Fwd: GCC 4.2.0 Status Report (2007-02-19)

[Janne Blomqvist]

On Tue, Feb 20, 2007 at 01:36:16PM +0100, Tobias Burnus wrote:
> (1) Should there be a 4.2 (based on the current 4.2 branch) or not?
> The replys seems to indicate a yes; Brooks stressed that it is important
> for gfortran.
> cf. thread at http://gcc.gnu.org/ml/gcc/2007-02/msg00427.html

FWIW I agree with Brooks and Mark. It's a bit too late to start
back-pedalling on 4.2 now. It won't be a perfect release, but then
again nothing is. Perfection is the enemy of good enough, as the
saying goes.

> (2) Performance regressions of SPEC
> cf. thread at (1) and
> http://gcc.gnu.org/ml/gcc/2007-02/msg00424.html

Seems like backporting the relevant 4.3 stuff is a bit too much work
(according to the mails exchanged between M.M and D.B), so the choice
assuming that 4.2 should be relased as originally planned is then
between reverting the patch or living with the same bugs as
4.1. Personally, I can live with both choices.

Does anyone have any idea how this stuff affects gfortran? Given the
aliasing rules in Fortran, we should be able to tell the middle end
that procedure arguments are all the equivalent of C99 "const
restrict" (and an additional const for INTENT(IN) arguments), thus
making Fortran less dependent on heroic alias analysis.

-- 
Janne Blomqvist