Re: Andreas Barth 2013-05-12 <[🔎] 20130512231053.GP28240@mails.so.argh.org> > Also, I think the appropriate answer would be to have the sockets be > created somewhere configurable, e.g. within /tmp (or symlink from /tmp > to build/contrib/pg_upgrade/ or whatever). Building in a symlink from /tmp/pg-$something to /build/... was my first thought, but that won't work because getcwd() is used to determine the path name (I haven't specifically tried, but I'm pretty sure this canonicalizes the path). I was looking closer at this yesterday. First I was reluctant to actually change the socket location from getcwd() to /tmp because that seems pretty instrusive given the way pg_upgrade works, but then I had the idea to only fall back to /tmp if the path is too long. I have uploaded a new version 9.3~beta1-2 with that patch and it built successfully on i386 and powerpc. (Now we can inspect those plperl segfaults on kfreebsd-* more closely.) Christoph -- cb@df7cb.de | http://www.df7cb.de/
Attachment:
signature.asc
Description: Digital signature