feat(gallery): sort/group/jump by COALESCE(post.post_date, image_record.created_at) — surface migrated content at its original publish date, not FC scan date

Operator hit this 2026-05-25 after the IR tag_apply landed: ~57k images all scanned into FC in the same week share image_record.created_at, so the gallery timeline collapses them into a single month bucket and scroll orders them all together at the top. Their actual publish dates (spread over years) were already available in Post.post_date but the gallery never read it. Backend wire-up: - tag_apply phase 4 now sets ImageRecord.primary_post_id when creating ImageProvenance (only if currently NULL — preserves the canonical download-time linkage set by the importer for new FC ingests). - gallery_service.py introduces _effective_date_col() = COALESCE(post.post_date, image_record.created_at), used in: * scroll() ORDER BY + cursor WHERE clauses * timeline() year/month group-by * jump_cursor() year/month filter * _neighbors() prev/next ordering - Each method LEFT OUTER JOIN Post on primary_post_id so the COALESCE works for images without a post (NULL on the Post side, fall back to created_at). - GalleryImage gains posted_at + effective_date fields; API /gallery /scroll exposes both alongside the existing created_at so the UI can render 'Posted on X (imported Y)' if desired. - get_image_with_tags() returns posted_at for the modal. Cursor format unchanged — the encoded datetime is now the effective_ date (whichever column won the COALESCE) and pagination remains consistent. To pick up new behavior for an already-migrated IR set: re-run /api/migrate/tag_apply on the existing manifest (phase 4 is idempotent; the new primary_post_id assignment backfills). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-25 12:30:18 -04:00
parent 9f54efdedf
commit c361032554
5 changed files with 328 additions and 55 deletions
@@ -133,3 +133,135 @@ async def test_get_image_with_tags_includes_integrity_status(db):
    svc = GalleryService(db)
    payload = await svc.get_image_with_tags(img.id)
    assert payload["integrity_status"] == "ok"
+
+
+async def _seed_image_with_post(
+    db, *, sha: str, image_created_at, post_date, artist_name="test-artist",
+    platform="patreon", external_post_id="42",
+):
+    """Helper: seed an Artist + Source + Post and one ImageRecord whose
+    primary_post_id points at that Post. Used for date-coalesce tests."""
+    from backend.app.models import Artist, Post, Source
+    artist = Artist(name=artist_name, slug=artist_name.lower().replace(" ", "-"))
+    db.add(artist)
+    await db.flush()
+    source = Source(
+        artist_id=artist.id, platform=platform,
+        url=f"https://www.{platform}.com/{artist.slug}",
+    )
+    db.add(source)
+    await db.flush()
+    post = Post(
+        source_id=source.id, external_post_id=external_post_id,
+        post_title="A Post", post_date=post_date,
+    )
+    db.add(post)
+    await db.flush()
+    img = ImageRecord(
+        path=f"/images/test/{sha[:8]}.jpg",
+        sha256=sha, size_bytes=1000, mime="image/jpeg",
+        width=100, height=100,
+        origin="imported_filesystem", integrity_status="unknown",
+        primary_post_id=post.id,
+    )
+    img.created_at = image_created_at
+    db.add(img)
+    await db.flush()
+    return img, post
+
+
+@pytest.mark.asyncio
+async def test_scroll_sorts_by_post_date_when_available(db):
+    """Operator-flagged 2026-05-25: ~57k IR images all imported in the
+    same week sort by image.created_at and pile up in one month bucket.
+    Once primary_post_id is wired (via tag_apply phase 4), the gallery
+    should sort by Post.post_date instead, spreading them across the
+    actual publish years."""
+    base_import = _now()
+    # Image A: imported NOW, but post was made 2 years ago.
+    img_a, _ = await _seed_image_with_post(
+        db, sha="a" * 64,
+        image_created_at=base_import,
+        post_date=base_import - timedelta(days=730),
+        artist_name="Aria", external_post_id="A-1",
+    )
+    # Image B: imported NOW (1 min later), post made YESTERDAY.
+    img_b, _ = await _seed_image_with_post(
+        db, sha="b" * 64,
+        image_created_at=base_import - timedelta(minutes=1),
+        post_date=base_import - timedelta(days=1),
+        artist_name="Bea", external_post_id="B-1",
+    )
+    # Image C: filesystem-imported, no primary_post_id, created 5 days ago.
+    img_c = ImageRecord(
+        path="/images/test/c.jpg", sha256="c" * 64,
+        size_bytes=1000, mime="image/jpeg",
+        width=100, height=100,
+        origin="imported_filesystem", integrity_status="unknown",
+    )
+    img_c.created_at = base_import - timedelta(days=5)
+    db.add(img_c)
+    await db.flush()
+
+    svc = GalleryService(db)
+    page = await svc.scroll(cursor=None, limit=10)
+    # Effective-date order: B (yesterday) > C (5 days ago) > A (2 years ago)
+    assert [i.id for i in page.images] == [img_b.id, img_c.id, img_a.id]
+
+    # API exposes both fields explicitly so the UI can show "Posted X / Imported Y".
+    a_payload = next(i for i in page.images if i.id == img_a.id)
+    assert a_payload.posted_at is not None
+    assert a_payload.posted_at < a_payload.created_at
+    c_payload = next(i for i in page.images if i.id == img_c.id)
+    assert c_payload.posted_at is None
+    assert c_payload.effective_date == c_payload.created_at
+
+
+@pytest.mark.asyncio
+async def test_timeline_buckets_use_post_date_when_available(db):
+    """Timeline group-by must follow the same effective_date rule so the
+    UI's year/month navigation surfaces publish-date buckets, not the
+    single FC-scan bucket all migrated images share."""
+    base = datetime(2026, 6, 15, 12, 0, tzinfo=UTC)
+    await _seed_image_with_post(
+        db, sha="1" * 64,
+        image_created_at=base,
+        post_date=datetime(2024, 3, 10, tzinfo=UTC),
+        artist_name="Carl", external_post_id="C-1",
+    )
+    await _seed_image_with_post(
+        db, sha="2" * 64,
+        image_created_at=base,
+        post_date=datetime(2024, 3, 11, tzinfo=UTC),
+        artist_name="Dee", external_post_id="D-1",
+    )
+    await _seed_image_with_post(
+        db, sha="3" * 64,
+        image_created_at=base,
+        post_date=datetime(2025, 9, 1, tzinfo=UTC),
+        artist_name="Eli", external_post_id="E-1",
+    )
+    svc = GalleryService(db)
+    buckets = await svc.timeline()
+    bucket_keys = {(b.year, b.month, b.count) for b in buckets}
+    # Two posts in 2024-03, one in 2025-09 — even though all imported in 2026-06.
+    assert (2024, 3, 2) in bucket_keys
+    assert (2025, 9, 1) in bucket_keys
+    # The FC-import bucket should NOT appear since all 3 images have post_date.
+    assert not any(b.year == 2026 and b.month == 6 for b in buckets)
+
+
+@pytest.mark.asyncio
+async def test_get_image_with_tags_includes_posted_at_when_present(db):
+    base = _now()
+    img, _ = await _seed_image_with_post(
+        db, sha="f" * 64,
+        image_created_at=base,
+        post_date=base - timedelta(days=365),
+        artist_name="Fred", external_post_id="F-1",
+    )
+    svc = GalleryService(db)
+    payload = await svc.get_image_with_tags(img.id)
+    assert payload["posted_at"] is not None
+    # Image's own created_at is still surfaced separately.
+    assert payload["created_at"] != payload["posted_at"]
@@ -227,6 +227,67 @@ async def test_image_posts_creates_source_post_provenance(db, tmp_path):
    )).scalar_one()
    assert prov_count == 1

