GCC 3.4.1 Status

[Mark Mitchell <mark at codesourcery dot com>]

The 3.4 branch is now slushy -- from this point forward, all patches
need to be explicitly approved by yours truly.

Please read this entire message to see if you can help out.

There are still a large number of regressions here -- even after
pruning out most of the relatively unimportant ones, there are some
nasties left.  And I've been generous; some relatively nasty problems
have been postponed until GCC 3.4.2, including preprocessor crashes on
incorrect uses of macros and PR 15114, which is a -funit-at-time
problem with VLAs, which might bite a fair number of people.

We've got some kind of EH regression on IA64 in ILP32 mode (PR 16129),
an ICE on interactions between the named-return value optimization and
inlining (PR 15461) and we've still got several unit-at-a-time
problems -- in both directions, meaning that, as far as I can tell,
there is no safe setting, especially for C++ users.  We've got an
ICE-on-valid for IA32 on PR 15340.  We've got a case where a small C
program causes the optimizer to go into an infinite loop (PR 15320).
We need a review for the loop patch in PR 15089, which is a one-liner
-- probably obvious to someone whether this is correct or not, but we
haven't managed to get a review for this.  We have a crash in CSE on
IA32 GNU/Linux on valid code (PR 15340).  PR 16026 is waiting on Geoff
Keating to run a bootstrap/test cycle on Darwin; we'll just ignore
that one if it doesn't happen.

This is an embarassing set of failures; I'd like to see these fixed
before I give up and ship these bits.

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