Date: Tue, 21 Sep 2010 16:23:21 -0400 (EDT) From: Benjamin Kaduk <kaduk@MIT.EDU> To: Jan Henrik Sylvester <me@janh.de> Cc: afs-list freebsd <freebsd-afs@freebsd.org> Subject: Re: 1.5.77 port of openafs Message-ID: <alpine.GSO.1.10.1009211559470.9337@multics.mit.edu> In-Reply-To: <4C990A1A.7040407@janh.de> References: <4C990A1A.7040407@janh.de>
next in thread | previous in thread | raw e-mail | index | archive | help
On Tue, 21 Sep 2010, Jan Henrik Sylvester wrote: > I wanted to give 1.5.77 a try after seeing some FreeBSD related improvements > in the changelog. > > Is http://web.mit.edu/freebsd/openafs/openafs.shar the working version of > something that might end up in ports or do you have another one in progress? > (I assume that the git version is not mend for ports.) Oops, I started to make an updated sharball but didn't get around to testing it. I think it was at http://web.mit.edu/freebsd/openafs/tmp/openafs.shar (off the top of my head), but it wouldn't have the plist or package fixes that you describe below. > > Since http://web.mit.edu/freebsd/openafs/openafs.shar is still at 1.5.75, I > started updating it. > > To get it to compile, I removed two patches that do not apply to 1.5.77 > anymore, after updating the version and distinfo. > Those patches (or something equivalent to them) have been merged into upstream. > Verifying the pkg-plist, I realized that 'pkg_create -b' does not produce a > clean package (giving the same files as the port that can cleanly be > removed): Besides some files missing in pkg-plist, lib/openafs is eventually > emtpy and missing from the package and ThisCell.sample contains the sample > line twice when installed from the package. Moreover, whether afsd.fuse is > installed or not depends on the presence of the fuse-libs port. > > I have attached an updated version of the port that fixes issues I found > related to packaging. I tried to make the diff to the old version > minimalistic by not doing anything like sorting the pkg-plist. > > Does it look ok? > Well, I'm still at work, so I can't test or look too closely, but it seems pretty reasonable. (I think that the et_lex.lex.l patch is not needed anymore, but don't remember the details of what fix made it where offhand.) Thanks for putting the work in to do the updates and testing! > I hope I verified all correctly: > - 'pkg_create -b' produces a package that installs the same files as the > port. > - The port or package can be removed cleanly WITH_FUSE or without, with > modified configuration files present or without. > - The binaries seem to link to the same files in a clean environment and in a > polluted one (1155 packages installed, including fuse-libs). > Excellent. Have you also followed the steps here? http://www.freebsd.org/doc/en/books/porters-handbook/porting-testing.html (The porter's handbook is a pretty nice resource for ports tips, though it sadly doesn't give much guidance for how to deal with kernel modules.) > All testing was on 8.1-RELEASE. > > As for functionality: I did a little bit of copying from and to AFS and using > basic fs and pts commands. I did not experience any problems. I have yet to > test the issues that I had with 1.5.75 (copying files larger than the cache > size, unmounting the AFS, stopping the afsd, ...). Okay. As I noted in the tmp/openafs.shar README, I think this code has survived a buildworld in AFS, but deadlocks on a parallel make. (Actually most of the cycles in the past couple weeks that I would have put into openafs were devoted to fixing FreeBSD's ar(1) to not die when presented with a file owned by a machine principal in AFS, but I think I have a working patch for that.) > > Possible improvements before it goes to ports: Nicer error for emtpy > /usr/src/ or /usr/obj/, creation of /afs, creation of a sample for > /usr/local/etc/cacheinfo, sorting pkg-plist. Before I start anything in that > direction: > > Are you planing to bring something to ports anytime soon -- maybe as > net/openafs-devel -- or is it still too early? I hadn't thought too hard about when to bring something into ports, but I do think the code is reaching a point where it is actually useful to non-developers. In my dream world, we would figure out the locking issues that cause parallel make to deadlock in time for a 1.6 release, and that would go into ports, but I think that fixing up your packaging of 1.5.77 and making it net/openafs-devel would be a reasonable thing to do. Thanks again for moving this forward, Ben
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?alpine.GSO.1.10.1009211559470.9337>