```html
<span>
<a href="https://open.audio/library/artists/25">Jonathan Coulton</a> &
<a href="https://open.audio/library/artists/395">John Roderick</a>
</span>
```
#### "Also appears in" section
To keep artist discographies properly catalogued, only albums attributed to the artist, not albums with tracks featuring the artist, should appear in the "Albums by" section of the artist page.
Releases to which the artist has contributed should be separated into a new section labeled "Also appears in". This ensures that it is easy to discover content to which an artist has contributed while also prioritizing their own content on the artist page.
#### "Compilations" page
The "Various Artists" artist is a special artist used in MusicBrainz to catalogue compilation content not attributed primarily to a specific artist or artists. This artist is widely used in Funkwhale as the Album Artist tag for compilation content. To preserve this content and make compilation content easier to find, we should create a special "Compilations" page to house compilation content.
### Documentation
The following needs to be documented:
- The artist splitting mechanism. Meaningful examples should be provided to help users tag their content
- The environment setting to override the join phrase list
- Any tasks or script functions which are created to automate the update of historical content
## Availability
- [ ] Admin panel
- [x] App frontend
- [ ] CLI
## Responsible parties
- The backend developers in the **Development** group are responsible for:
- Creating the new `ArtistCredit` model
- Updating the `Album` and `Track` models to link to `ArtistCredit` entries
- Creating the strategies for populating `ArtistCredit` during the upload process
- Updating the API/creating new endpoints
- Creating the database migration to add the new field and that artist information is copied over
- The frontend developers in the **Development** group are responsible for:
- Updating the artist representations in the Album and Track pages to use the new `artist_credit` field in the API response
- Creating the new "Also appears in" section for the artist discography pages
- The **Design** group is responsible for:
- Providing designs for how multi-artist content should be rendered responsively
- Providing designs for the new "Compilations" page
- Providing designs for the updated Artist page
- The **Documentation** group is responsible for:
- Documenting the new feature behavior
- Documenting the new environment settings
- Providing docstrings for API development
## Open questions
- What should be used as the default join words list?
- To maintain compatibility with API v1, should we continue to populate the `artist` tag on `Track` and `Album` entries with the first credited artist?
## Minimum viable product
The MVP for this product is an implementation of the backend behavior. It should:
1. Set up the new models
2. Create the database migration
3. Implement the token splitting and data fetching for new files
### Next steps
1. Create a `manage.py` script that enables admins to update entries in the database
2. Create a task that updates files imported in-place