MySQL on -current (possible gcc issue?)

[Jeremy Chadwick]

I've been trying to figure this one out for a few weeks now, maybe
someone can shed some light on the problem.  This may or may not be
cross-posted to freebsd-threads (if someone wants to forward it over
there -- I'm not on the list) -- I'm starting to wonder what the
_true_ source of the problem is (compiler bug?  I don't know for sure).

Here's an initial comment about the problem, dated back in March:

http://lists.freebsd.org/pipermail/freebsd-threads/2004-March/001791.html

The issue is that MySQL (4.0.x, 4.1.x, and 5.x) all sig11 on machines
which lack a defined CPUTYPE in make.conf (which translates over to a
missing -march flag to gcc during compile-time).  I've confirmed this
on two -current machines, and one 5.2.1-RELEASE machine, all with
different hardware and different configurations.  The 5.2.1-RELEASE
test I did solely to see if it was due to the latest gcc -- doesn't seem
to be.  Also, WITH_PROC_SCOPE_PTH makes no difference either.

Details about the sig11: mysqld will sig11 on a TCP socket connect to
the server itself.  UNIX domain socket access does not cause the crash.
Defining CPUTYPE (in my case, CPUTYPE=p4) fixes this problem and things
work as they should.

Can someone give me some pointers as to how I can assist in debugging
what exactly is going on?  Why would using -march fix this?  Is this a
scheduler problem per chance (a shot in the dark, I know...)?

Thanks.

-- 
| Jeremy Chadwick                                 jdc at parodius.com |
| Parodius Networking                        http://www.parodius.com/ |
| UNIX Systems Administrator                   Mountain View, CA, USA |
| Making life hard for others since 1977.                             |