[Numpy-discussion] C99 math funcs and windows

Pauli Virtanen pav@iki...
Wed Aug 13 15:52:16 CDT 2008


Hi,

Wed, 13 Aug 2008 13:06:47 -0500, David Cournapeau wrote:
> Many tests (14) are failing on windows because of the recent changes in
> some maths functions (C99 conformance). I believe the problem is caused
> by mingw compiler (which is stuck at 3.4.5), but I still have to check
> that. To solve this, I see three solutions:
>  - disabling the C99 tests on windows. Easy 
>  - using a more recent
>  version of mingw: mingw does not officially
> support gcc 4, but we can built it ourselves. Not too difficult (I have
> some makefiles to build a toolchain from scratch on linux/mac os x), but
> I don't like the idea of using our own compiler.
>  - we don't care, tests fail on windows.
> 
> Do some people have a strong opinion on that ? Disabling sounds like the
> way to go for 1.2.0, but at some point, we will have to deal with this
> C99 problem on windows if people add more C99 code in it.

I went ahead and backed out the TestC99 testcases in r5644.

They don't currently really serve any purpose as they are not guaranteed 
to succeed on all platforms, and we'll definitely want all tests to pass. 
I suggest we put them back in only after we actually have some guarantees 
about inf/nan handling.

-- 
Pauli Virtanen



More information about the Numpy-discussion mailing list