[HuaTNA]

SetCookie(expires=...) formats non-UTC datetimes with their local wall-clock fields followed by GMT, shifting the expiration instant. A future cookie can consequently appear expired to HTTPX.

Convert aware datetimes to UTC before formatting. The focused parameterized regression covers both offset directions (including a half-hour offset), a year boundary, and the existing naive-datetime/string behavior; the existing all-attributes test covers UTC.

Fixes #336.

Review follow-up: adds one public respx.mock / respond(cookies=...) / httpx.Client regression asserting the stored cookie-jar expiration matches the aware datetime's epoch. It fails against the original implementation and passes with this fix.

Validation on macOS:

• Regression before the fix: 2 failed, 3 passed; after the fix: 5 passed.
• nox -s test-3.11: 326 passed, 2 remote pass-through tests skipped, 100% configured statement/branch coverage (Python 3.11.15).
• Project mypy session on Python 3.10 and all pinned pre-commit hooks passed.
• Local public respx.mock / respond(cookies=...) / HTTPX 0.28.1 client checks confirm identical cookie-jar expiry timestamps for UTC and both offset directions. Additional local controls cover Toronto DST folds and a naive-like tzinfo.

AI assistance was used for code inspection, implementation, and local test execution. The issue was found through code analysis and reproduction, not a production incident.

[Pavel-Shcherbo]

Reviewed commit 0d0984e on CPython 3.14.6 with HTTPX 0.28.1.

tests/test_utils.py --no-cov: 5 passed. I also checked the cookie jar through respx.mock / respond(cookies=...) / httpx.Client: UTC offsets -12:00, -05:00, -03:30, +00:00, +05:30 and +14:00, plus both Toronto DST folds on 2030-11-03. All eight retain int(expires.timestamp()) at this head; seven fail against the base implementation. Three further controls preserve naive datetime, naive-like tzinfo (utcoffset() is None) and string behavior.

The utcoffset() guard handles awareness correctly without converting naive-like values through the machine's local timezone. I found no blocking issue in this diff.

One optional follow-up: retain a public-API cookie-jar assertion in the project suite, alongside the formatting test. That would directly cover the expiration behavior users observe. This review covers the cookie change and focused controls; I have not run the full RESPX compatibility matrix.

[HuaTNA]

Thanks for the detailed review. Added a focused public-API regression in 2ea43a1: a response configured with SetCookie(expires=...) passes through respx.mock, respond(cookies=...) and httpx.Client, then asserts the stored cookie-jar expiry equals int(expires.timestamp()).

It fails against the original implementation (a 5h30m shift) and passes with the fix. The documented Python 3.11 nox session now has 326 passing tests, 2 remote pass-through skips and 100% configured statement/branch coverage; pinned hooks and mypy pass. I have not run the full compatibility matrix.

[Pavel-Shcherbo]

Independently checked 2ea43a1 on Python 3.12.9 / HTTPX 0.28.1: the new public-API regression plus tests/test_utils.py pass (6 tests). Restoring only the original SetCookie implementation makes the unchanged new regression fail: the cookie jar stores 4102464600 instead of 4102444800, a 19,800-second shift. This covers the end-user behavior I suggested retaining. Thanks for adding it.

This follow-up is limited to those synchronous checks; I have not independently verified the full compatibility matrix or coverage result.