Re: BUG #3362: xlog corruption just after initdb on irix

"Herve Boulouis" <amon@sockar.homeip.net> writes:
> - trigger the bug

> /etc/init.d/pgsql start
> /etc/init.d/pgsql stop
> /etc/init.d/pgsql start
> /etc/init.d/pgsql stop
> /etc/init.d/pgsql start
> <signal 6 on postmaster startup, see logfile>

> from that point, postmaster won't start unless another initdb is done.

That's pretty dang weird.  It looks like it's failing to write the
shutdown checkpoint record (but is updating pg_control anyway).  But if
that's the explanation then it should have failed during initdb, because
each step of initdb goes through the start/stop process.

> I had already noticed that problem on this platform with 8.0 or 8.1, but
> didn't report it at the time.

You should have mentioned it earlier :-( ... whatever the problem is
is surely now long-forgotten.  Although I see we have a report of 8.1
working with IRIX 6.5, so apparently it works for some people.  Are you
sure your compiler is up-to-date?

> If necessary, I can provide access to an irix machine if some pgsql
> developpers need it to test/maintain the irix port.

I'm willing to take a look, but first please see if you can eliminate
the compiler-bug theory.  Check to see if there's a newer compiler
version available; check whether building with CFLAGS=-O0 makes the
problem go away.

            regards, tom lane