This is the mail archive of the
binutils@sources.redhat.com
mailing list for the binutils project.
Re: [PATCH] ia64 unwind directive semantics
- From: James E Wilson <wilson at specifixinc dot com>
- To: Jan Beulich <JBeulich at novell dot com>
- Cc: binutils at sources dot redhat dot com
- Date: Fri, 28 Jan 2005 17:49:03 -0800
- Subject: Re: [PATCH] ia64 unwind directive semantics
- References: <s1fa08eb.028@emea1-mh.id2.novell.com>
On Fri, 2005-01-28 at 01:42, Jan Beulich wrote:
> Jim wilson wrote:
> >Hopefully gcc emits unwind info that passes all of these tests.
> So do I hope. But if not, they're at fault...
I already noticed one. The gcc -pg support is broken, it both emits
instructions without unwind info, and does a section switch. This will
need to be fixed. It won't work anymore after this patch goes in. This
is GCC PR 12455, and Steve Ellcey is looking at the problem because gcc
-pg doesn't work right under HPUX.
> Actually, if you use expr_build_dot at the point or .proc, things may
> get entirely wrong if the primary entry point doesn't immediately
> follow.
Yes, either way, we need to make assumptions, and it isn't clear that
either set of assumptions is worse than the other. I'm willing to go
along with your patch, but if I find a problem I will switch back to the
old way.
In theory, everyone should be emitting the function label immediately
after the .proc, and giving the function name as the first argument to
the .proc, in which case it doesn't matter which way we do this. Either
way we get the same answer, so in theory this change should be harmless.
This patch is OK.
--
Jim Wilson, GNU Tools Support, http://www.SpecifixInc.com