Skip to content

CI: Windows test jobs fail on every run since 2026-08-26 (journal rollup PermissionError + CLI-outside-PATH lookup) #297

Description

@CryptoJones

Both Windows matrix jobs (windows-latest, 3.10 and 3.14) have failed on every run since 2026-08-26 — on main and on every PR — independently of #294 (which is now fixed by #295). Linux and macOS are green.

The 8 failures are the same on every run:

FAILED tests/test_ai_usage.py::test_resolve_finds_a_cli_outside_path - AssertionError: installed CLI outside PATH was not found
FAILED tests/test_cli.py::test_rollup_compacts_a_week_of_journals - PermissionError: [Errno 13] Permission denied
FAILED tests/test_journal.py::test_rollup_week_produces_parseable_summary_and_archives - PermissionError: [Errno 13] Permission denied
FAILED tests/test_journal.py::test_rollup_delete_removes_dailies - PermissionError: [Errno 13] Permission denied
FAILED tests/test_journal.py::test_rollup_default_respects_retention - PermissionError: [Errno 13] Permission denied
FAILED tests/test_journal.py::test_rollup_is_idempotent_per_week - PermissionError: [Errno 13] Permission denied
FAILED tests/test_journal.py::test_rollup_note_stays_out_of_listings - PermissionError: [Errno 13] Permission denied
FAILED tests/test_journal.py::test_rollup_late_daily_keeps_archived_days - PermissionError: [Errno 13] Permission denied

Two distinct causes:

  1. Journal rollup on Windows — seven PermissionError: [Errno 13] failures. The rollup archives/deletes dailies; on Windows a file that is still open (or was just written) cannot be renamed/unlinked, which is the classic NTFS sharing-violation shape. Likely a handle left open by the writer or the reader before the archive move.
  2. test_resolve_finds_a_cli_outside_path — the resolver that finds model CLIs installed off PATH (fix(verify): find model CLIs that are installed but not on a hook's PATH #271) does not find the fixture on Windows; probably an executable-suffix (.exe/.cmd) or a ~/.local/bin-style path assumption.

Neither job is a required check, so mergeStateStatus reads UNSTABLE rather than BLOCKED, which is why PRs still merge — but it means the Windows matrix currently carries no signal at all. Example run: https://github.com/CryptoJones/omind/actions/runs/33800438606

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions