I think is is great. 

I haven’t tested yet, but a suggestion to make the process simpler is to let 
PETSc build suitesparse, etc. PETSc is a C library but can be installed with 
pip (it has a Python-based build system). It can take care of a number of 
dependencies (solvers, graph partitioners, etc).

I’ve copied Andy Terrel at Conitnuum Analytics who might have something to chip 
in with.

Garth 


> On 5 Jan 2015, at 13:07, Juan Luis Cano <[email protected]> wrote:
> 
> Hello all,
> 
> My name is Juan Luis Cano, I'm studying a MSc in Aerospace Engineering in 
> Madrid and I started recently to play with FEniCS for my final degree 
> project. For my day to day work I am using a virtualized Linux Mint and 
> everything works like a charm thanks to the Ubuntu PPA, but as it is not the 
> distribution which I normally use I tried to build a conda package these 
> holidays.
> 
> I noticed there are a couple of build systems out there (dorsal, hashdist) 
> but, as the Anaconda distribution[1] is getting popular in the scientific 
> Python world these days, I really wanted to try to provide FEniCS packages 
> for it (at least in Linux). For those who don't know it, Anaconda's package 
> manager, conda, is open source[2] and provides a nice build system[3].
> 
> You can try out my progress so far with a Linux 64 bit box and a Python 2.7 
> environment:
> 
> $ conda create --name py27 python=2.7
> $ source activate py27
> (py27)$ conda install fenics --channel juanlu001
> 
> The build process itself was painful because I knew very little about FEniCS 
> dependencies a week ago but right now I managed to run the `demo_poisson.py` 
> (_without_ plotting, see below). The results seem OK from Paraview.
> 
> The good thing is that I made the builds in an Ubuntu Server box but it works 
> the same in an Arch Linux machine too. I didn't try to compile it against 
> PETSc, Trilinos and such yet because I wanted some feedback from the 
> community first, and know if this is something useful for anybody!
> 
> The trick here was avoiding the Ubuntu packages (via apt-get) and compile the 
> dependencies in the form of conda packages themselves. I did such with boost 
> and suitesparse, for instance[4]. This way there are no linking problems 
> across different Linux distros. I am stuck with VTK though because it seems 
> to look for libGL.so, which in turn pulls from X11... and everythings gets 
> messy very quickly[5].
> 
> So if I can get some feedback about how does this work in others' computers, 
> if this is any useful and which packages should I try to build next that 
> would be great. Anybody can reproduce the build process using my 
> conda-recipes fork.
> 
> Kind regards and happy new year!
> 
> Juan Luis
> 
> [1] https://store.continuum.io/cshop/anaconda
> [2] https://github.com/conda/
> [3] http://conda.pydata.org/docs/build.html
> [4] https://binstar.org/juanlu001/
> [5] 
> https://github.com/Juanlu001/conda-recipes/commit/a18cedc56e330ba09961b8ddaeb86f580e22f3cc
> _______________________________________________
> fenics-support mailing list
> [email protected]
> http://fenicsproject.org/mailman/listinfo/fenics-support

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

Reply via email to