+    # Phase 4 must also set ImageRecord.primary_post_id so the gallery's
+    # effective_date COALESCE can surface Post.post_date. Operator-flagged
+    # 2026-05-25: without this, IR-migrated images keep sorting by FC's
+    # scan date instead of the original publish date.
+    primary_post_id = (await db.execute(
+        select(ImageRecord.primary_post_id).where(ImageRecord.id == img_id)
+    )).scalar_one()
+    canonical_post_id = (await db.execute(
+        select(Post.id).where(Post.external_post_id == "10001")
+    )).scalar_one()
+    assert primary_post_id == canonical_post_id
+
+
+@pytest.mark.asyncio
+async def test_image_posts_primary_post_id_not_clobbered(db, tmp_path):
+    """If the importer already set primary_post_id (e.g. a downloaded
+    image with a known provenance), phase 4 must NOT overwrite it when
+    re-running tag_apply against the IR migration. The existing
+    download-time linkage is the source of truth."""
+    sha = "9" * 64
+    await _seed_image(db, sha, suffix="9")
+    # Pre-set primary_post_id to a sentinel Post so we can detect a clobber.
+    img_id = (await db.execute(
+        select(ImageRecord.id).where(ImageRecord.sha256 == sha)
+    )).scalar_one()
+    # Build an existing Source + Post for the sentinel.
+    art = Artist(name="Pre-existing", slug="pre-existing")
+    db.add(art)
+    await db.flush()
+    src = Source(
+        artist_id=art.id, platform="patreon",
+        url="https://www.patreon.com/pre-existing",
+    )
+    db.add(src)
+    await db.flush()
+    sentinel_post = Post(
+        source_id=src.id, external_post_id="sentinel-99",
+        post_title="Pre-existing",
+    )
+    db.add(sentinel_post)
+    await db.flush()
+    await db.execute(
+        ImageRecord.__table__.update()
+        .where(ImageRecord.id == img_id)
+        .values(primary_post_id=sentinel_post.id)
+    )
+    await db.commit()
+
+    _write_manifest(tmp_path, image_posts=[
+        {**_POST_ENTRY, "image_sha256s": [sha]},
+    ])
+    await tag_apply.apply_async(db, images_root=tmp_path, dry_run=False)
+
+    # The migration created a NEW Post (external_post_id="10001") and a
+    # new ImageProvenance, but primary_post_id must still point at the
+    # original sentinel.
+    primary_post_id = (await db.execute(
+        select(ImageRecord.primary_post_id).where(ImageRecord.id == img_id)
+    )).scalar_one()
+    assert primary_post_id == sentinel_post.id
+

@pytest.mark.asyncio
 async def test_image_posts_idempotent_on_rerun(db, tmp_path):