Ok, I realise I'm probably one of the only people that's got clustering software installed on the same machine as my music player daemon, but there is good reason for it... 5:) I'm currently testing out a patch for johntheripper that allows john to crack passwords on a cluster, or (more aptly) on a multiprocessor system. During the course of trying to get this to work, I got a bit of a shock when mpirun tried to start reading my mpd.conf. It turns out both systems use the same config file. It's probably too rare a problem to be worth fixing, but I thought I'd file it anyway just in case. If you'd like to mark this as WONTFIX, that's fine by me. 5:)
Actually, this is slightly worse than I first thought. It turns out they both write to /usr/bin/mpd, so installing one after the other will break the first one...
FEATURES="collision-protect" for t3h win! BTW, isn't it mandatory for devs? :P
We can't really change the mpich daemon configuration file location, that's just too nonstandard and would break stuff. It wouldn't surprise me if the same were true for mpd. Maybe a blocker's the way to go.
Jakub: you can have collision-protect zillion times in your FEATURES, if you're not actually using both packages - and how many people use both mpich2 and mpd (as Mike already mentioned) ? As for mpd.conf and /usr/bin/mpd files installed by media-sound/mpd, I really do not want to change them from upstream's designated names. I imagine this is the same for sys-cluster/mpich2 maintainer, so I'm afraid blocker is the only way to go. As a side note, Debian call their mpich2's mpd "mpich-mpd-bin".
Yep, not a problem. Blocker sounds like the best way to go (don't forget mpd-svn). Thanks... 5:)
OK, added a block to both mpd an mpd-svn. Donnie added the block to mpich2, so we're good to close this bug. Thanks everybody.
*** Bug 179477 has been marked as a duplicate of this bug. ***
*** Bug 342851 has been marked as a duplicate of this bug. ***
One way of partly solving this is to make mpd optional: Index: mpich2-1.2.1_p1-r1.ebuild =================================================================== RCS file: /var/cvsroot/gentoo-x86/sys-cluster/mpich2/mpich2-1.2.1_p1-r1.ebuild,v retrieving revision 1.8 diff -u -B -r1.8 mpich2-1.2.1_p1-r1.ebuild --- mpich2-1.2.1_p1-r1.ebuild 27 Oct 2010 01:16:51 -0000 1.8 +++ mpich2-1.2.1_p1-r1.ebuild 27 Oct 2010 01:44:38 -0000 @@ -15,7 +15,7 @@ LICENSE="as-is" SLOT="0" KEYWORDS="amd64 ~hppa ppc ppc64 x86" -IUSE="+cxx debug doc fortran threads romio mpi-threads" +IUSE="+cxx debug doc fortran mpd threads romio mpi-threads" COMMON_DEPEND="dev-libs/libaio romio? ( net-fs/nfs-utils )" @@ -25,7 +25,7 @@ sys-devel/libtool" RDEPEND="${COMMON_DEPEND} - !media-sound/mpd + mpd? ( !media-sound/mpd ) !sys-cluster/openmpi !sys-cluster/lam-mpi" @@ -111,7 +111,7 @@ c="${c} --sysconfdir=/etc/${PN}" econf ${c} ${romio_conf} \ --docdir=/usr/share/doc/${PF} \ - --with-pm=mpd:hydra \ + --with-pm=$(use_enable mpd mpd:)hydra \ --disable-mpe \ $(use_enable romio) \ $(use_enable cxx) \ Needless to say, the rest of the ebuild would need to be revised too, especially src_test() which relies heavily on mpd and would need to be rewritten for hydra (if that works for the test suite at all). Also, media-sound/mpd would need to block sys-cluster/mpich2[mpd].
fine with me, just change it in the mpd ebuilds as soon as you get the ok from cluster
(In reply to comment #9) > One way of partly solving this is to make mpd optional: mpd will be leaving the default configuration of mpich2 in the next release.
(In reply to comment #11) > mpd will be leaving the default configuration of mpich2 in the next release. There is an mpich2-1.3 ebuild in science overlay, could this improvement be integrated there? (or in the current upstream version 1.3.2p1?)
(In reply to comment #12) > There is an mpich2-1.3 ebuild in science overlay, could this improvement be > integrated there? (or in the current upstream version 1.3.2p1?) That's the plan, yes. Last time I worked on it, tests were failing to pass.
(In reply to comment #9) > One way of partly solving this is to make mpd optional: > @@ -111,7 +111,7 @@ > c="${c} --sysconfdir=/etc/${PN}" > econf ${c} ${romio_conf} \ > --docdir=/usr/share/doc/${PF} \ > - --with-pm=mpd:hydra \ > + --with-pm=$(use_enable mpd mpd:)hydra \ While reading the Installation PDF docs for mpich2 I suspect that the currently we force mpd to be default on Gentoo, while upstream docs say that since mpich2-1.1 the default is the hydra (if I remember right). Therefore, I think when they state in their docs that this enables all (well, except smpd) the first is the default: --with-pm=hydra:mpd:gforker I propose to be sure you swap the order like --with-pm=hydra:$(use_enable mpd mpd:) \ but have to fix the trailing colon when -mpd. :(
MPD has been deprecated in MPICH2. So, the Gentoo ebuilds should consider dropping it as a configure option as well. Apart from the conflict with the MPD music player, it would also remove the python dependencies.
+*mpich2-1.4_rc2 (04 May 2011) + + 04 May 2011; Justin Bronder <jsbronder@gentoo.org> +mpich2-1.4_rc2.ebuild: + Version bump (#362655). Use system hwloc. Use hydra instead of mpd for pm + (#145367). Disable more tests as recommended by upstream.
cvs/gentoo-x86/media-sound/mpd $ ebuildvar RDEPEND|grep mpi mpd-0.15.8.ebuild : RDEPEND="!sys-cluster/mpich2 mpd-0.15.15.ebuild : RDEPEND="!sys-cluster/mpich2 mpd-0.15.16.ebuild : RDEPEND="!sys-cluster/mpich2 mpd-0.16.2.ebuild : RDEPEND="!sys-cluster/mpich2 These need to be changed to !<sys-cluster/mpich2-1.4_rc2
MPICH2 no longer includes MPD (process manager) by default, so there is no conflict with the MPD music player.
(In reply to comment #18) > MPICH2 no longer includes MPD (process manager) by default, so there is no > conflict with the MPD music player. Yes, and? angelos@/sound@ herd still need to do their bit so we can have sys-cluster/mpich2 and media-sound/mpd installed at the same time, so it's down to a Gentoo thing now.
Changed the blockers accordingly