dolfin team mailing list archive

[Johannes Ring <johannr@xxxxxxxxx>]

On Thu, Nov 17, 2011 at 4:47 PM, Anders Logg <logg@xxxxxxxxx> wrote:
> On Thu, Nov 17, 2011 at 02:31:51PM +0000, Garth N. Wells wrote:
>> On 17 November 2011 13:49, Andre Massing <massing@xxxxxxxxx> wrote:
>> > -----BEGIN PGP SIGNED MESSAGE-----
>> > Hash: SHA1
>> >
>> > On 11/17/2011 02:17 PM, Anders Logg wrote:
>> >> Looks like someone forgot to run the tests before pushing to
>> >> 1.0.x.
>> >>
>>
>> Test were run.
>>
>> >> I just wanted to point this out since it wasn't me this time. :-P
>> >
>> > Actually, the change in the GlobalAssemble call in EpetraMatrix is the
>> > culprit, it says:
>> >
>> > [ 61%] Building CXX object
>> > dolfin/CMakeFiles/dolfin.dir/la/EpetraFactory.cpp.o
>> > /home/buildbot/buildbots/main/fenics-src/dolfin/dolfin/la/EpetraMatrix.cpp:
>> > In member function 'virtual void
>> > dolfin::EpetraMatrix::apply(std::string)':
>> > /home/buildbot/buildbots/main/fenics-src/dolfin/dolfin/la/EpetraMatrix.cpp:404:
>> > error: no matching function for call to
>> > 'Epetra_FECrsMatrix::GlobalAssemble(bool, Epetra_CombineMode)'
>> > /home/buildbot/buildbots/main/local/include/trilinos/Epetra_FECrsMatrix.h:527:
>> > note: candidates are: int Epetra_FECrsMatrix::GlobalAssemble(bool)
>> > /home/buildbot/buildbots/main/local/include/trilinos/Epetra_FECrsMatrix.h:554:
>> > note:                 int Epetra_FECrsMatrix::GlobalAssemble(const
>> > Epetra_Map&, const Epetra_Map&, bool)
>> >
>> > Is there an older trilinos version installed on the buildbot (which
>> > one and which version do we require at least?)
>> >
>>
>> The Ubuntu Trilinos package is old (10.4). It is a modified  version,
>> because the 10.4 source fails to compile with gcc 4.6.
>>
>> There is a nasty change in the Epetra interface - the old code still
>> compiled, but thanks to an interface change the code did something
>> different.
>>
>> I would prefer to support just the latest (10.8). Is there demand for
>> 10.4? I wouldn't advocate using any of the Ubuntu/Debian linear
>> algebra packages (Trilinos/PETSc/BLAS) for performance.
>
> Supporting only 10.8 sounds fine to me. Less worries for us and those
> using Trilinos will probably know how to build it.

The Debian packages will then be built without Trilinos support
(unless someone package 10.8 for Debian).

Johannes