Row-scan loops swallow SQLITE_CORRUPT — queries return plausible truncated/empty results with no error

[kriswill]

Version: codebase-memory-mcp 0.8.1 (built from source)
Platform: macOS (Apple Silicon) · Install channel: Built from source · Binary variant: standard

What happened, and what did you expect?

Every row-scan loop in the store layer has the shape while (sqlite3_step(stmt) == SQLITE_ROW) with the terminal code discarded, so SQLITE_CORRUPT mid-scan is indistinguishable from a clean end of results. Against a page-corrupted DB, counts are answered from covering indexes (correct) while row fetches die at the first corrupt table page (truncated/empty) — producing contradictory, error-free output across every query surface:

• search_graph label mode: {"total": 379, "results": [], "has_more": true}
• query_graph MATCH (n:Function) RETURN count(n) → 64, while get_graph_schema reports 379 Function nodes
• some labels unreachable entirely via Cypher (zero rows even for count()) while the schema reports dozens of nodes
• BM25 query= search: total: 0 despite matching symbols (corrupt FTS shadow tree)
• get_architecture with path=: correct scoped_total_nodes, but node_labels: [], edge_types: [], scoped_total_edges: 0
• search_code: grep matches present, graph enrichment total_results: 0

Expected: a loud "database is corrupted — re-index" error instead of plausible wrong answers. Silent truncation cost weeks of confusion in the field: the symptoms read as extraction/marshaling bugs and indexing blind spots, when the store data was simply unreadable.

Reproduction

Any public repo (e.g. colinhacks/zod), released binary + sqlite3 + dd only:

codebase-memory-mcp cli index_repository --repo-path /tmp/zod --mode full

# corrupt three mid-file pages of the cache DB (simulates a torn artifact — see #895)
DB=~/.cache/codebase-memory-mcp/<project>.db
PS=$(sqlite3 "$DB" 'PRAGMA page_size;'); PC=$(sqlite3 "$DB" 'PRAGMA page_count;')
dd if=/dev/zero of="$DB" bs=$PS seek=$((PC/2)) count=3 conv=notrunc

codebase-memory-mcp cli get_graph_schema --project <project>        # full label counts (index-covered)
codebase-memory-mcp cli query_graph --query 'MATCH (n:Function) RETURN count(n)' --project <project>   # truncated count, no error
codebase-memory-mcp cli search_graph --label Function --limit 5 --project <project>  # total=N, results=[], no error

Project scale (field occurrence)

3,977 nodes / 6,666 edges / ~480 files

Cascade

This is the concealer: it let the torn artifact from #895 serve wrong answers indefinitely. Filed alongside two related same-class issues (stale-WAL replay on DB swap; malformed-JSON emission), linked in a comment once created.

Maintainer note

I have a working private fix for this (with reproduce-first regression tests) that I can submit as a PR. Before I do, I'd prefer you groom this issue — confirm the framing, tag, and prioritize — so the PR lands against an agreed shape.

Confirmations

• I searched existing issues and this is not a duplicate.
• My reproduction uses shareable code (a dummy snippet or a public OSS repository), not proprietary code.

[kriswill]

Companion issues in this cascade:

#	Issue	Role in the cascade
#895	Fast artifact export raw-copies a live WAL-mode DB	Producer — the torn artifact this bug concealed
#896	Row-scan loops swallow SQLITE_CORRUPT	Concealer — total=N, results=[] with no error (this issue)
#897	Fresh DB install leaves old -wal/-shm behind	Second producer — stale WAL replayed over the new file
#898	Sequential pipeline emits invalid JSON	Gate blocker — aborts a quick_check-based integrity gate