Freeze the clock in the conference archiving regression tests - #524

Routine review

1. Request

No linked issue; the PR's own words: freeze the clock for TestConferenceArchiving in tests/regression/test_conference_archiving.py so its date-relative fixtures stop crossing New Year, covering test_archive_boundary_conditions, test_identify_past_conferences, test_bulk_archiving_process, test_archive_preserves_data_integrity, test_archive_with_extended_deadline and test_archive_with_timezone_handling, and make test_archive_boundary_conditions assert real AoE/UTC behaviour instead of only isinstance(..., bool).

2. Checked against

• date: Mon Nov 2 11:50:36 UTC 2026.
• No external facts in this review. The AoE claim was checked against utils/sort_yaml.py:72-79 (no timezone → Etc/GMT+12, so a naive "now − 1 h" deadline normalises to now + 11 h UTC) and utils/sort_yaml.py:111-112.

3. Needs your decision

None. The boundary-test rewrite is named in the PR body and stays inside the file being fixed.

4. Findings

Important: None.

Nits:

• Possible nit: FROZEN_NOW = "2026-06-15 12:00:00" (tests/regression/test_conference_archiving.py:27) has no zone. freezegun reads a naive string as UTC, so it works, but "2026-06-15 12:00:00+00:00" would make the boundary test's AoE/UTC reasoning readable without knowing that default.
• The PR body lists 8–9 Jan as a failure window for test_archive_with_timezone_handling, but the base file passed under a clock frozen at 2027-01-08 12:00 UTC. The window may be narrower or depend on the hour. This doesn't affect the fix.

5. Tests

The invariant ("no fixture date span crosses a year boundary, whatever the real date") is covered by the autouse _frozen_clock fixture, which applies to every test in the class, including the date-building fixtures. The boundary test now pins both sides: AoE still open (False), UTC passed (True).

I ran the base and changed versions of the file with the whole session frozen at each listed window (a pytest_configure plugin calling freeze_time(...).start()). Python 3.11 through uv with the REVIEW.md package set:

So the base fails and the change passes in every window I could reproduce.

Change, real clock:

tests/regression/test_conference_archiving.py: 14 passed in 2.32s
tests/ (full): 2560 passed, 5 skipped, 3 xfailed, 2 xpassed, 251 warnings in 141.09s (0:02:21)
uvx ruff@0.4.4 check utils tests: All checks passed!

GitHub statuses on b2e9032 at review time: pre-commit.ci - pr: success, Pre-commit Checks, Data Validation, CodeQL and labeler success, Python Tests in_progress.

6. Bounds

• Route: cloned the repo, fetched refs/pull/524/head, and diffed against origin/main. PR metadata came from the REST API.
• The trigger's head was 940e967. The PR head is now b2e9032, a pre-commit.ci auto-fix that only removes one blank line. I reviewed b2e9032.
• Read in full: the diff, REVIEW.md, the edge_case_conference fixture, and sort_by_cfp / sort_by_date_passed in utils/sort_yaml.py. Skimmed: CONTRIBUTING.md and AGENTS.md (they cover data, and none of it applies to a test-only change).
• freezegun is already a declared test dependency (pixi.toml:50).
• Not run: the REVIEW.md data check (sort_yaml.py), because no _data files changed; npm test, because no JS changed; pixi itself, which isn't in the sandbox.
• Pre-existing: none noticed.

Generated by Claude Code