test_file_utils umask tests assert against a umask the code never reads (import-time _UMASK), so they only pass on a 0o022 host

オープン 初心者向け
  • 何か問題がありますか？
#1,071 コメント 0 件 リアクション 0 件 担当者 0 名 GitHub で見る

メンテナーはふだん 1 日以内に返信

@neverbiasu がすでに取り組んでいます。

2026年11月4日 から。

  • #1105 @neverbiasu による — オープン

評価

難易度
2/5
見積もり時間
1〜3時間
初心者へのやさしさ
88/100
issue の種類
バグ
明瞭さ
明確に書かれている
活発さ
活発
技術スタック
python
領域
testing

調査の方向性

tests/comfy_cli/test_file_utils.py と comfy_cli/file_utils.py で、_UMASK の取得箇所を確認してください。実装が読み取る値を制御できるように、アトミック書き込みの2つのテストを更新し、0o077 など2つ目の umask のテストも追加してください。対象を絞った file-utils テストを実行してください。プロセスの umask を変更せずに、期待されるファイルモードを検証する必要があります。

索引モデルが issue の本文から書いたものです。

説明

bug

Summary

test_atomic_write_text_new_file_uses_umask_default and test_atomic_write_bytes_new_file_uses_umask_default set a umask at runtime and assert the resulting file mode. file_utils captures the umask once at import, so the os.umask(0o022) the tests call has no effect on the code under test. The assertion therefore only holds on a host whose ambient umask already happens to be 0o022, and fails outright on any other.

Observed on a host with umask 0007:

tests/comfy_cli/test_file_utils.py:163: AssertionError
assert 432 == (438 & ~18)
#      ^ 0o660, i.e. 0o666 & ~0o007 (the AMBIENT umask)
#              ^ 0o644, i.e. 0o666 & ~0o022 (the umask the test set)

Both tests fail the same way; test_atomic_write_bytes_… at :313.

Cause

comfy_cli/file_utils.py:98-105:

# The process umask, captured once at import. os has no getter, so reading it
# ...
# overlapping probes can restore each other's zero and strand the umask at 0.
_UMASK = os.umask(0)
os.umask(_UMASK)

Capturing once is deliberate and the comment gives a good reason for it. The tests simply do not account for it: they manipulate the real process umask, which _atomic_write no longer consults.

So this is a test defect, not a product one. The product behaviour the tests mean to pin — "a new destination gets the umask-derived default, not mkstemp's hardcoded 0600" — is correct and worth keeping pinned; the tests just pin it against the wrong knob.

Suggested fix

Patch the captured value rather than the process umask, so the test controls what the code actually reads:

def test_atomic_write_text_new_file_uses_umask_default(tmp_path, monkeypatch):
    monkeypatch.setattr(file_utils, "_UMASK", 0o022)
    target = tmp_path / "fresh.json"
    atomic_write_text(target, "data")
    assert stat.S_IMODE(os.stat(target).st_mode) == (0o666 & ~0o022)

That also removes the os.umask save/restore, which is process-global state the current tests mutate.

A second case worth adding while in there: assert the same for a non-0o022 umask (0o077, so 0o600), which would have caught this — a single hardcoded umask value cannot distinguish "the code honours the umask" from "the host umask matches the literal in the assertion".

Context

Found while verifying that pre-existing failures were unrelated to a converter change. Two other environment-sensitive failures turned up in the same run and are worth separating:

  • Five tests/comfy_cli/command/test_code_search.py::TestCodeSearchCLI cases fail when the suite runs from inside an AI coding tool, because CLAUDECODE / AI_AGENT in the environment flip detect_caller() and code-search then passes agent_origin=True. Fixed in #1072.
  • test_node_deps.py::test_non_pep440_installed_version_is_unknown_not_a_crash reports mismatch where it expects unknown. Not agent-env related (it fails with those vars unset) and I have not diagnosed it; noting it so it is not assumed covered by the other two.

None of these fail in CI, which is why they have survived: CI's environment happens to satisfy every one of the assumptions.

Task Decomposition

Not decomposed: this is a two-test change in one file (swap os.umask(...) for monkeypatch.setattr(file_utils, "_UMASK", ...), plus one extra parametrised umask value). Stated explicitly per the issue-filing convention.

主要言語
Python
スター
1k
フォーク
176
平均マージ
1日 11時間
マージ済み PR（30日）
87

環境構築

はじめの一歩

  1. issue を最後まで読み、次にプロジェクトのコントリビューションガイドを読みます。
  2. 着手することを issue にコメントします — 二人が同じ作業をするのを防げます。
  3. リポジトリをフォークし、ブランチを切って変更します。
  4. issue 番号を参照したプルリクエストを送ります。

Comfy-Org/comfy-cli のほかの issue

Comfy-Org/comfy-cli の issue をすべて見る

似ている issue

Python の issue をもっと見る

新しい issue をメールで受け取る

初心者向けの GitHub issue を短くまとめたダイジェスト。