dolfin team mailing list archive

["Garth N. Wells" <gnw20@xxxxxxxxx>]

On 20 January 2013 18:33, Kent-Andre Mardal <kent-and@xxxxxxxxx> wrote:
>
>
> On 19 January 2013 23:48, Johan Hake <hake.dev@xxxxxxxxx> wrote:
>>
>> On 01/19/2013 10:46 PM, Garth N. Wells wrote:
>> > I've got all of DOLFIN and the unit tests working nicely with OSX 10.8
>> > / Macports / clang, except the JIT unit test
>> >
>> >     test_compile_extension_module
>> >
>> > I have no idea how to debug it. The seg fault is triggered by the line
>> >
>> >     VecExp(*x);
>> >
>> > My best guess is that the pkg-config linking is bad. Any other ideas
>> > or suggestions on how to fix it?
>>
>> Do you have several libpetsc installed?
>>
>> > Can we finally get rid of the pkg-config garbage and use CMake for the
>> > jit linking?
>>
>> It might be a combination of pkg-config and distutils. In the latter
>> library_dirs and libraries are two different arguments to the Extension
>> class. This means that we split the library dirs and libraries from each
>> other destroying all information about library_dir to library order.
>> This is actually preserved in the pkg-config file.
>>
>> If we are going to shift to CMake we need not only ditch pkg-config, but
>> also distutils. Kent has added support for CMake compilation for VTK and
>> ITK based modules, which more easily is find using find_package calls.
>>
>> I can have a look at this next week. I think using CMake instead of
>> distutils also opens up for easier compiler configurations, as this can
>> be quite obscure in distutils.
>
>
>
> I agree that it is time to use cmake instead of pkg-config.
>
> Dolfin needs to provide a Dolfin.cmake file. The rest is simple, from
> Instant's point of view.
>

DOLFIN provides a CMake file already.

I did some more testing, and the OSX bug appears only with static
PETSc libs, which lends weight to it being a linking issue.

Macports builds static PETSc libs, so it would be good to have the
tests running perfectly for this case.

Garth

> Kent
>