On 1 Aug 2014, at 19:54, Mikael Mortensen <[email protected]> wrote:

> Thank you both for the tips. I’ll look into it some more. BTW, do you create 
> your own modules for all dependencies Garth? The module system is used on our 
> cluster, but I have never tried to create modules of my own.
> 

I haven’t done it, but have seen colleagues do it for FEniCS (on workstations 
and on clusters). It’s easy.

Dorsal should create module files.

Garth


> Best regards
> 
> Mikael
> 
> On 01 Aug 2014, at 18:20, Garth N. Wells <[email protected]> wrote:
> 
>> Use the 'module' system to support multiple installations on a cluster.
>> 
>> Garth
>> 
>> On Fri, 1 Aug, 2014 at 3:36 PM, Jan Blechta <[email protected]> 
>> wrote:
>>> On Fri, 1 Aug 2014 16:16:38 +0200
>>> Mikael Mortensen <[email protected]> wrote:
>>>> Hi,
>>>> I'm trying to install master version of fenics on top of a stable
>>>> version installed by Johannes on the Abel cluster. Johannes has more
>>>> or less everything, including dolfin, installed under
>>>> /usit/abel/u1/johannr/nobackup/src/hashstack/default/, which is on my
>>>> path. The stable version here works just fine. I now want to use all
>>>> the dependencies here and build a master version, so I'm using dorsal
>>>> with unstable for fiat/ffc/ufl/instant, and install in my own folder.
>>>> I then clone dolfin in a different directory and run cmake.local. It
>>>> stops at MultiMeshAssembler.cpp because for some reason it picks up
>>>> ufc.h from the stable and not the unstable branch, even though the
>>>> unstable is sourced last. From my .bashrc:
>>>> source /usit/abel/u1/johannr/fenics-1.4.0-gcc.conf  # stable
>>>> everything
>>>> source /usit/abel/u1/mikaem/Fenics/fenics-1.4/share/fenics/fenics.conf
>>>> # unstable fiat/ffc/ufl/instant/ufc.h
>>>> With some hacking of paths I am able to fix MultiMeshAssembler and
>>>> everything compiles just fine. However, the python version still does
>>>> not work and I find that it is because the module-libraries are
>>>> picking up the stable dolfin version. Here is ldd of
>>>> dolfin/cpp/_common.so:
>>>> [mikaem@login-0-0 ~]$ ldd
>>>> /usit/abel/u1/mikaem/Fenics/dolfin/local.master/lib/python2.7/site-packages/dolfin/cpp/_common.so
>>>>    linux-vdso.so.1 =>  (0x00007fff7816d000)
>>>>    libdolfin.so.1.4 =>
>>>> /usit/abel/u1/johannr/nobackup/src/hashstack/default/lib/libdolfin.so.1.4
>>>> (0x00007fb14c5ee000)
>>>>    libpython2.7.so.1.0 =>
>>>> /usit/abel/u1/johannr/nobackup/src/hashstack/default/lib/libpython2.7.so.1.0
>>>> (0x00007fb14c1e7000)
>>>> ...
>>>> Evidently _common.so has picked up the stable libdolfin.so.1.4. Any
>>>> hints at how this can be fixed? An alternative I don't like is to
>>>> give up Johannes' dependencies all together and install everything
>>>> for myself.
>>> You should not use Johannes' fenics.conf, it probably sets
>>> LD_LIBRARY_PATH. Linker can't distinguish between stable and unstable as
>>> they have same name libdolfin.so.1.4. Rather you should tell to DOLFIN's
>>> CMake about all the dependencies which you want to use from Johannes.
>>> Or maybe try a little bit messy solution, keep the build you have and
>>> then get rid of Johannes' LD_LIBRARY_PATH in your own fenics.conf a
>>> check what linker says.
>>> Generally, it will be everytime tricky. You should convince Johannes to
>>> install dependencies you need separately from stable FEniCS. (You could
>>> simulate this with symlinks.)
>>> Jan
>>>> Thanks for any help.
>>>> Best regards
>>>> Mikael
>>> _______________________________________________
>>> 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