Merge branch 'fix/media-meta-unenriched-guard'

1 files changed, 60 insertions, 1 deletions

diff --git a/packages/meshbay-node/tests/test_media_meta_request.py b/packages/meshbay-node/tests/test_media_meta_request.py
index a752825..0a3df12 100644
--- a/packages/meshbay-node/tests/test_media_meta_request.py
+++ b/packages/meshbay-node/tests/test_media_meta_request.py
@@ -9,7 +9,7 @@ real risk here too, since a season folder routinely holds many episodes.
 
 import pytest
 from cryptography.hazmat.primitives.asymmetric.ed25519 import Ed25519PrivateKey
-from meshbay_common.protocol import IndexEntry
+from meshbay_common.protocol import MNP, IndexEntry
 from meshbay_node.indexer.group_index import GroupIndex
 from meshbay_node.media_cache import MediaCache
 from meshbay_node.transport.webrtc_server import WebRTCPeerSession
@@ -165,3 +165,62 @@ async def test_unknown_file_id_is_refused(media_cache):
     await session._do_media_meta_request({"file_id": "nope"})
 
     assert session.sent == [{"type": "error", "detail": "File not found"}]
+
+
+async def test_a_not_yet_enriched_entry_with_no_cache_is_answered_without_a_search(media_cache):
+    """
+    A video the indexer has seen but not enriched yet has no display_title
+    (enrich.py always sets one) and season/episode still None — which the
+    movie/show split reads as "movie" and hands its raw filename to TMDB's
+    movie search. During a slow initial scan with a browser on the Videos
+    tab that is a storm of `search/movie?query=<raw filename>` and bogus
+    cached matches (found live 2026-08-29). With nothing cached it must
+    answer confidence 0 and let the client refetch once the enriched fields
+    arrive — never guess a match from the raw filename.
+    """
+    index = GroupIndex(group_id="g" * 32, sk_node=Ed25519PrivateKey.generate())
+    entry = IndexEntry(id="id-raw", name="Show.S01E01.1080p.WEB.mkv",
+                       path="shows/Show", size=1, type="video", added_at=0)
+    index.add_entry(entry)
+    client = FakeTmdbClient()
+    session = _session(index, media_cache, client)
+
+    await session._do_media_meta_request({"file_id": "id-raw"})
+
+    assert len(session.sent) == 1
+    resp = session.sent[0]
+    assert resp["type"] == MNP.MEDIA_META_RESP
+    assert resp["file_id"] == "id-raw"
+    assert resp["confidence"] == 0
+    assert "tmdb_id" not in resp
+    assert client.searched == [], "no TMDB search for a not-yet-enriched video"
+
+
+async def test_a_not_yet_enriched_entry_is_served_from_cache_without_a_search(media_cache):
+    """
+    The point the operator raised: a restart must not re-query TMDB for a
+    file already resolved. An un-enriched entry (season/episode not yet
+    populated) whose content hash already has a cached match is served
+    straight from that cache, honouring the cached *kind* rather than the
+    provisional "movie" the split would pick — so a show episode keeps its
+    real "tv" match instead of triggering a fresh movie search on its raw
+    filename.
+    """
+    index = GroupIndex(group_id="g" * 32, sk_node=Ed25519PrivateKey.generate())
+    entry = IndexEntry(id="id-known", name="Show.S02E05.1080p.WEB.mkv",
+                       path="shows/Show", size=1, type="video", added_at=0)
+    index.add_entry(entry)
+    await media_cache.set_file_tmdb("id-known", "1396", "tv")
+    await media_cache.set_tmdb_meta("1396", "tv", {
+        "title": "The Cached Show", "original_title": "The Cached Show",
+        "first_air_date": "2008-01-20", "confidence": 1.0,
+    })
+    client = FakeTmdbClient()
+    session = _session(index, media_cache, client)
+
+    await session._do_media_meta_request({"file_id": "id-known"})
+
+    resp = session.sent[0]
+    assert resp["tmdb_id"] == "1396"
+    assert resp["title"] == "The Cached Show"
+    assert client.searched == [], "a cached match must not be re-searched on restart"