octave-maintainers
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [fem-fenics] Build system


From: Eugenio Gianniti
Subject: Re: [fem-fenics] Build system
Date: Sun, 9 Mar 2014 15:53:32 +0000


On 09 Mar 2014, at 01:50, Marco Vassallo <address@hidden> wrote:




On Sat, Mar 8, 2014 at 1:07 PM, Eugenio Gianniti <address@hidden> wrote:

On 06 Mar 2014, at 17:30, Marco Vassallo <address@hidden> wrote:




On Thu, Mar 6, 2014 at 4:09 PM, Eugenio Gianniti <address@hidden> wrote:

First of all I’d like to know if the openmpi issue is known and, in case, I would kindly ask for a reference to where it is addressed.

Try to look at this post, it could fix the problem [1].


[1] http://octave.1599824.n4.nabble.com/fem-fenics-0-0-1-released-td4662182.html#a4662234
 

I managed to install the package locally *without* the need to set the environment variable proposed in [1], via the attached pkg-config-patch.diff. After the installation both the already cited variable and the one mentioned in your previous reply still need to be set to run the examples due to the just in time compilation, and plot is to be avoided as reported in [2]. However I obtain the results shown on the Octave wiki fem-fenics page and on the FEniCS project documentation, if I save and visualise them using Paraview. Please tell me if I should submit the diff through the bug or patch tracker, or if the attachment is enough.

I would like to discuss a couple of points to further improve the build system of the package. The first one is that currently the configure script, if not finding the dependencies, reports a quite misleading message telling that they weren’t found and the compilation will just proceed with missing functionality. The source code is not meant for such a thing to happen, namely it unconditionally includes dolfin.h whether it was found or not. Since fem-fenics is a wrapper for FEniCS, I would say that the configure script should be modified to fail when dolfin is not available.

It was supposed to be like this, but I think that JuanPablo inadvertently changed it. It is really easy to fix it [1].

Ok, I attach a patch in order to do so.

Eugenio

Attachment: configure-failure.diff
Description: configure-failure.diff



Marco


  
Another aspect is that the configure script explicitly looks for the Eigen headers, whilst the package sources never explicitly include them. When using pkg-config these recursive dependencies are handled by the utility itself, so in my opinion the Eigen checks can be removed.

As a last consideration, even if the package now installs without setting manually the compilation flags, as already said they are still necessary for the just in time compilation, so probably a PKG_ADD with proper instructions should be provided.

If you agree with my points I am willing to work on them and provide further patches to address the issues these days.

Eugenio





reply via email to

[Prev in Thread] Current Thread [Next in Thread]