fix: FTS query tokenization and daemon/foreground DB contention (v0.9.1) - #86
Merged
Merged
Conversation
ashu17706
commented
Aug 25, 2026
ashu17706
commented
Contributor
- fix(db): let QMD own busy_timeout and WAL so cold-open DDL is protected
- fix(search): tokenize FTS queries instead of interpolating them raw
- chore(release): 0.9.1
createStore() runs the WAL migration and the FTS trigger DDL before it returns, so the busy_timeout Smriti set on the returned handle arrived after the only section that needed it. That DDL raced at bun:sqlite's default of 0. The daemon makes this reachable rather than theoretical: defaultFlushAgent opens the store, ingests and closes it per flush, so a watching daemon re-runs createStore() on every debounced change. flushChain serializes flushes only within the daemon process, and there is no cross-process lock, so any overlap with a foreground recall/embed/ingest failed on contact with 'database is locked'. Upstream moved busy_timeout and a retrying WAL migration into openDatabase(), the only place that runs before the DDL. Bumps the qmd submodule to that (155 commits, v2.1.0 -> v2.8.3). Six concurrent cold opens on one fresh DB: 1/6 succeeded before, 6/6 after. Claude-Session: https://claude.ai/code/session_01HWzCvpW1yLZD3SJTXEojvR
searchFiltered pushed the user's query straight into the FTS5 MATCH expression, so punctuation was parsed as FTS5 grammar rather than as text. 'smriti search node-llama-cpp' failed with 'no such column: llama' and 'smriti search qmd 2.8.3' with 'fts5: syntax error near "."'. Hyphenated names are everywhere in these transcripts, so this hit ordinary queries, not edge cases. buildMemoryFTS5Query had the matching defect on the recall path: it stripped punctuation rather than splitting on it, collapsing '2026.4.10' to '2026410' while the porter unicode61 tokenizer had indexed '2026', '4' and '10'. Silent zero-hit rather than an error. Both now split on the same boundaries the tokenizer uses and AND the parts, so a query tokenizes the way the indexed text did. The builder is exported and shared, and searchFiltered parenthesises the group so the column filter binds to all of it. Same class as QMD upstream #563, which does not reach Smriti because these are Smriti's own query builders. Claude-Session: https://claude.ai/code/session_01HWzCvpW1yLZD3SJTXEojvR
Contributor
Benchmark Scorecard (ci-small)Bench Scorecard (ci-small)threshold: 20.00%
Summary: WARN (4 metrics) |
v0.9.1 was already tagged on 2026-08-02 (pointing at eaf82b8), so the next stable version is 0.9.2. Also realigns package.json, which the auto-release job does not bump and which still read 0.9.0 at v0.9.1. Claude-Session: https://claude.ai/code/session_01HWzCvpW1yLZD3SJTXEojvR
ashu17706
force-pushed
the
release/v0.9.1
branch
from
August 25, 2026 07:22
fa560d4 to
8c1c641
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.