Installation fails on Wakari: can't find Lapack about riemann

[ketch]

Maybe we should only add **lapack_args when compiling the few Riemann solvers that need it, and put those inside a try block so the whole install doesn't fail if it can't find LAPACK.

[ahmadia]

Yes, it would not be hard to modify the Riemann installer to only build the
kernels that require LAPACK when LAPACK is available through NumPy.

On Sat, Feb 1, 2014 at 6:42 AM, David Ketcheson [email protected]:

Maybe we should only add **lapack_args when compiling the few Riemann
solvers that need it, and put those inside a try block so the whole install
doesn't fail if it can't find LAPACK.

Reply to this email directly or view it on GitHubhttps://github.com//issues/62#issuecomment-33869466
.

[aterrel]

Do these old instructions not work?

Just for grins I opened up a shell and did:

$ pip install git+https://github.com/clawpack/clawpack.git

But that resulted in "cannot find -lblas"

Being more creative let's try setting up links to mkl:

[~]$ ln -s /opt/anaconda/envs/np17py27-1.5/lib/libmkl_intel_lp64.so libblas.so
[~]$ ln -s /opt/anaconda/envs/np17py27-1.5/lib/libmkl_intel_lp64.so liblapack.so
[~]$ export LD_LIBRARY_PATH=$PWD:$LD_LIBRARY_PATH
[~]$ export LIBRARY_PATH=$PWD:$LIBRARY_PATH
[~]$ pip install git+https://github.com/clawpack/clawpack.git

Then clawpack installs fine (although you might want to clean the paths up a bit).

After that opening a new notebook, import clawpack works.  

[ketch]

@aterrel Thanks for that. However, I tried those 5 lines, followed by 'pip install clawpack' and got

/usr/bin/ld: cannot find -lf77blas

Is my environment somehow different from yours?

[ketch]

By the way, I was able to install clawpack successfully by doing

pip install -U numpy
pip install clawpack

As @mandli pointed out, our setup.py relies on flags available only in numpy 1.8. I think I'll post these instructions to the docs and close this issue.

[mandli]

I am not sure if it's the version of NumPy that's the issue but your fix @ketch seems to imply that. My guess is that the upgrade somehow either finds the BLAS and LAPACK libraries correctly or finds it's own.

I think the puzzling thing here is why numpy.__config__.lapack_opt_info does not have the correct information.