Re: 9.3.9 and pg_multixact corruption

On Fri, Sep 25, 2015 at 3:41 AM, Andreas Seltenreich
<andreas.seltenreich@credativ.de> wrote:
> I think the intention was to make configure complain if there's a -O > 2
> in CFLAGS.

-1 on that idea.  I really don't think that we should categorically
decide we don't support higher optimization levels.  If the compiler
has a bug, then the compiler manufacturer should fix it, and it's not
our fault.  If the compiler doesn't have a bug and our stuff is
blowing up, then we have a bug and should fix it.  I suppose there
could be some grey area but hopefully not too much.

> OTOH, a unit test for multixact.c that exercises the code including
> wraparounds sounds like a desirable thing regardless of the fact that it
> could have caught this miscompilation earlier than 6 months into
> production.

Definitely.

-- 
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company