fix(mcp): keep search index type in vector hydration#983
Conversation
Co-authored-by: Cursor <cursoragent@cursor.com>
|
Closing in favor of #984, which contains the same patch with the required DCO sign-off trailer. |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 16144196fc
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
| for si_key, similarity in similarity_by_si_key.items(): | ||
| row = search_index_rows.get(si_key) |
There was a problem hiding this comment.
Now that vector hydration can append multiple SearchIndexRows with the same numeric id but different type, default hybrid searches still drop one of them because _search_hybrid stores FTS/vector rows in rows_by_id, fts_scores, and vec_scores keyed only by row.id (see src/basic_memory/repository/search_repository_base.py around lines 2207-2239). In semantic-enabled clients where hybrid is the default, a vector match for entity:4 and relation:4 will collapse to whichever row is assigned last, so the collision this patch fixes for vector-only remains user-visible in hybrid mode; carry the (type, id) key through the fusion maps as well.
Useful? React with 👍 / 👎.
1 participant
Fixes #982.
Summary
(type, id)instead of bareid, so entity/relation/observation rows with the same search_index id no longer overwrite each other.WHERE id IN (...)query shape and apply type discrimination in Python, matching the existing filtered-search guard.entity:4:0andrelation:4:0both hydrating successfully.Test plan
uv run pytest tests/repository/test_vector_threshold.py tests/repository/test_vector_pagination.pyuv run ruff format src/basic_memory/repository/search_repository_base.py tests/repository/test_vector_threshold.py tests/repository/test_vector_pagination.pyuv run ruff check src/basic_memory/repository/search_repository_base.py tests/repository/test_vector_threshold.py tests/repository/test_vector_pagination.py