On Mon, Oct 7, 2013 at 8:52 PM, Björn Grüning <bjoern.gruening@pharmazie.uni-freiburg.de> wrote:
Hi Dave, Bjeorn,
Right now NumPy isn't installing at all on the Test Tool Shed.
Dave - is is possible that the system was updated and the Fortran compiler removed? e.g.
...
Since this seems to be breaking lapack, perhaps we should further simplify package_numpy_1_7 not to use lapack or ATLAS? What do you think Bjoern?
I'm a little bit sceptic to remove features only because the Test Utility can not handle it. On the other hand I understand that numpy should just work and that your test failing since month because of my dependencies. I will get rid of it tomorrow. As a result we can setup_virtualenv and install numpy from pip directly I think.
Is that ok for you? Cheers, Bjoern
Hi Bjoern, I think the real problem here is we don't yet have a clear base installation: https://trello.com/c/gYlRdqwr/774-toolshed-define-a-base-set-of-utilities-th... https://trello.com/c/7VTlX9rD/844-toolshed-tool-dependency-installation-requ... What is worse is the system used by the ToolShed for running the nightly tests appears to change without notice, e.g. here it looks like something with the Fortran compiler has changed, for the BLAST+ wrappers the default shell seems to have changed between bash and sh). I think the ToolShed system should be a role model of the agreed standard set of system level dependencies. Given that, it seems your lapack package has some dependencies which are not necessarily present, and this failure is cascading up to break NumPy and then Biopython, and thus multiple tools on the Tool Shed :( I do think that in the short term the best option is to further simplify package_numpy_1_7 not to use lapack or ATLAS - although this is not ideal :( On the positive side, your hard work setting up packages for NumPy, lapack and Altas is drawing attention to the need for a clearer set of base packages. :) Thanks, Peter