dorsal team mailing list archive

[Harish Narayanan <harish.mlists@xxxxxxxxx>]

On 8/19/10 1:26 AM, Garth N. Wells wrote:
> On Wed, 2010-08-18 at 20:51 +0200, Anders Logg wrote:
>> On Wed, Aug 18, 2010 at 06:58:33PM +0200, Anders Logg wrote:
>>> I'm now getting to the point where DOLFIN is building, but then at the
>>> end I get this:
>>>
>>> Implicit dependency '/Users/log/Work/FEniCS/lib/libpetsc.a' not found,
>>> needed by target 'dolfin/libdolfin.dylib'.
>>>
>>> Any ideas what goes wrong? The file in question seems to be a symlink
>>> that points to itself (with an extra slash)...
>>
>> I see more strange things.
>>
>> If I look in the directory ~/Work/FEniCS/lib (default Dorsal
>> installation location for libraries), there's a bunch of symlinks that
>> point to themselves:
>>
>>   libamd.a, libslepc.a, libpetsc.a, libumfpack.a
>>
>> Any idea why this is so?
>>
>> Is it a Mac thing or is it something with the PETSc/SLEPc
>> configuration?
>>
>> Here's how it's currently done:
>>
>> 1. When PETSc is configured, it knows PETSC_DIR=pwd and it knows
>> PETSC_ARCH=<some random darwin-blabla string>.
>>
>> 2. When SLEPc is configured, it knows PETSC_DIR=INSTALL_PATH but it
>> doesn't know PETSC_ARCH.
>>
> 
> Somewhat related, you could try the PETSc and SLEPc packages from
> MacPorts. The down side is that the MacPorts version is not compiled
> with UMFPACK, so there is not decent LU solver via PETSc. I imagine it
> would be easy to add though since UMFPACK is in MacPorts.

I am not so sure this is a good idea for the same reason dorsal doesn't
tend to use petsc/slepc packages for the other platforms: these packages
aren't always compiled with optimal options.

The dorsal petsc/slepc packages should be corrected.

Harish