On Thu, 16 Jan 2014 13:13:34 +0100
Jan Blechta <[email protected]> wrote:

> On Thu, 16 Jan 2014 10:08:54 +0100
> Johannes Ring <[email protected]> wrote:
> 
> > On Thu, Jan 16, 2014 at 10:06 AM, Johan Hake <[email protected]>
> > wrote:
> > > I am not sure this is the way we want to do it. get_swig_binary()
> > > relies on check_and_set_swig_binary() has been called previously.
> > > The latter can cause problems on clusters. The logic for the
> > > generated CMake file is that it should contain all information
> > > needed to compile the extension module and that it should get that
> > > from including the DOLFIN_USE_FILE. In that file we do not store
> > > any SWIG information. Instead we just call:
> > >
> > > find_package(SWIG REQUIRED)
> > > include(${SWIG_USE_FILE})
> > >
> > > essentially re-configuring SWIG. We cannot avoid that, as we need
> > > to load the swig specific cmake macros from the SWIG_USE_FILE, but
> > > we could for example set SWIG_EXECUTABLE in the DOLFINConfig.cmake
> > > file, which is generated during configuration of DOLFIN?
> > 
> > Yes, that sounds like a better fix.
> 
> Now I see that there may be two (DOLFIN+UFC+...)s with different
> __swigversion__ which generate same code (with same hash) but in a
> cache there is stored one with swigversion which was triggered
> earlier. So the second one fails on
> ffc/jitcompiler.py:check_swig_version(compiled_module)
> 
> So bassically, it's unsafe now to have FEniCSes with different SWIG on
> one system. This should be taken into account when fixing this.

Possible workaround is to define environmentals

export INSTANT_CACHE_DIR=$HOME/.instant/cache-swigX
export INSTANT_ERROR_DIR=$HOME/.instant/error-swigX

for each FEniCS installation and given swigX. Then instant/paths.py
will use different directories for JITing. Is it sufficient? Does all
the cache usage goes through instant/paths.py?

Jan

> 
> Jan
> 
> > 
> > Johannes
> > _______________________________________________
> > fenics-support mailing list
> > [email protected]
> > http://fenicsproject.org/mailman/listinfo/fenics-support
> 
> _______________________________________________
> fenics-support mailing list
> [email protected]
> http://fenicsproject.org/mailman/listinfo/fenics-support

_______________________________________________
fenics-support mailing list
[email protected]
http://fenicsproject.org/mailman/listinfo/fenics-support

Reply via email to