GCC 3.4 Release Status (2003-12-16)

[Mark Mitchell <mark at codesourcery dot com>]

GCC 3.4
=======

GCC 3.4 has, at this point, a ways to go.

There are 186 bugs in bugzilla targeted for GCC 3.4, and that,
historically means months of effort.  There are about 50 C++ bugs, but
that number is falling rapidly.  There are about 40
optimization/middle-end bugs.  We have 13 "debug" regressions.  

I suspect some of these bugs are not actually regressions in 3.4, and
some are not problems on high-priority platforms -- but there are
still a ton of open bugs.

I don't intend to create a GCC 3.4 branch until we've made a
significant dent in the outstanding regressions -- say, until we've
cracked the 100 regression limit, and maybe not until we break 50.
Experience has taught me that creating a release branch when there are
this many bugs is a prescription for people doing development on the
mainline, leaving the release branch to languish indefinitely.  I'll
be going through the open PRs and trying to see what can be pruned
away.

I think that we could create a 3.4 branch in as little as a few weeks,
if we knuckle down on the outstanding PRs.

There are six open PCH bugs.  That's a new feature -- but it's one
that people are going to try to use, and, from their point of view,
problems with PCH that didn't occur without PCH are going to be
regressions.  Geoff, would you take a look at these problems?
Alternatively, how would you feel about documenting PCH as
"experimental" so that people realize that this feature is not as
solid as we might like?

What is the realistic status of Java maintenance at this point?  Is it
reasonable to look for Java regressions to get fixed, or do we not
have enough personpower to do that?  For the last several releases, we
do not seem to have been able to fix Java PRs very effectively.  A
similar question applies to G77.  If we can't fix problems in these
languages effectively, that's OK -- we just need to admit that to
ourselves and move forward.

Do we still need -fwritable-strings?  We have a 3.4 regression with
that option, and I would prefer just to remove the option, since we're
moving away from supporting traditional C anyhow.

Previous Status Report
======================

http://gcc.gnu.org/ml/gcc/2003-10/msg00119.html

--
Mark Mitchell
CodeSourcery, LLC
mark@codesourcery.com