GeezerVison

Copy link

Copy Markdown

Member

picnixz requested a review from gpshead as a code owner October 19, 2026 22:37

gpshead approved these changes Oct 19, 2026

View reviewed changes

Sorry, something went wrong.

Uh oh!

There was an error while loading. Please reload this page.

62 checks passed

picnixz deleted the meta/crypto/c/hmac/33-45/gil-issue-159786 branch October 20, 2026 13:01

…asing the lock (GH-159805) (#159825)

gh-159786: Convert HACL* update() errors in `_hmac` after releasing the lock (GH-159805)

* gh-159786: Convert HACL* update() errors in `_hmac` after releasing the lock

`_hacl_hmac_state_update_once()` converted the HACL* exit code into an
exception while the thread state was detached and, for HMAC.update(),
while the object's mutex was held. On debug builds, this re-attached the
thread state under the mutex and then called PyErr_Occurred() with the
thread state detached in the chunking loop of `_hacl_hmac_state_update()`.

Return the raw HACL* exit code instead and convert it in `_hmac.new()` and
`_hmac.HMAC.update()` once the thread state is re-attached and the mutex
is released, as `hmac_digest_compute_locked()` already does. Release
builds keep ignoring the update() exit code.

Since `_hacl_convert_errno()` is now always called with an attached
thread state, remove the `PyGILState_Ensure()` added by gh-135561 and
assert that a thread state is attached instead.

* simplify ifdefs: always check for the hacl error

---------
(cherry picked from commit 0fba211)

Co-authored-by: Bénédikt Tran <[email protected]>
Co-authored-by: Gregory P. Smith <[email protected]>