On 2020/01/10 16:04, Paul Irofti wrote: > Hi, > > Here is a new port of the popular convex optimization toolbox for > Python. > > I need this for my daily research, but unfortunately it depends on a > newer numpy version. I am currently running with the one from this > thread without any issues (yes, with clang): > > https://marc.info/?l=openbsd-ports&m=157267780518610&w=2 > > Can we please update it with that diff? It seems that the talk around > that update were actually unrelated to the actual change and all the > issues were fixed in the meantime.
I don't want to stand in the way of updating NumPy but it's quite far from being ready. The diff forces use of flang, but the flang port is only for aarch64/amd64. Large parts of the tree indirectly depend on numpy (mostly via boost/opencv/openexr). daniel@ also said "numpy's fortran compiler can't be switched in isolation. We'd have to do a switch across all the forward and reverse dependencies in the tree at the same time." I haven't tested that but I do think his suggestion of having runtime tests done (as well as a bulk build) is justified. Since it needs to work with gfortran for !flang arches anyway, we probably might as well continue with gfortran everywhere to avoid yet another variation between arches.
