fix(genesis-writer): emit six state fields the indexer reads

[rickyrombo]

The bug class

The indexer's create handlers read a metadata key the writer never emits, so the column silently lands on its default. Row-count parity is blind to it — the counts match while the columns are empty. Tracks were fixed separately in #486; these are the six remaining instances.

metadata key	indexer reads at	target column	source rows
is_save_of_repost	social_save.go:89	saves.is_save_of_repost	28,155
is_repost_of_repost	social_repost.go:83	reposts.is_repost_of_repost	21,366
is_image_autogenerated	playlist_create.go:94	playlists.is_image_autogenerated	4,865
is_scheduled_release	playlist_create.go:96	playlists.is_scheduled_release	566
is_members_only	comment_create.go:44	comments.is_members_only	39
video_url	comment_create.go:45	comments.video_url	15

Every count re-measured on audius_discovery_2026_08_07; the two playlist reads had shifted from the line numbers originally reported.

omitempty, decided per field

The convention is that a boolean whose read-side default is true must be serialized without omitempty (a dropped false flips it on read) — that is why is_delete and is_available carry an "always serialized" comment. Each of the six was checked individually:

field	read-side default	conclusion
is_save_of_repost	MetadataBoolOr(..., false)	omitempty — safe, and keeps the key off ~28M saves
is_repost_of_repost	MetadataBoolOr(..., false)	omitempty — same
is_image_autogenerated	MetadataBoolOr(..., false)	omitempty — matches its neighbours is_album / is_private / is_stream_gated
is_scheduled_release	MetadataBoolOr(..., false)	omitempty — same
is_members_only	MetadataBoolOr(..., false)	omitempty — 39 rows carry it, none of the rest
video_url	MetadataString → nullString → NULL	omitempty — absent, "" and NULL are the same state here, so nothing is lost

None of the six defaults to true, so no field needed the always-serialized treatment. video_url is nullable text in the source and is carried as *string, so an absent URL stays NULL rather than becoming ''.

is_members_only is emitted verbatim from the column: all 39 source rows that set it are entity_type='FanClub', which is exactly what validateCommentWrite requires and what insertCommentWithState honors. (The 15 video_url rows are 11 FanClub + 4 Event.)

Scan alignment

Each of the four queries touched had its SELECT list diffed pairwise against its rows.Scan argument list, mechanically rather than by eye. Counts and every position match:

• saves — 7 cols / 7 args, s.is_save_of_repost → &s.isSaveOfRepost
• reposts — 7 / 7, r.is_repost_of_repost → &r.isRepostOfRepost
• playlists — 28 / 28, p.is_image_autogenerated → &p.IsImageAutogenerated (pos 8), p.is_scheduled_release → &p.IsScheduledRelease (pos 17)
• comments — 11 / 11, c.is_members_only → &c.IsMembersOnly (10), c.video_url → &c.VideoURL (11)

The two playlist columns were inserted next to their semantic neighbours rather than appended, and the Scan was moved in step, so the mirrored SELECT/Scan grouping the file already had is preserved.

Testing

TestWriterEmitsStateFieldsTheIndexerReads in cmd/genesis-writer/entities_state_fields_test.go follows the entities_comment_pin_test.go pattern: build a DP-shaped source schema, run the real writer step, then replay the emitted transactions through the real migration dispatcher (production handlers + RegisterMigrationOverrides, as indexer.go builds it) and assert the indexed column. Asserting the emitted JSON alone would not catch a key the indexer spells differently. Playlists and comments each include a negative-control row so a blanket true fails too.

No restructuring was needed: all four metadata builders are inline closures, but each step is callable on its own (writeSaves, writeReposts, writePlaylists, writeComments), so the whole path is exercised end to end without extracting a builder the way #486 had to.

Verified red by reverting the three entities_*.go files and keeping the test:

--- FAIL: TestWriterEmitsStateFieldsTheIndexerReads
    saves.is_save_of_repost = false, want true (the source row sets it;
      metadata was {"created_at":"2025-06-01T12:00:00Z","is_delete":false})
    reposts.is_repost_of_repost = false, want true (the source row sets it;
      metadata was {"created_at":"2025-06-01T12:00:00Z","is_delete":false})
    playlist 9301 is_image_autogenerated = false, want true
    playlist 9301 is_scheduled_release = false, want true
    comment 9401 is_members_only = false, want true
    comment 9401 video_url = <nil>, want "https://example.com/clip.mp4"

All six fail on their own assertion. With the emission restored the package is green (TestWriteCommentPins_SetsPinnedCommentID, TestWriteMutedUsers_ReachesMutedUsersTable, and the new test). go build ./cmd/genesis-writer/, gofmt -l and go vet are clean.

Coordination

Rebased onto origin/main after #489 landed; the entities_social.go overlap auto-merged. No new time.RFC3339 formatting is introduced, so nothing here interacts with the .UTC() guard in #488.

🤖 Generated with Claude Code