Re: Philipp Kern 2013-05-12 <[🔎] 20130512195219.GA621@hub.kern.lc> > It does seem to me that running it with a socket in $TMPDIR instead of the > current directory would solve the problem, though. Is there no env var > for this? Also during regular operation it looks like this would be not > insane to do if it invokes temporary daemons. During regular operation, these are the standard unix sockets for the database clusters in /var/run/postgresql/. > The current directory design is due to the following: > > * /build is a separate persistent partition. > * /build/$pkg-$ver-$arch-$random is the base dir to host the source files and > the unpacked source > * /build/$pkg-$ver-$arch-$random/$pkg-$ver is the result of the dpkg unpack. > > Hence you can just remove /build/$pkg-$ver-$arch-$random and be guaranteed > to have no remainder on disk. Yeah I understand that part, I'd just like "$pkg-$ver-$arch-$random" to be shorter... I will try to see if I can easily make this regression test use a different socket location. On first sight, it looked like changing the buildd part of this would be the cleaner solution. Btw, if Roger's "this will trigger random build failures" argument is a valid one, the buildd base dir should really be > 107 chars so things *do* fail. Christoph -- cb@df7cb.de | http://www.df7cb.de/
Attachment:
signature.asc
Description: Digital signature