Skip to content

armgit / avibe

Public Imported from GitHub
0

Make your computer a home for your AI partner. Avibe is the local-first Agent OS: Vibey, its built-in AI partner, works out of the box, and Claude Code, Codex & OpenCode run unchanged beside it — reach them from your browser, phone, or any chat app.

5.0 KB
Raw Blame

Python Test Value and CI Performance Audit

This lane audited Python tests under tests/, excluding test_ci_*.py, from the shared contract at e089ae442. The goal was to remove tests without an independent contract and reduce measured setup or harness overhead while preserving real API, authentication, state, migration, and failure semantics.

Deletions

tests/test_memory_removal.py::test_config_write_transaction_loads_test_owned_snapshot

The test only loaded a freshly saved default config through config_write_transaction and compared one default field. It did not exercise locking, nesting, concurrent writers, mutation, or fresh state. Those are the actual transaction contract and remain covered by the primary owner tests:

  • tests/test_config_write_transaction.py::test_shared_config_lock_allows_nested_load_save_transaction
  • tests/test_config_write_transaction.py::test_transaction_reentrant_with_save
  • tests/test_config_write_transaction.py::test_transaction_load_and_mutator_wait_for_migration_file_lock
  • tests/test_config_write_transaction.py::test_transaction_blocks_second_process_until_first_releases

The retired-memory tests still cover removal from round trips, migration persistence, concurrent-writer custody, opaque user data preservation, and legacy shape handling.

tests/test_removed_feature_contract.py::test_removed_cli_command_is_unknown

The test launched the source checkout with python -m vibe memory status and asserted argparse exit code 2. This is the same parser contract already probed by the installed artifact owner:

tests/test_distribution_artifacts.py::test_installed_wheel_imports_and_cli_without_memory

That test imports the built wheel and runs the same unknown command, alongside the installed-package import and removed API route checks. The distribution lint/installer job requires this installed-wheel test. The removed HTTP method matrix and SPA fallback test remain because they cover distinct route and UI semantics.

Fixture and harness changes

Reuse migrated SQLite templates

test_vibey_agent.py and test_claude_output_recovery_ledger.py now use the existing sqlite_db_factory template facility instead of replaying the full SQLite importer migration chain for every test. Each test still receives its own mutable database copy. Vibey tests continue to use the default SQLite path because the real adapter and message mirror resolve that path themselves; the module fixture reuses the shared test-owned copy when the autouse seeder has already created it.

The factory's no-overwrite and isolation guarantees were checked separately by tests/test_ci_test_fixtures.py; that file was not modified.

Reduce feedback test server shutdown polling

test_feedback_intake_recovery.py keeps the real 10-second client network timeout in both stall cases. Its local HTTP server now runs serve_forever(poll_interval=0.01), reducing only the time for fixture shutdown to observe the stop request. No production timeout, barrier, retry, 429 persistence, or recovery assertion changed.

Measurements

The complete CI attribution files /tmp/avibe-ci-attribution/38258683681-metrics.json and /tmp/avibe-ci-attribution/38261750469-metrics.json contained all 536 files and successful exits. They identified repeated migration setup and feedback fixture teardown as actionable costs.

Focused local measurements used isolated uv Python 3.12.

Focus Before After Result
test_feedback_intake_recovery.py 43 passed, 88.29s pytest 43 passed, 66.61s pytest 21.68s faster; stall calls remained 10.30s and 10.33s
test_vibey_agent_context.py 27 passed, 121.00s pytest / 134.54s real 27 passed, 22.30s pytest / 27.22s real migrated template reused
test_claude_output_recovery_ledger.py CI reference: 38.35s PR / 30.02s master 75 passed, 23.44s local migrated template reused
test_vibey_agent.py + test_vibey_agent_context.py — 157 passed, 281.40s pytest full focused consumer verification
test_ci_test_fixtures.py — 12 passed, 30.20s factory consumer contract

The full changed-test verification also passed:

  • test_memory_removal.py plus test_removed_feature_contract.py: 36 passed
  • test_claude_output_recovery_ledger.py: 75 passed on an isolated rerun
  • test_feedback_intake_recovery.py: 43 passed

The first ledger run was concurrent with the removal tests and had one time-sensitive retry assertion; the isolated rerun passed all 75 tests.

Retained slow coverage

No deletion was justified in the remaining high-cost files. The auth setup scenarios and SQLite migration tests exercise durable custody and released schema transitions. The pairing and scheduled-task suites cover independent recovery and lifecycle boundaries. The model-hub source-probe matrix varies protocol, outcome, credential identity, and mutation races; the shell-writer matrix reaches the real scan and apply boundary with distinct parser roles. They remain follow-up profiling candidates after the integrated CI run.