Re: GCC 3.3, GCC 3.4

[Joe Buck <jbuck at synopsys dot com>]

On Fri, Jan 31, 2003 at 01:40:56AM +0100, Gerald Pfeifer wrote:
> Agreed.  What worries me more, though, is that currently we have
> ~59 high priority PRs = regressions for the 3.2-branch (most of
> which also apply to 3.3).

69 against 3.2, 47 of these also againt 3.3 (meaning that 22 of the
69 are fixed in 3.3).
 
> I believe we should try to tackle these first (and deliver a
> sensibly improved GCC 3.2.3 after the forthcoming 3.2.2).

Well, maybe.  But ...

If 3.3 can be stabilized to the point where almost all the high-priority
bugs are fixed in it, then 3.3 (with its other improvements) would
be preferable to 3.2.3 (which may have bug fixes, but doesn't have 3.3's
new features).  This is especially the case for any bugs that depend
on new infrastructure to fix.

As the third digit increases, we should be more and more cautious about
what kinds of changes we apply.  If we wind up backporting large patches,
it might be safer to just postpone fixes until 3.3.

So, if stabilization work got the 69/47 number above down to, say,
50/15, then we have the choice of either releasing 3.2.3 with 50
regressions, or 3.3 with 15 regressions but more likelihood of having
some new failure.

On the other hand, if most fixes are easily done in both branches, so
the number looks more like 20/15, then 3.2.3 would be a good idea.

Still, I'm worried about whether we really have the resources to have
three flows going at once on a continuing basis.  As long as we do,
I think that the middle one (3.3) will suffer: lots of testing and
fixing on 3.2, lots of gurus working on the cool stuff in 3.4, but
little effort on 3.3.  Then I'll probably have to issue another
"what's the point" message again. :-)


 
> >> There's more I'd like to say, but I can't.  Hopefully soon.
> > This continues to annoy me, but I understand why you do it. The longer
> > you say this line, the less I believe there's anything to say.
> 
> Given how many ABI bugs CodeSourcery developers have already fixed
> (and the fact that I cannot remember Mark ever promising something
> he hasn't kept), I'm much more optimistic.
> 
> BTW, if you consider my finding of C ABI breakage for several, not so
> minor platforms today, apparently we could also use a C ABI testsuite.
> 
> > I think the longer gcc, as a project, goes on without an autobuild
> > continuous regression checker, the worse off things will get. Somebody
> > [...] will have to step up with the bandwidth, time, and machine to do
> > this.
> 
> Fully agreed.
> 
> Well, why don't we send a "Call for support" to the gcc-announce list?
> Seriously.
> 
> Gerald
> -- 
> Gerald "Jerry"   pfeifer@dbai.tuwien.ac.at   http://www.pfeifer.com/gerald/

-- 
Q. What's more of a headache than a bug in a compiler.
A. Bugs in six compilers.  -- Mark Johnson