Re: GCC 3.1 released

[dewar@gnat.com (Robert Dewar)]

"Jeffrey Creem" <jeff@thecreems.com> wrote in message news:<PmtF8.29827$sg2.7831839@typhoon.ne.ipsvc.net>...

> Yup..There are problems.. There are also problems in GNAT 
> 3.14, 3.13, 3.12, 3.15 ... etc.

But none of these releases failed ACATS tests and tests in
our test suite. GNAT 5 has significant numbers of failures
in all categories on all targets. Several of these are
of the form of incorrect code being generated silently
and resulting in wrong results. I am not saying that the
build is unusable, not at all, it will probably work fine
for a lot of stuff, but it is still not at the product
releasable level for us. Actually my current view is for
our most up to date internal tree, which is quite a bit
beyond the 3.1 release, so the 3.1 release may well have
additional problems, we have not run the 3.1 release itself
against our test suites at all. I believe Laurent Guerby
is working to make the ACATS suites usable in the FSF
context.

One thing to understand here is that in the past when we
have released a version like 3.15p it has passed all our
internal qualification tests at the point corresponding
to its internal freeze date. That's always been a criterion
for any release (of GNAT Pro or the public version). A
consequence was that the public versions were always quite
a bit behind, but reasonably reliable. 

Our new approach for public releases, integration into the
GCC 3 tree, does not give any such guarantees. The plus is
that it is far closer to our development wavefront, the minus is that
it is likely to be somewhat unstable. Our
feeling is that students can probably stay with 3.14p just
fine, and hobbyists who want to fiddle with the latest and
greatest and don't care whether it's 100% reliable can 
benefit from the GCC 3 version.