Dear all, 
This is a failure report using the hashdist method
In a HPC environment (linux, infiniband), the petsc compilation fails with
*******************************************************************************
2014/12/15 20:51:22 - INFO: [package:run_job]          UNABLE to CONFIGURE with 
GIVEN OPTIONS    (see configure.log for details):
2014/12/15 20:51:22 - INFO: [package:run_job] 
-------------------------------------------------------------------------------
2014/12/15 20:51:22 - INFO: [package:run_job] Could not find a functional BLAS. 
Run with --with-blas-lib=<lib> to indicate the library containing BLAS.
2014/12/15 20:51:22 - INFO: [package:run_job]  Or --download-fblaslapack=1 to 
have one automatically downloaded and installed
2014/12/15 20:51:22 - INFO: [package:run_job] 
*******************************************************************************


10 Ara 2014 tarihinde 18:50 saatinde, Anders Logg <[email protected]> şunları 
yazdı:

> Yes, we likely need a couple of different variants. To begin with, I think 
> the main focus should be on the kitchen sink variant to get something that is 
> as close to foolproof as possible for users.
> 
> Note that the current version does not (as promised) build from master. I 
> made a mistake in how hashdist decides which changesets to use from git. So 
> the dev version currently builds the stable predefined version. Fix in 
> progress..
> 
> --
> Anders
> 
> 
> Wed Dec 10 2014 at 5:34:42 PM skrev Martin Sandve Alnæs <[email protected]>:
> Hashdist is not like any package manager. It doesn't only build packages, but 
> also manages entire stacks of packages with particular versions of each 
> living side by side. This allows reproducible stacks if used right, as well 
> as variations of stacks without rebuilding more than the stack diffs. When 
> using anything outside of the hashdist stack, reproducibility may be lost. 
> Pragmatic reasons compel the use of e.g. system compilers, and IMO using 
> system python as well is best for most linux users.
> 
> I think we need more than one install script. In particular, the kitchen sink 
> variant which builds e.g. python and numpy etc, and minimal variants that use 
> system installed python etc. I'd like a variant that doesn't install fenics 
> so I can install master/next/wip branches myself in another directory. Of 
> course this means maintenance of stack variants for different platforms, 
> similar to the dorsal platform scripts.
> 
> Martin
> 
> 10. des. 2014 17:10 skrev "Garth N. Wells" <[email protected]>:
> 
> 
> On Wed, 10 Dec, 2014 at 11:13 AM, Anders Logg <[email protected]> wrote:
> ok, so is there a limited set of Python packages that we could agree on?
> 
> I assume numpy is already included. What about matplotlib?
> 
> A package manager should only install dependencies; user should install other 
> packages independently, i.e. like Debian/Ubuntu 'recommended' and 'suggested' 
> packages.
> 
> Garth
> 
> Wed Dec 10 2014 at 12:05:38 PM skrev Johannes Ring <[email protected]>:
> On Wed, Dec 10, 2014 at 11:40 AM, Anders Logg <[email protected]> wrote:
> > Good points.
> >
> > iPython seems like a good thing to include.
> >
> > It would be good to make the system Python libraries visible (but after
> > hashdist in the search path) but I don't know how to automatically extract
> > the correct path to the system Python libraries.
> >
> > I think Johannes knows this better than I.
> 
> If we want to use system Python modules, then I don't think we should
> build our own Python.
> 
> Johannes
> _______________________________________________
> fenics mailing list
> [email protected]
> http://fenicsproject.org/mailman/listinfo/fenics
> _______________________________________________
> fenics mailing list
> [email protected]
> http://fenicsproject.org/mailman/listinfo/fenics

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

Reply via email to