codesearch index: incremental sync skips build_index on interrupted DB

[v1b3coder]

Problem

When codesearch index . is cancelled/interrupted mid-way (e.g. timeout during Phase 2), the HNSW vector index is never built — build_index() is intentionally skipped in the cancellation path.

On the next run, codesearch index . (incremental sync) checks file_meta, finds no changed files, and exits early with "Database is up to date!" — without ever calling build_index(). The database remains in a broken state and search fails with:

Error: Index not built. Call build_index() after inserting chunks.

Root cause

In src/index/mod.rs, the cancellation handler skips build_index():

// Handle cancellation: exit quickly without blocking on build_index
if cancelled {
    // Don't call build_index() — it blocks for 10-30 seconds on large datasets.
}

The incremental sync path (perform_incremental_refresh) only calls build_index() per-file during reindexing of changed files. If no files changed (they were already indexed before cancellation), build_index() is never called.

Compare: serve daemon handles this correctly

In src/serve/mod.rs, when opening an existing DB, the serve daemon checks vstore.stats().indexed and auto-calls build_index():

match vstore.stats() {
    Ok(s) if s.total_chunks > 0 && !s.indexed => {
        info!("Building vector index for ...");
        if let Err(e) = vstore.build_index() { ... }
    }
}

Suggested fix

Add the same check to the incremental sync path in codesearch index .: after processing file changes, check if the vector store has chunks but no index, and call build_index() if needed.

The StoreStats.indexed field already exists and is exposed via codesearch stats ("Indexed: ✅ Yes / ❌ No"). The CLI just needs to act on it.

[flupkede]

Fixed in v1.0.160

This is resolved as of v1.0.160 (released 2026-06-01, also in v1.0.162).

What was done

A new ensure_hnsw_index_if_needed() helper was added to src/index/mod.rs (commit in the Release v1.0.160 PR #104). It is called as a safety-net at the end of every incremental sync — even when no files changed:

// Safety net: if a previous run was cancelled/interrupted mid-way,
// the HNSW vector index may never have been built. Detect this and
// rebuild now so the database is usable without requiring --force.
match ensure_hnsw_index_if_needed(&db_path, model_type.dimensions()) {
    Ok(true) => { /* log: rebuilt */ }
    Ok(false) => {}
    Err(e) => { /* log warning, continue */ }
}

The helper checks VectorStore::stats():

• total_chunks > 0 && !indexed → calls build_index() and returns true
• already indexed or empty DB → no-op, returns false
• stats unreadable → logs a warning, returns Ok(false) (non-fatal)

This mirrors the existing check in src/serve/mod.rs and closes the gap for the CLI path. Three unit tests cover the three cases (needs rebuild, already indexed, empty DB).

Result: Running codesearch index . after an interrupted run now detects and repairs the unbuilt HNSW index automatically, without requiring --force.