Created attachment 152723 [details] devel_py-venusian.diff
This ports had been updated to 1.0 by bofh@ at the same day of the submit. Since 1.0 < 1.0a8, PORTEPOCH should be added in the commit as the PR. wen
Over to bofh@ in order to: - Fix package version going backwards - Decide about NO_ARCH as the new maintainer
(In reply to Jan Beich from comment #2) Somehow I am not sure about the phrase NO_ARCH. As far as I know that this means the port has no architecture dependent files. On the other hand all ports are by default arch independent. Then what's the point of this feature. Looking forward for some lights. And apologize for the PORTEPOCH issue.
NO_ARCH is part of long-term goal inspired by other package systems: https://wiki.freebsd.org/PortsToDo#Architecture_independent Here're a few relevant excerpts from IRC logs. #freebsd-python@Freenode: <jbeich> Any reason py-* ports aren't marked NO_ARCH en masse? <jbeich> I think, very few have binary modules which are easy to spot in plist <koobs> jbeich: we should organise a sweep <koobs> jbeich: want to add an entry to the Python wiki page mentioning a few notes on what you look for? ... <jbeich> koobs: no wiki account. Try adding: - Mark ports that install only .py files as NO_ARCH. * Investigate if there're any ARCH-specific steps in Makefiles, setup.py, etc. #bsdports@EFnet: <jbeich> some > 1G packages are actually NO_ARCH which *should* alleviate the time spent in pkg-create <marino_> jbeich: i don't know the purpose of NO_ARCH <marino_> nor why it would affect packaging? <marino_> bofh_ asked me this but I couldn't answer him <jbeich> marino_: it strips part of ABI string making it possible to install same package on i386, amd64, etc. Unfortunately, poudriere doesn't optimize builds with such packages. <jbeich> it'd be even more nice to have NO_OSREL or NO_OPSYS... <marino_> so it's faster because it only builds once instead of once per arch? <bapt> it should be faster because it should only build once <bapt> but we are not there yet <bapt> NO_ARCH is for now pretty stupid <bapt> and should be improved <marino_> okay -- in my case it sometimes doesn't build once <bapt> yes NO_OSREL and NO_OPSYS would be nice as well <bapt> but lots of work is needed to get there <marino_> if there is a lot of contention, a 1G package can take far longer than normal to package <bapt> to be able to build only once over the repos we should also work on repo layout way more and on pkg repo itself (In reply to Muhammad Moinur Rahman from comment #3) > On the other hand all ports are by default arch independent. The opposite. It's a legacy of how the first ports were for C-based projects. # currently $ pkg info -fx venusian | fgrep -i arch Architecture : freebsd:11:x86:64 # with NO_ARCH set $ pkg info -fx venusian | fgrep -i arch Architecture : freebsd:11:* which partially satisfies the following warning when invoked by poudriere: ===> Registering installation for py27-venusian-1.0 pkg-static: DEVELOPER_MODE: Notice: arch "FreeBSD:10:i386" -- no architecture specif ic files found: **** could this package use a wildcard architecture?
(In reply to Jan Beich from comment #4) Thanks for the detailed. I think we can move it to NO_ARCH. :)
A commit references this bug: Author: bofh Date: Thu May 14 09:03:12 UTC 2015 New revision: 386304 URL: https://svnweb.freebsd.org/changeset/ports/386304 Log: devel/py-venusian: Fix PORTEPOCH - Mark ARCH independent PR: 197424 Submitted by: tkato432@yahoo.com Changes: head/devel/py-venusian/Makefile