I think the patch should be upstreamed and then we can update the vendored version of qhull (perhaps keeping a similar approach as #11246 in mind).

Looks like upstream isn't active, but I've raised an issue about the patch. I'm a little confused as to why qhull itself builds but the matplotlib extension won't build.

@s-sajid-ali I was suggesting you put the patch in your spak recipe, I strongly suspect that the spak package for qhull is already carrying such a patch.

I am skeptical of merging this for the same reason as @anntzer , but leave the final decision to @ianthomas23 .

Okay, I misunderstood that part. I'll see if the issue I opened upstream gets any response and if that fails see if it can be fixed within spack.

Actually looking at this again the issue appears to arise because you are compiling poly2.c using icpc (intel c++ compiler) as opposed to icc (intel c compiler). On a bland linux install this file is correctly compiled by gcc, not g++. (And indeed, assigning the result of malloc without a cast is valid C, not valid C++.)

Thus I think the fix is actually on spack's side (or whoever set up the compilation machinery to use icpc/icc).

I agree with @anntzer , this should be fixed on the spack side (either to fix the compiler selection, to find the system qhull, or to carry the patch).