fix(incremental): fall back to mtime when stale last_indexed_sha exists — closes #117

[Wolfvin]

What

Fixes RED TEAM issue #117: find_changed_files() silently returned empty (changed=[], new=[], deleted=[]) on non-git workspaces that had a stale last_indexed_sha bookmark in the DB, causing incremental scan to miss file modifications.

Root cause

find_changed_files() in scripts/incremental.py tried the git-aware path first. If a last_indexed_sha bookmark existed in the DB (from a previous scan when the workspace was a git repo), it called _git_changed() and _git_untracked(). Both return [] when:

• (a) there are no changes, OR
• (b) the workspace is not a git repo / git unavailable

The two cases are indistinguishable, so on a non-git workspace with a stale SHA, the function returned empty lists without falling back to the mtime path.

Fix

After retrieving last_indexed_sha, verify the workspace is still a git repo by calling get_current_sha(). If it returns None (workspace is not a git repo or git unavailable), fall through to the mtime-based _find_changed_files_mtime() path instead of trusting the empty git result.

last_sha = _git_last_sha(workspace, db_path)
if last_sha:
    # Issue #117: verify the workspace is STILL a git repo before
    # trusting the git-diff result.
    current_sha = _git_current_sha(workspace)
    if current_sha is None:
        # Workspace is no longer a git repo (or git unavailable).
        # Fall through to mtime path so changes are still detected.
        logger.debug(...)
    else:
        # ... existing git-diff logic ...
        return changed_abs, [], deleted_rel

Verification

Reproduced the bug from #117 and verified the fix:

# 1. Copy clean_app fixture to temp dir (non-git workspace)
# 2. Run full scan (creates DB)
# 3. Inject stale last_indexed_sha into registry_meta
# 4. Modify src/utils.py
# 5. Run incremental scan

# Before fix: changed_files_count: 0  (BUG)
# After fix:  changed_files_count: 1  (correct)

All 6 previously-failing test_graph_incremental.py tests now pass:

• test_renamed_function_reflected_in_graph ✅
• test_added_function_reflected_in_graph ✅
• test_removed_function_reflected_in_graph ✅
• test_graph_has_no_orphan_edges_after_removal ✅
• test_incremental_scan_with_modification_updates_graph ✅
• test_auto_scans_on_fresh_workspace ✅

Test plan

• Reproduction script confirms fix (changed_files_count: 1)
• 6 previously-failing incremental tests now pass
• No new regressions in full test suite (1183 passed, 1 pre-existing [RED TEAM] test_returns_correct_counts expects 97 CALLS edges but gets 76 — pre-existing parser under-count #118)
• CI: full test suite passes

Risk

Low. The fix only adds a git-repo verification check before trusting git-diff results. If the workspace IS a git repo, behavior is unchanged. If it's NOT a git repo, we now correctly fall back to mtime instead of silently returning empty.

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud