On 2020/11/07 13:26, Jeremie Courreges-Anglas wrote: > On Fri, Nov 06 2020, Daniel Dickman <didick...@gmail.com> wrote: > > This diff sets NO_TEST=Yes for cython under python2 so numpy can be > > updated. > > > > If we don't go this route, we'd have to convert all the cython consumers > > to python3 so cython can be made python3-only. That means updating > > consumers like py-sphinx. So going down that path doesn't look very > > simple. > > As a policy I think that the TEST_DEPENDS of python2-only ports > shouldn't block updates of other ports to python3-only versions.
Agreed. We are still some way off being able to retire a py2 version of numpy though. > > ok on the simpler approach below so we can move forward? > > Makes sense to me, no need to bump REVISION though. I'm not sure the > comment helps as the .if block looks self-explanatory to me. +1 to all of that. Here's what I make the list of victims of removing py-numpy would be. Some of the ports involved aren't that important but others definitely are. devel/boost - (lots) games/angrydd games/fretsonfire games/singularity graphics/opencv - graphics/digikam - graphics/nomacs graphics/piglit lang/cython lang/pcbasic math/lpsolve/python print/printrun x11/py-gtk2 - audio/solfege - comms/chirp - devel/dissy - devel/py-hachoir-metadata - games/childsplay - geo/gmapcatcher - graphics/comix - graphics/gimp/stable - graphics/gimp/deskew - graphics/gimp/lensfun - graphics/gimp/liquid-rescale - graphics/gimp/resynthesizer - graphics/xsane,gimp - graphics/gnofract4d - graphics/ufraw - graphics/zbar - net/nmap,-zenmap - net/pidgin - net/microblog-purple - net/pidgin-sipe - net/pidgin-window-merge - security/pidgin-otr - x11/pidgin-guifications - x11/pidgin-libnotify - print/pdfshuffler - print/py-poppler - x11/compiz/ccsm - x11/driconf - x11/gammapage - x11/obmenu I don't think anyone will disagree with removing the dependency on py2 numpy where it's not actually useful. For opencv, digikam and nomacs don't appear to need python parts at all, let alone py2, I don't think dropping py2 from there is likely to cause much upset. For boost, I don't see anything with a listed wantlib on any type of the boost-numpy libs. Worth a bulk (on amd64; i386 misses too much these days to really be a good test) to make sure but dropping that should be possible too I think. For py-gtk2, IMHO gimp is an absolute show stopper. And for some of the others while not exactly critical software, there's no alternative. For this reason I think that we need a py2-numpy for now, and if we're doing that anyway then it makes sense to use it in cython TEST_DEPENDS too.
