Thread

Re: pgsql: Remove "fmgr.h" include in cube contrib --- caused crash on a Ge

Bruce Momjian <bruce@momjian.us> — 2011-09-02T15:20:50Z

Jeremy Drake wrote:
> On Thu, 1 Sep 2011, Bruce Momjian wrote:
> 
> > Jeremy Drake wrote:
> >
> > > I am the owner of both mongoose and okapi.  Let me know if there's
> > > anything you want me to try.
> >
> > Thanks.  I would either like to email you patches to test or get ssh
> > access so I can compile it myself.
> 
> You can send me patches if you want, but I spent a little time with it
> tonight and it seems to be the change to src/include/access/xlog.h:
> http://git.postgresql.org/gitweb/?p=postgresql.git;a=blobdiff;f=src/include/access/xlog.h;h=1fd60fb98d7362b677112517a20a41e32227a34f;hp=cdbf63fa76e0e7b154c084191d0df6138e1cbfcc;hb=4bd7333;hpb=d010391ac8f706e17998671534ca1230f68d2f38
> 
> Unfortunately, I also had to revert commit
> 6416a82a62db4e66b2edb0fa8fc83a580c3f1931 to fix compile errors.  I expect
> you would be able to do something a little more surgical than that...

Wow, that is interesting.  So the problem is the inclusion of
replication/walsender.h in xlog.h.  Hard to see how that could cause the
cube regression tests to fail, but of course, it is.

I noticed you are using these compile flags:

	'CFLAGS' => '-O3 -xN -parallel -ip',
	'CC' => 'icc'

Can you test it with lower optimizations?

I looked at the contrib/cube compile messages and didn't see anything
unusual. 

The only other idea I have is to try the attached patch which changes
the offsetof() call to mention a struct field name, and not the first
element of the field.  However, I see other uses of accessing the
element of a struct field, so I might be wrong here.

I will say that our buildfarm is great at giving developers information
to diagnose the cause of failures!  It just isn't helping me find the
cause in this particular case.

-- 
  Bruce Momjian  <bruce@momjian.us>        http://momjian.us
  EnterpriseDB                             http://enterprisedb.com

  + It's impossible for everything to be true. +