Date: Sat, 1 Oct 2022 11:50:17 +0200 From: Matthias Andree <mandree@FreeBSD.org> To: d@delphij.net, ports-committers@FreeBSD.org, dev-commits-ports-all@FreeBSD.org, dev-commits-ports-main@FreeBSD.org Subject: Re: git: 525e857368c8 - main - sysutils/e2fsprogs: revert bogus qsort_r() patch. Message-ID: <e026e87b-6d27-ae3b-aa7a-1058272391d8@FreeBSD.org> In-Reply-To: <eaa1a01a-f52b-70c6-e506-d7efb7345464@delphij.net> References: <202210010713.2917DPgq077535@gitrepo.freebsd.org> <eaa1a01a-f52b-70c6-e506-d7efb7345464@delphij.net>
next in thread | previous in thread | raw e-mail | index | archive | help
Am 01.10.22 um 10:30 schrieb Xin Li: > First of all, thanks for connecting me with the upstream developer. Thanks for going in so much detail with him. > On 10/1/22 00:13, Matthias Andree wrote: >> The branch main has been updated by mandree: >> >> URL: >> https://cgit.FreeBSD.org/ports/commit/?id=525e857368c8c2de355ca00b0c35008be6ee8a3c >> >> commit 525e857368c8c2de355ca00b0c35008be6ee8a3c >> Author: Matthias Andree <mandree@FreeBSD.org> >> AuthorDate: 2022-10-01 07:10:08 +0000 >> Commit: Matthias Andree <mandree@FreeBSD.org> >> CommitDate: 2022-10-01 07:13:09 +0000 >> >> sysutils/e2fsprogs: revert bogus qsort_r() patch. > > Which part of it was bogus? Xin, Applying it was, in spite of the discussion and better approaches in the pipeline -- and the patch that got committed here also is not the one that got into sort_r. Look, it's not as if FreeBSD 14-CURRENT were datacenter production ready, and -CURRENT isn't Tier #1 for ports either. It does not matter if it breaks for a few days. > > It was a working fix, and the goal was to ensure that the upcoming > 2022Q4 branch will be working with -CURRENT, by the way. Well, we can MFH the fix any time. > >> delphij@ and the upstream maintainer were working on a >> autoconf-based >> solution, and this patch was not approved. Remove it and mark >> port BROKEN on systems that changed qsort_r() for the GNU-like API. > > I think I have made it pretty clear in our discussion with upstream > maintainer that the current versions of e2fsprogs are NOT really using > qsort_r at all, and it is always using sort_r_simple. > > sort_r_simple is platform independent, there is no compelling to leave > the port broken just because the sort_r author decided it was a good > idea to redefine qsort_r (the maintainer also accepted my proposal to > change qsort_r to (qsort_r) earlier this week: > https://github.com/noporpoise/sort_r/blob/master/sort_r.h . > > So based on the reasons above, my change shouldn't even affect the > resulting binary. Then the proper fix is to remove the unused ballast if it is getting in the way. (Having to parenthesize a function name is an abomination I have not seen in 32 years of dealing with C and C++ code.) > > And it's my believe that -CURRENT users deserves better treatment than > marking a "trunk" port broken like this, just because a better > solution is being worked on by the upstream. The ports patching > mechanism exists for a reason, and we can do better than this. I am compiling 14-CURRENT from Git as we speak so I have a test bed. But this takes a while, even with a 8-core 16-thread CPU. > > This has unnecessarily broke all ports that uses sysutils/e2fsprogs as > MASTERDIR, which includes misc/e2fsprogs-libuuid, which do not > necessarily use sort_r.h, please fix or revert. > Oops. I will fix that first. Regards, Matthias
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?e026e87b-6d27-ae3b-aa7a-1058272391d8>