CLN: fix build warning in c_timestamp.pyx

[ghost]

pandas/_libs/tslibs/c_timestamp.c: In function ‘__pyx_f_6pandas_5_libs_6tslibs_11c_timestamp_10_Timestamp__get_start_end_field’:
pandas/_libs/tslibs/c_timestamp.c:7257:25: warning: ‘__pyx_t_10’ may be used uninitialized in this function [-Wmaybe-uninitialized]
 7257 |   __pyx_t_5numpy_int8_t __pyx_t_10[1];
      |  

Retrying fix dropped from #27371

a review comment there disliked reintroducing ndarray here because the file had been converted to using memoryviews. get_start_end_field returns a view of type npy_bool soout should therefore be either a memoryview or an ndarray, and bool memoryviews will only be supported in cython 0.30
(see cython/cython/pull/2676).

[WillAyd]

Haven't closely followed that conversation or checked links so apologies if stupid question but does a unit8 memory view not work in the meantime? IIRC we've used that in the past instead

[ghost]

No, that's a reasonable question. But no, that doesn't work either because on cython<0.30 bool arrays require an explicit cast in the cdef even for assigning to a ndarray[npy_bool], while memory views just don't work. If you try a memory view (bool or int8_t) what you get is

ValueError: Does not understand character buffer dtype format string ('?')

The linked cython PR fixes this in the next release.

[jbrockmendel]

It looks like the linked cython issue makes it unnecessary to specify cast=True in some ndarray declarations. We use that in a few places, so once we update to require cython>=0.30.0, feel free to take point on removing that no-longer-necessary usage

[ghost]

ok. There's also the conditional gil support in 0.30 which is needed. Should probably update soon after release.