Re: GCC 4.0 Status Report (2004-01-05)

[Mark Mitchell <mark at codesourcery dot com>]

Paul Brook wrote:

> > This is where our disagreement is. You can rewrite as much as you like of
> > GCC, but if there is something which makes the user unable to compile its
> > code with the new version, while it works well in the old version, then it
> > is a regression. As a user, I don't care if you rewrite everything or just
> > keep fixing bugs.

I agree with the Fortran maintainers.


Of course, I've got nothing against fixing wrong-code (or other) bugs in 
gfortran.

However, it's not accurate to think of these bugs as regressions, as 
gfortran is not a replacement for g77.  We've removed g77; we've added 
gfortran, but the latter is not a substitute for the former.

That is in contrast to other rewrites; for example, when I rewrote the 
C++ parser, I knew it would be stricter than the previous version, but I 
still meant for it to be a replacement.  Therefore, I was certainly 
concerned about situations where valid code that worked with the 
previous version did not work with the rewritten version.

Of course, gfortran would ideally correctly compile all valid F77 code, 
but, realistically that's not going to happen any time soon.  I see 
little value in marking these bugs as 4.0 regressions; I don't think 
they'll be fixed before the release, and I wouldn't imagine holding up 
the release to try to fix them.  The value in marking something as a 
regression in bugzilla is that (a) I can see the open issues, and make 
judgements about how to manage the release, and (b) the maintainers of 
various parts of the compiler can attack these problems.  With respect 
to gfortran, I am not interested in these regressions, and, so far as I 
can tell, the gfortran maintainers feel they have bigger fish to fry.

Perhaps the best way to think about it is that 4.0.x will not contain a 
Fortran compiler at all.  It will contain a technology preview of a 
Fortran 95 compiler; that compiler is already useful for some things, 
and therefore is being included in the hopes that it might be of 
utility.  But, it is a distinctly second class citizen relative to (say) 
the C and C++ compilers, which are intended to be of production quality. 
 (And, certainly, adequate replacements for GCC 3.4.x.)

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