Re: [gentoo-alt] numpy-1.3.0 problems with freebsd-7.2

[Fabian Groffen]

On 25-07-2009 19:30:22 -0400, Michael Yang wrote:
> Hi guys,
>
> I'm using freebsd-7.2 Release/i386.  Bootstrapped successfully
> (gcc/gfortran built with USE="fortran") and I ran into problems
> emerging dev-python/numpy-1.3.0:
>
> $ emerge numpy-1.3.0
> ...
> build/src.freebsd-7.2-RELEASE-i386-2.6/numpy/core/src/umathmodule.c
> /usr/include/fenv.h: Assembler messages:
> /usr/include/fenv.h:156: Error: suffix or operands invalid for `fnstsw'
> ----------------------------------------------------------------------------------------------------------
> I think what's happening is that gcc is somehow pulling in
> /usr/include/fenv.h into the ebuild process.  I can successfully build
> directly from the var/tmp/.../work directory:
>
> $ python setup.py build
>
> build/temp.freebsd-7.2-RELEASE-i386-2.6/build/src.freebsd-7.2-RELEASE-i386-2.6/numpy/core/src/umathmodule.o
> -Lbuild/temp.freebsd-7.2-RELEASE-i386-2.6 -lnpymath -lm -lpython2.6 -o
> build/lib.freebsd-7.2-RELEASE-i386-2.6/numpy/core/umath.so
> ...
>
> So I don't know where in the ebuild the /usr/include/fenv.h is getting
> sucked in instead of telling python to use the
> numpy/core/include/numpy/fenv/fenv.h (which is newer)
> This may be related to the numpy issue on
> http://projects.scipy.org/numpy/ticket/811.

either your manual build uses the host compiler/linker, or it includes
an -Ipath/to/var/tmp/portage/numpy/work/include

> A workaround is to copy the numpy/core/include/numpy/fenv/fenv.h to
> /usr/include/ as root but that's obviously cheating :)

That's for most of us not even an option, let alone about the possible
damage that can do to your system.

We need to know what setup.py is doing exactly to find out what's going
on here.


-- 
Fabian Groffen
Gentoo on a different level