Mainline is now regression and documentation fixes only

[Richard Guenther]

On Wed, 23 Jan 2008, Peter Bergner wrote:

> On Wed, 2008-01-23 at 17:19 +0100, Richard Guenther wrote:
> > The backend changes should be acked by a backend maintainer, the 
> > middle-end changes look non-intrusive enough that they are obvious if
> > the rest is considered ok.  I would say an ABI fix like this would
> > qualify as a P1 bug
> 
> When I created the bugzilla:
> 
>   http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34814
> 
> I set the severity to blocker, but was under the impression that the RM
> would set the priority.  Was I supposed to set that too?

We use the priority only for categorizing regressions - a priority
on a non-regression doesn't have a meaning currently.

> > - how is its regression state?  DFP support was
> > added for 4.2, right?  Does this make 4.3 incompatible with 4.2?
> > If so it should be documented in the changes document at least,
> > possibly configurable with a target switch?
> > 
> > So in the end target maintainers should weight the risk of late
> > (but important in their view) fixes like this.
> 
> And there's the rub.  This has been broken from day one on GCC, so
> technically, it is not a regression (IIRC, DFP support was added in
> early GCC 4.3, Janis and/or Ben would know for sure).  However, it has
> worked on XLC, so our obj files are incompatible with XLC compiled
> objects.  XLC is also dependent on DFP routines in libgcc and libdfp
> which are compiled with GCC, so it's somewhat broken too until this
> gets fixed.

If DFP support went in in 4.3 we should fix the ABI definitely, but
I at least see the modes appearing in 4.2 which is why I ask for
a decision from the target maintainers here.  (And yeah, I cross-checked
that we didn't ship DFP support backported to 4.1 with SLES, so from
my personal point of view the change is ok - but that should not matter
here).

Richard.