Re: [RFC] Merge criteria for tree-ssa

[Robert Dewar <dewar at gnat dot com>]

Mark Hahn wrote:

> > The criteria that makes most sense from a users viewpoint is that
> > tree-ssa could not be merged until gfortran was as good as g77 at
> > compiling f77.

But that's not fair, the quote was in a context that clearly
understood this as unreasonable.

> as another user, I strongly disagree.  can't people with f77 code
> simply compile with older versions of gcc?

Not if they need other aspects of a newer gcc (e.g. more functional
ia64 support)

> gcc badly needs the optimization benefits promised by ssa

When you "badly need" something that is only "promised", you
may be setting yourself up for disappointment :-)

> not to mention
> support for f90, etc.  there's nothing wrong with temporary regressions 
> that are a side-effect of progress elsewhere; in fact, it's just silly 
> to think that quality should increase monotonically along every axis.

Commercial products are usually constrained to exactly this rule, and
users expect it. I can't imagine our Ada customers being happy to be
told. Never mind if some things don't work any more, those that still
work work must faster!

So in terms of releases I think this is a dangerous principle, and that
indeed for releases quality must increase, or at least not decrease
on every axis.

Of course for fortran, you can just decide that you are skipping a
release if necessary. That might for example happen with Ada, though
we hope not!

> regards, mark hahn.