The state of DragonFly and pkgsrc 2008Q4

[Matthew Dillon]

:Assuming these building issues are fixed, it'd take about a week to build
:pkgsrc-2008Q4 and then another period of time (another week?) for it to
:propagate out to mirrors.  So, it seems like it would work.
:
:However, the build happening now, assuming there isn't significant
:breakage, would work well as an upgrade from the existing set of binaries.
: I'd hate to hold up the release for what may be a relatively
:inconsequential update to a number of packages.
:
:A better way to say this: re-entrant functions aside, if the bulk build
:running now finishes with lots of working packages, let's go with that.

    Ok, I mentioned this on IRC but I should post it here too.  The 2.2
    release will occur closer to the end of January, not mid-january.  This
    will help with the package building pressure but the main reason is that I
    am way behind on my 'hammer fsck' work, and I really want that code
    to be in the release so I can put a 'production ready' stamp on HAMMER.

    The new directory hash for HAMMER has been canceled.  It turns out that
    the hash isn't the cause of the performance issues it was supposed to
    fix.  Non-correlation between inode numbers and directory search ordering
    (whether sorted or not) is the problem.  I have some idea in that
    regard but nothing that will make it into the release.

    Amd64 and netmp will not be ready for the 2.2 release (unless Aggelos
    thinks differently re: netmp) but really excellent progress is being
    made on both.  Amd64 is now in our master repo and netmp is now a
    branch in our master repo.  Amd64 now boots through init and /bin/sh
    but has issues with signal state and a few other things that cause
    user programs to seg-fault fairly quickly.  Netmp is in even better
    shape but still has some gotchas and it may be safer to go through
    another release cycle with netmp integrated into the master branch
    for wider testing before getting it into a release.

					-Matt
					Matthew Dillon 
					<dillon at backplane.com>