BLD: py2.7 failing on perf hash changes (for windows)

[jreback]

xref #13436

so this is failing on windows py2.7 (3.5 works fine).

pandas\algos.c(113380) : warning C4244: '=' : conversion from 'int' to '__pyx_t_5numpy_float32_t', possible loss of data
pandas\algos.c(114441) : warning C4244: '=' : conversion from '__pyx_t_5numpy_int64_t' to '__pyx_t_5numpy_float64_t', possible loss of data
pandas\algos.c(114538) : warning C4244: '=' : conversion from '__pyx_t_5numpy_int64_t' to '__pyx_t_5numpy_float64_t', possible loss of data
pandas\algos.c(114647) : warning C4244: '=' : conversion from '__pyx_t_5numpy_int64_t' to '__pyx_t_5numpy_float64_t', possible loss of data
pandas\algos.c(114744) : warning C4244: '=' : conversion from '__pyx_t_5numpy_int64_t' to '__pyx_t_5numpy_float64_t', possible loss of data
pandas\algos.c(116387) : warning C4244: '=' : conversion from 'npy_double' to '__pyx_t_5numpy_float32_t', possible loss of data
pandas\algos.c(118081) : warning C4244: '=' : conversion from 'npy_double' to '__pyx_t_5numpy_float32_t', possible loss of data
pandas\algos.c(119587) : warning C4244: '=' : conversion from 'npy_double' to '__pyx_t_5numpy_float32_t', possible loss of data
pandas\algos.c(119609) : warning C4244: '/=' : conversion from 'double' to '__pyx_t_5numpy_float32_t', possible loss of data
pandas\algos.c(121324) : warning C4244: '=' : conversion from 'npy_double' to '__pyx_t_5numpy_float32_t', possible loss of data
pandas\algos.c(124112) : warning C4244: '=' : conversion from 'npy_double' to '__pyx_t_5numpy_float32_t', possible loss of data
pandas\algos.c(126309) : warning C4244: '=' : conversion from 'npy_double' to '__pyx_t_5numpy_float32_t', possible loss of data
pandas\algos.c(128960) : warning C4244: '=' : conversion from 'npy_double' to '__pyx_t_5numpy_float32_t', possible loss of data
pandas\algos.c(131798) : warning C4244: '=' : conversion from 'npy_double' to '__pyx_t_5numpy_float32_t', possible loss of data
pandas\algos.c(133366) : warning C4244: 'function' : conversion from 'Py_ssize_t' to 'int', possible loss of data
pandas\algos.c(134168) : warning C4244: '=' : conversion from 'long' to 'char', possible loss of data
pandas\algos.c(139210) : warning C4244: '=' : conversion from 'Py_ssize_t' to 'int', possible loss of data
pandas\algos.c(163983) : warning C4244: 'function' : conversion from 'Py_ssize_t' to 'int', possible loss of data
pandas\algos.c(164199) : warning C4244: 'function' : conversion from 'npy_int64' to 'int', possible loss of data
error: command 'C:\\Program Files (x86)\\Microsoft Visual Studio 9.0\\VC\\BIN\\amd64\\cl.exe' failed with exit status 2

would be ok with just having this change affect py3 instead

[jreback]

cc @chris-b1

[chris-b1]

I at least got a more helpful error message locally, I can take a look later. It might be preferable to keep this in py2 too, since I think it's a little faster than the py2 hash, and for consistency.

pandas/src/klib\khash_python.h(16) : error C2054: expected '(' to follow 'inline'
pandas/src/klib\khash_python.h(16) : error C2085: 'asint64' : not in formal parameter list
pandas/src/klib\khash_python.h(16) : error C2143: syntax error : missing ';' before '{'

[jreback]

yes I think that Py_ssize_t is 32-bits on windows is the issue. But I think for simplicity just make this for py3, then should be fine.

[chris-b1]

Older MSVC is apparently weird about inline, it looks like all that needs to be done is replace it with the PANDAS_INLINE macro, which expands to __inline