mirror of
https://github.com/LearningCircuit/local-deep-research.git
synced 2026-06-16 03:51:07 +03:00
057a4ceab8
Two tests in test_credential_store_behavior.py used real time.sleep(1.5)
to wait for TTL expiration:
test_credentials_expire_after_ttl (TTL=1, sleep 1.5)
test_expired_entries_cleaned_on_store (TTL=1, sleep 1.5)
The SUT (credential_store_base.py) compares time.time() to expires_at:
src/local_deep_research/database/credential_store_base.py:47
\"expires_at\": time.time() + self.ttl
src/local_deep_research/database/credential_store_base.py:73
if time.time() > entry[\"expires_at\"]:
freezegun fully mocks time.time(), so swapping time.sleep(1.5) for
frozen.tick(1.5) gives identical behavior without the wall-clock wait.
NOT migrated:
test_credential_store_extended.py:172 -
test_expired_entry_returns_none_after_ttl
Its docstring explicitly says \"This test deliberately uses a real
time.sleep instead of freeze_time so that the rest of the freezegun-
based suite is validated against the actual time.time() clock at least
once. Don't migrate this one without keeping another integration
anchor.\" — respecting that intent.
Time saved per CI run: ~3s (was 1.5 + 1.5 = 3s real wait, now near-zero).
Eliminates wall-clock-jitter flakes around the int(time.time())
truncation that the original tests already worked around with 1.5s buffer.
Verified locally:
pytest tests/database/test_credential_store_behavior.py
=> 30 passed in 12.45s
test_credentials_expire_after_ttl: 6.37s (was longer)
test_expired_entries_cleaned_on_store: 0.52s (was ~1.5s sleep)
Twenty-ninth PR in the systematic test-quality review series.