GCC Status Report (2004-10-08)

[Mark Mitchell <mark at codesourcery dot com>]

GCC 3.4.3
=========

There are 129 regressions targeted at 3.4.3, of which 77 are also 4.0
regressions.  If you're interested in fixing 3.4.3 regressions, you
might give special preference to those that are also 4.0 regressions.
Although new regressions are definitely still being found in the 3.4.x
series, I see no problem with releasing 3.4.3 at the end of this
month.

GCC 4.0
=======

I've been informed that there's considerable confusion about what
exactly is permissible in Stage 3, especially given that we've given
some special dispensation to certain projects during this particular
incarnation of Stage 3.

The short answer is "bug fixes, but not new features".  That, of
course, begs the question "what's a bug fix and what's a feature?"  I
don't think there's a bright-line rule that will distinguish these
cases.  However, there are a few areas that can be clarified.  First,
patches for performance improvements (either at compile-time or
run-time) are acceptable, if they're of a relatively limited nature.
For example, tweaking an optimization pass to use different heuristics
would be acceptable, if the relevant maintainers agree that makes
sense.  Adding a new optimization pass would not be acceptable,
though. Please let me know if there's continuing unclarity here, and I
will try to help resolve questions.  It seems to me as though people
are exercising good judgement and erring on the side of caution, as is
desirable.

There are 223 PRs targeted at GCC 4.0, including the 77 mentioned
above.  That's rather a large number, but I still have not fully
examined the open PRs; I suspect that some of these PRs will be closed
or postponed.  Also, I'm pleased to see lots of gcc-patches traffic
with PRs getting fixed.  It's been suggested that we again make it a
goal to get the total number of PRs targeted at 4.0 down to 100 before
creating the release branch.  I think that's a good goal, so we'll
make that an informal milestone target. I still think we can get there
soon.  If not by the end of October, at least before Thanksgiving!
(For those of you not in the US: an holiday on the fourth Thursday of
November involving turkeys, corn, Native Americans, pies, potatos, and
football, in no particular order.)

In looking at the outstanding list of projects, as maintained on
Daniel Berlin's Wiki, we've still got a few we want to get reviewed.
I see that RTH is reviewing the vectorization patches, which is
great.  I'm going to look at variable expansion and Sebastian's data
dependence analysis.  I'd appreciate other maintainers helping out
with any of the other patches.

--
Mark Mitchell
CodeSourcery, LLC
(916) 791-8304
mark@codesourcery.com