In win7 x64 debug mode with thread_debug=1, every PyThread_release_lock() is accompanied with a message on stderr:

Could not PyThread_release_lock(<id>) error: 0

Looks okay, but I wonder whether the implementation of PyMUTEX_UNLOCK should return something other than zero? What is it meant to return?

In all PyMUTEX_* and most PyCOND_* functions, all the underlying APIs are void. So there's no point for the functions to return anything other than as convention.

A comment at the start of condvar.h says: "APIs generally return 0 on success and non-zero on error".
In Python\ceval_gil.h:57, there are boilerplace checker macros for them
(which will never fire).

So I guess this is done for uniformity.
Regardless, PyMUTEX_UNLOCK and PyMUTEX_LOCK at least are called both with and without checking the return value.


So I would drop the pretense and make these fns void.

---

That said, the reported error still persists after the fix.
The code confusing about which value LeaveNonRecursiveMutex should return on success. Pushed another commit, confirmed to work now.

> So I would drop the pretense and make these fns void.

Scratch that. In POSIX, they are #define'd as pthread fns which do return an error code.

So, nothing more to do here.

> So, nothing more to do here.

In case that was cryptic, it means: the PR can be acted upon, no other changes are needed.

New changeset 05e922136a3286893bd489a8f2ecfa0dba4da368 by Steve Dower (native-api) in branch 'master':
bpo-33316: PyThread_release_lock always fails (GH-6541)
https://github.com/python/cpython/commit/05e922136a3286893bd489a8f2ecfa0dba4da368

New changeset c851dfc99b28c7335d42abd8250bb77c11ce23c0 by Miss Islington (bot) in branch '3.7':
bpo-33316: PyThread_release_lock always fails (GH-6541)
https://github.com/python/cpython/commit/c851dfc99b28c7335d42abd8250bb77c11ce23c0