fenics team mailing list archive

[Johan Hake <hake.dev@xxxxxxxxx>]

On 02/02/2013 11:33 PM, Garth N. Wells wrote:
> On 2 February 2013 22:30, Johan Hake <hake.dev@xxxxxxxxx> wrote:
>> Yes, I reported it, but I cannot either find the message...
>>
>> Anyhow it is related to the one of the tests in the
>> ufl-jit-assemble-chain test suit, which uses boost_math. In the
>> distutils system we always linked to boost_math and the problem was
>> avoided.
>>
>> We can fix it by:
>>
>> 1) Add boost_math when configuring ufc and also add it to the
>> UFCConfig.cmake file.
>>
>> 2) Add boost_math as an additional lib dependency when JIT compiling UFC
>> modules. This will add back runtime configuration to JII compiling.
>>
>> 3) Expose a number of configuration options like libs, lib_dirs aso to
>> the cmake build system and let the user provide the whereabout of an
>> external library.
>>
>> There are pros and cons with all of these. The last one looks most
>> attractive to me, but requires the most work. 1 also seams attractive as
>> the bessel function is exposed in UFL and hence should be available when
>> generating UFC forms.
>>
> 
> I'm about to push a 'fix' - use Boost math special functions instead
> of boost::math::tr1. The former is header only, hence no linking
> issues.

Sounds very good.

> With the recent change, is there somewhere that the flags, libraries,
> etc, that are being used by the JIT compiler can be inspected?

Put a message("${FOO}") statement in the UFCConfig.file which get read
by the instant JIT compiler and then have a look in the config.log file
in the cache directory.

Johan

> 
> Garth
> 
>> Johan
>>
>> On 02/02/2013 10:51 PM, Garth N. Wells wrote:
>>> I have some trouble with the  ufl-jit-assemble-chain in DOLFIN related
>>> to Boost math libraries. I get
>>>
>>> ......93b919307320.so: undefined symbol: boost_cyl_bessel_j;
>>>
>>> I recall somebody reporting this already, but I can't find the message.
>>>
>>> Garth
>>>
>>> On 1 February 2013 11:46, Johan Hake <hake.dev@xxxxxxxxx> wrote:
>>>> Hello!
>>>>
>>>> Recently there have been some annoyance with distutils and how we
>>>> runtime check where swig is. Kent, Johannes and I have now added a CMake
>>>> based system. Both UFC and DOLFIN use that now to JIT compile extension
>>>> modules. It uses what ever compiler defined in the UseFOO file. The
>>>> default configuration system of instant is still distutils, retaining
>>>> all backward compatibility.
>>>>
>>>> The configuability of instant using cmake is reduced compared to
>>>> distutils, as it just reads in UseFOO files. It can easily be extended
>>>> to be more general, like the present distutils system.
>>>>
>>>> Please report any issues that most probably crops up.
>>>>
>>>> Johan
>>>>
>>>> _______________________________________________
>>>> Mailing list: https://launchpad.net/~fenics
>>>> Post to     : fenics@xxxxxxxxxxxxxxxxxxx
>>>> Unsubscribe : https://launchpad.net/~fenics
>>>> More help   : https://help.launchpad.net/ListHelp
>>