BUG: Fix M_PI

[larsoner]

Might fix https://ci.appveyor.com/project/scipy/scipy/builds/32959985 and related failures:

  scipy\stats\_stats.c(19781): error C2065: 'M_PI': undeclared identifier

Seems like this must be an underlying Cython issue (?), but maybe this will allow us to work around it.

[WarrenWeckesser]

The definition of M_PI in math.h is not standard. Perhaps what we really need is to define _USE_MATH_DEFINES when building with a Microsoft compiler: https://docs.microsoft.com/en-us/cpp/c-runtime-library/math-constants?view=vs-2019

[larsoner]

It seems like something that Cython should take care of when you use libc.math, no?

In any case, it seems to have succeeded building with this:

https://ci.appveyor.com/project/scipy/scipy/builds/32961266

Might be worth merging this as-is to get things back to green, then if someone has a more principled solution, that can undo the change here (if desired, it seems fairly harmless).

[larsoner]

The build passed but it's not functional, trying to fix the shim...

[larsoner]

Actually it looks like elsewhere in the codebase we solve this in various different ways. It seems like the most expedient here would be just to use PI from numpy.math, which is already cimported. Let's see if that one passes.

[larsoner]

Seems to have fixed AppVeyor https://ci.appveyor.com/project/scipy/scipy/builds/32962123

[WarrenWeckesser]

Even simpler than the first version!

[WarrenWeckesser]

Any idea why there is no Travis-CI run?

[larsoner]

Could be just some slow status update issue (or maybe someone restarted it?), but it's there now and green

[WarrenWeckesser]

Yes, I guess it is just very slow getting started. It wasn't even in the list when I made my comment

[WarrenWeckesser]

Merged. Thanks, @larsoner.