Johan Hake wrote:
> On Sunday 27 September 2009 20:18:07 Garth N. Wells wrote:
>> Johan Hake wrote:
>>> On Sunday 27 September 2009 19:49:09 Garth N. Wells wrote:
>>>> Where's the appropriate place in the build system to test for the
>>>> version number of external libraries? Is it in dolfin/SConscript (using
>>>> the simula-scons function 'checkVersion')?
>>> I do not think we have any defined sections in SConstruct/SConscript
>>> where this should be done. We have used checkVersion to check for the
>>> swig version some years ago... It is still there as out-commented code in
>>> the SConscript file.
>>>
>>> I think the function just compare a version string with another handed
>>> version string, where the numbers are delimited by '.'.
>>>
>>> Aren't there any version checks done in the pkg-config generator files?
>> No, but I guess that's not the place to do it since they are unrelated
>> to DOLFIN and just produce a foo.pc file.
>>
>> I guest the right way is to use 'pkg-config --version foo.pc' rather
>> than checkVersion.
>
> Again, checkVersion is just a function that check if a version string is
> larger or equal than a passed version string. So the test might look
> something
> like this:
>
> foo_version = getoutput("pkg-config --version foo.pc")
> if scons.checkVersion(foo_version,"3.0.0"):
> raise Error
>
Yes, I know how it works ;).
>
>> Still, I don't know where this check is best performed.
>
> Not sure either. I added an extensive check for the SWIG version in
> SConscript. I have never felt good about that piece of code...
>
> With the present state of the build system it is not easy to answer questions
> like this. However it would be nice to add some functionality to the
> scons.cfg
> file. Instead of passing a list of strings, stating the dependencies we pass
> a
> dict.
>
> Something like:
>
> OptDependencies = {'petsc':'3.0.0',
> 'slepc':'2.3.3':
> 'foo':None,
> ...}
>
> The checks could then be done in the simula-scons library, when the
> dependencies are checked for.
>
pkg-config does has some fancy features with max and min version
numbers, but it could get complicated.
>> My motivation for this is to require PETSc >= 3 so the that PETsc
>> wrappers can cleaned up.
>
> Yes, I just saw your blueprint :)
>
It will help clean up some of the logic in the PETSc wrappers.
Garth
> Johan
_______________________________________________
DOLFIN-dev mailing list
[email protected]
http://www.fenics.org/mailman/listinfo/dolfin-dev
