[blfs-dev] BLFS status

Ken Moffat zarniwhoop at ntlworld.com
Wed Sep 11 13:08:15 PDT 2013


On Wed, Sep 11, 2013 at 01:15:59PM -0500, Bruce Dubbs wrote:
> The question now is how to handle a BLFS-7.4 release.  We can get the 
> current 4 (or 5) packages incorporated in a day or two, but new versions 
> of packages will keep coming.  I can tag a 7.4 version, make a few 
> editorial updates and call it BLFS-7.4.  That can be done pretty much 
> any time.
> 
> Or we can do a -rc for a week with a package freeze.
> 
> Or call for comments now, continue updating, and do a tag in a week or so.
> 
> Or something else.
> 
> What's the best way to go?
> 
>    -- Bruce

 I suggest kicking it out as soon as possible, with the current
version of ImageMagick.  I'm sure there are lots of odd little issues
lurking in the corners, but I don't see how to get people to test it
enough to find them.

 But I'm stalled on postgresql, and not from the new version.

 I had overlooked part of r11169, namely a one-line sed to put the
socket in /run/postgresql.  That is good (debian's reason for not
using /tmp is that it prevents ordinary users spoofing postgresql),
but it breaks the tests - either /run/postgresql doesn't exist, or
else it has been created with mode 755 so that an ordinary user
running the tests cannot write there (first action is to write a
lock file), and root cannot run the tests.

 Oddly, my normal user can 'mkdir /run/postgresql' and run the
tests, but if a correct /run/postgresql exists (e.g. if an older
version is already installed) the expected "Permission denied"
results.

 I see all manner of things in fedora and debian, and I was hopeful
that one of the debian patches to add EXTRA_REGRESS_OPTS to some of
the Makefiles (--host=/tmp and perhaps --port=57532 which is the
default for testing) would fix it, but it still dies with:

LOG:  could not translate host name "/tmp", service "57532" to
address: Name or service not known
WARNING:  could not create listen socket for "/tmp"
FATAL:  could not create any TCP/IP sockets

 so obviously it's trying to treat --host as a machine name :
supposedly --host can be a machine or a directory in postgres.
Setting it to localhost proves that, it goes back to trying to write
to /run/postgresql/ and PGHOST PG_HOST both seem to be ignored.

 I've also noticed that we don't build any of the contrib/ directory
which means that we don't build or install the pg_upgrade program.
I can live with that part (I've already upgraded 9.2 the long way),
but perhaps it is worth mentioning it somewhere.

ĸen
-- 
das eine Mal als Tragödie, dieses Mal als Farce



More information about the blfs-dev mailing list