This reverts commit 751532a5.

See https://github.com/cython/cython/issues/2529

Exclude a C++17 test (currently in PR #3294) on macOS since it suffers from the same incompatibility issues as the other C++17 tests.

Closes https://github.com/cython/cython/issues/3938

With Cython syntax, we ignore the exception_check when a Python object is returned.
In pure Python mode, with a Python object type as return type, we reject it.
Instead of raising an error, we now just reset exception_check to False.

Found in https://github.com/cython/cython/issues/2529

Closes https://github.com/cython/cython/issues/2304

* Warn when cyfunction.__[kw]defaults__ is changed

Cython can't make use of these new defaults so it seems sensible
to alert the user (even if we're choosing to allow the assignment
for compatibility reasons).

I know the warning mechanism is somewhat heavy. I think this is
OK because it's unlikely that users will be repeatedly assigning
to these attributes in a time-critical loop.

Doesn't fix https://github.com/cython/cython/issues/2650, but hopefully alerts users to it.

CPython added PyMem_RawCalloc and PyMem_Calloc in version 3.5.

The #defines of PyMem_RawMalloc, PyMem_RawRealloc, and PyMem_RawFree
are removed from Cython/Utility/ModuleSetupCode.c since the Raw versions allow
use without the GIL but the non-Raw versions do not.
Co-authored-by: William Schwartz <wkschwartz@gmail.com>

Closes https://github.com/cython/cython/pull/3050
Closes https://github.com/cython/cython/pull/3047

Closes https://github.com/cython/cython/issues/2760

It requires Py_DecodeLocale which appears in 3.5. This is causing
it to fail on Windows. It's somehow passing on Linux for reasons
that I don't understand (but it really shouldn't be)

Specifically this disallows memoryviews, but it extends to any future type that Cython has to generate manual reference counting code for.

Closes https://github.com/cython/cython/issues/3085

Fixes https://github.com/cython/cython/issues/4354

As suggested in https://github.com/cython/cython/pull/4471#issuecomment-979489947
I don't think that we're getting any benefit from running the pypy
known bugs test as part of the CI (and it's causing minor problems).

Fixes an issue which Pyston triggers.
See https://github.com/cython/cython/issues/4200

Exception messages had changed a little

Let a tracing test run slower to prevent div-by-zero crashes in line_profiler due to measured zero runtime.

Python 3.11 hid _PyLong_FormatAdvancedWriter and
_PyFloat_FormatAdvancedWriter. I've disabled USE_UNICODE_WRITER
for the moment but I suspect we want to find replacements in the longer term.

To remove a compiler warning since this slot was added in 3.11a2.

Fixes an issue which Pyston triggers.
See https://github.com/cython/cython/issues/4200

To silence a compiler warning since it was added in Python 3.11a2.

For 0.29.x branch only.

With reference to https://github.com/cython/cython/issues/4365#issuecomment-977023011
in Python 3 PyCode_NewEmpty does everything we need to set
exception traceback code objects.

This change is probably only a real improvement on Py3.11 onwards
(where the replacement for PyCode_new is currently pretty slow).
On earlier version it'll probably be fairly similar (maybe one extra allocation for the cline case?)

"Fixes" a few bugs that are just caused by very small implementation
details, recategorize some others (e.g. those that cimport array
are not expected to ever work)

Follow up to https://github.com/cython/cython/pull/4453 (01d323ab).
That PR creates a temp that's only used in the limited API
(and therefore causes warnings on all other paths)

Now that they've implemented the PyContextVar C API it segfaults instead of just failing.

With reference to https://github.com/cython/cython/issues/4365#issuecomment-977023011
in Python 3 PyCode_NewEmpty does everything we need to set
exception traceback code objects.

This change is probably only a real improvement on Py3.11 onwards
(where the replacement for PyCode_new is currently pretty slow).
On earlier version it'll probably be fairly similar (maybe one extra allocation for the cline case?)