Re: Beyond GCC 3.0: Summing Up

[Mark Mitchell <mark at codesourcery dot com>]

>
> [1] Some regressions we do not even care about.  So what if romp
>     doesn't build?

David asked about this on the SC list, too, and I think it's
an important question.  If we, as a community, decide to desupport
a particular platform, that's fine.  In fact, the less we do
the more successful we will be at what we do do.  I wouldn't mind
dropping support for a lot of old targets.  In fact, in one fit
of madness, I suggested desupporting all debugging formats except
DWARF2.

However, that decision should be made independently.  If a patch breaks 
ROMP, we should ask if we still want to support it.  If
we don't, then we don't care, and that failure is not a regression;
the spec has changed.  Ideally, we would decide what platforms,
if any, to obsolete at the *start* of a major release cycle,
and then make the remainder work well.  It seems a bit odd to
decide in the middle of things to desupport a target because some
patch breaks it, but seems otherwise good.

It's also not reasonable to discuss every change that causes
breakage in depth.  This flamewar has taken hours of everyone's
time over the last few days, and we can't afford that.  There
has to be a process by which we can reach a decision quickly
and efficiently.  Often, a good way to do that is to have a default
that requires a manual override; this is Per's notion of a
"rebuttable presumption".  I think that's a very good idea;
if a patch breaks something (but not by exposing a latent bug;
see earlier compromise!) and nobody is trying to fix it after
notice is given, then it goes -- unless you can persuade the
SC (or some other panel?) that the benefits somehow outweigh the costs.