GCC 4.2.0 Status Report (2007-02-19)

[Mark Mitchell <mark at codesourcery dot com>]

I've spent some time today looking at GCC 4.2.

I've heard various comments about whether or not it's worth doing a 4.2
release at all.  For example:

[Option 1] Instead of 4.2, we should backport some functionality from
4.2 to the 4.1 branch, and call that 4.2.

[Option 2] Instead of 4.2, we should skip 4.2, stabilize 4.3, and call
that 4.2.

[Option 3] Like (2), but create (the new) 4.2 branch before merging the
dataflow code.

One of the key points behind these suggestions is that Red Hat and
Novell plan to skip to 4.3 for their next releases, so we'll have a hard
time getting volunteers for stabilization of 4.2.0.  Another comment is
that the aliasing fixes on the 4.2 branch mean that 4.2's performance on
SPEC is likely to be inferior to that of the 4.1 releases.  (I'd like to
see a 4.1.2 vs. 4.2.0 comparison for SPEC so that we can evaluate that
more accurately.)

I've had a look at the state of 4.2.0, from Bugzilla, and observed the
following things:

1. There are 133 P3 and higher PRs, of which about 25 are P1s.

2. Virtually all of the PRs are also in at least one of 4.1 or 4.3 --
and most are in both.

A consequence of (2) is that -- from a correctness point of view --
there isn't all that much to prevent us from releasing 4.2.0 forthwith.
 By hypothesis, 4.1 is satisfactory (shipping with major GNU/Linux
distributions, and widely used throughout the entire GCC community), so
problems that existed in 4.1 must be survivable.  I count 7 P1s that are
new in 4.2, and of those, all but two are also in 4.3.  So, fixing the
4.2 P1s now just means less work on 4.3 in future.

GCC 4.2.0 also has some good new features that are not part of FSF 4.1.
 OpenMP is the most obvious of these, but there is also support for new
CPUs, and, as always, many-a-bug fixed.

Considering the options above:

* I think [Option 3] is unfair to Kenny, Songbae, and others who have
worked on dataflow code.  The SC set criteria for that merge and a
timeline to do the merge, and I believe that the dataflow code has met,
or has nearly met, those criteria.  We should not force the dataflow
folks to maintain that code on a branch any longer.

* I think [Option 1] is not terribly productive.  I'm not aware of
anything in 4.2 that's bad, per se, with the possible exception of the
performance regression from the aliasing changes.  And, we can undo
those by reverting Danny's patch.  So, to a first approximation, we can
have the performance of 4.1 with the bugs of 4.1.  If the bugs trigger
more often in 4.2, then we can change things so they don't.

* I think [Option 2] lengthens the time between releases (which several
people have recently told me is too long, although other people have in
past told me it was too short...), but doesn't save much effort.  The
minimal way of getting to 4.2.0 is to fix the P1s common to 4.2 and 4.3,
make a decision as to what to do about the aliasing safety patches, and
declare victory.

Also, I know of several operating system distributors who plan to ship
GCC 4.2.0.  Although they do not directly contribute to GCC in the same
way that Red Hat and Novell do, they do still provide support for GCC.
I think it would set a bad precedent to pull the plug on the 4.2.0
release after having created the branch, as it's reasonable for our
entire userbase to rely on that as a commitment to produce a release.

So, my feeling is that the best course of action is to set a relatively
low threshold for GCC 4.2.0 and target 4.2.0 RC1 soon: say, March 10th.
      Then, we'll have a 4.2.0 release by (worst case, and allowing for
lameness on my part) March 31.

Feedback and alternative suggestions are welcome, of course.

Thanks,

-- 
Mark Mitchell
CodeSourcery
mark@codesourcery.com
(650) 331-3385 x713