artifacts/codex-audit/test-gaps.md

Test gaps

Existing test coverage observed

Test modules inspected:

• tests/test_codex_data.py
• tests/test_processes.py
• tests/test_startup.py
• tests/test_packaging.py
• tests/test_crash_safety.py

Commands run:

• python3 -m pytest: 16 passed.
• python3 -m compileall -q src tests: passed.

Covered areas:

• Codex thread listing hides archived sessions by default.
• Archived search path works.
• Missing state DB returns zero stats.
• Locked SQLite DBs return quickly without exceptions.
• Process scanner filters simple ps output and reads /proc metadata in mocked live mode.
• Uptime formatting.
• Optional GUI dependency packaging.
• Missing GI/GTK startup message.
• Refresh fan-out and in-flight guards via AST tests.
• systemd restart policy guard.
• WebKit exclusion.

Important gaps

1. GUI behavior is mostly untested.

   • No tests instantiate/render rows under GTK.
   • No tests validate button wiring for resume/open-folder/stop.
   • No tests cover tray menu behavior, close-to-tray behavior, or status text transitions.
   • No tests cover search entry behavior beyond data-layer query filtering.

2. SQLite schema-drift handling is under-tested.

   • The locked-DB test uses a minimal table that triggers sqlite errors safely, but there are no explicit tests for real Codex schema changes.
   • Missing columns, renamed columns, null-heavy rows, unexpected timestamp units, or invalid token types are not separately asserted.
   • list_logs has no normal-path fixture test for logs_2.sqlite structure.

3. Silent failure behavior is untested from the UI perspective.

   • Data functions return empty lists/stats on sqlite errors. No tests assert whether the UI can tell "no sessions" apart from "database unreadable".

4. Process matching false positives are under-tested.

   • Current tests use simple positive/negative lines.
   • No tests cover command arguments, paths, or filenames containing keywords accidentally.
   • No tests define which process families should be stoppable in a Codex-first dashboard.

5. Destructive process action is minimally covered.

   • terminate_process directly sends SIGTERM, but no tests cover _confirm_stop wiring with mocked dialog response and mocked termination.
   • No test ensures the app never stops its own dashboard process from the UI list beyond scanner-level filtering.

6. Packaging/install behavior is lightly checked only.

   • No tests for desktop entry correctness beyond WebKit absence.
   • No tests that hard-coded service paths match current repo or installed package entry points.
   • No console_scripts entry point exists to test.

7. Claude data helpers have no tests.

   • claude_data.py includes JSON, JSONL, grouping, stats, session, and event parsing helpers but is not covered by the current test suite.
   • If this module stays in scope, it needs malformed-input and large-directory tests. If it is intentionally out of P0 scope, the repo should clarify that.

Suggested next test additions

• tests/test_logs_data.py: fixture logs database; normal listing, search, missing/corrupt DB behavior.
• tests/test_schema_drift.py: expected Codex schema variants; missing/renamed columns; invalid/null values.
• tests/test_process_false_positives.py: keyword-in-path/argument cases and family/stoppable policy.
• tests/test_ui_actions.py: command construction and stop confirmation with GTK/subprocess/os.kill mocked or moved behind adapter functions.
• tests/test_install_assets.py: service/desktop paths, relocatability assumptions, package launch entry point if added.
• tests/test_claude_data.py only if Claude helpers remain maintained code.