SuperBuild: SciPy for Windows

Greetings,

It seems that SciPy does not get appended in the list of projects when building on Windows. Are there any plans to include it?.. or am I wrong? I believe the logic is found here: https://gitlab.kitware.com/paraview/paraview-superbuild/blob/022257566f11c97522ed507db191800972e44e7f/CMakeLists.txt#L100

Thank you,

Patrick

1 Like

We don’t have logic which supports building scipy on Windows right now (mainly due to Fortran compiler availability on the build machine(s)). With the Python3 binaries, support will be provided by expecting Python3 to be provided externally where pip should work as it does in normal Python distributions.

2 Likes

It looks like this plan changed somewhere along the line (as of 5.8)? Is there a reasonable way to add scipy to the python that Paraview’s using on Windows?

Yes, external Python is a hairy endeavor (essentially, it is very confusing for users because there’s no simple way to just double click on the app anymore). There’s been no movement on scipy on Windows so far, sorry.

Is there a way to hack a solution for this? I suppose I could probably do it using $PYTHONPATH, but I would also be fine using the built-in pip (if it’s built-in), or something like that.

Would it make sense to include a Fortran dependency for the MS-MPI version, since that’s already requiring users to install special software?

Not right now, sorry. We avoid shipping OpenSSL (due to getting chased about updating it whenever a CVE comes out), so pip isn’t really useful since the shipped Python can’t talk with PyPI. You might be able to install a stack of wheels directly though?

The Fortran dependency doesn’t really affect the binaries (as I think we can redistribute those libraries). I don’t think we can redistribute the MSMPI libraries, though it is now on Github…). I suspect even if the licensing allows us, there are some registry and/or environment shenanigans which need done.

Have I misunderstood something? I thought that the reason that SciPy wasn’t included was because of a reluctance to package the Fortran libraries or include a dependency on an external compiler installation (which made sense to me). Is the problem really that it is hard to install gfortran onto the build machines and package scipy in directly?

It’s something we haven’t investigated much, honestly. We can look at it for 5.10 with the gitlab-ci migration.

Cc: @utkarsh.ayachit