Re: GNAT and GCC 3.0

[dewar@gnat.com (Robert Dewar)]

Ronald Cole <ronald@forte-intl.com> wrote in message news:<m3vghvau6m.fsf@yakisoba.forte-intl.com>...
> I remember offering to help since I was able to patch
> 3.11p to compile itself with gcc-2.95,

Well it was easy enough to do a patch that would sort of
allow 3.11 to work with gcc-2.95, but the result was hopeless from an
Ada point of view, since 2.95 had so
many serious unfixed bugs that might not have affected
g++ so much, but really resulted in a completely unusable
Ada port (as evidenced by a huge number of failures in
our test suites).

A major part of the effort in getting GNAT to some resaonable level of
compatibility with gcc was to check
in thousands of lines of patches to fix these deficiencies
in 2.95 (many of these were 2.8.1 patches that had never
made it to 2.95). That work is reasonably complete, and
although the current checked in sources are still not
at the point where we feel that a reliable commercial
product (or public release) can be generated from this
version of the sources, we are definitely getting nearer.

Internally at ACT, we are working on identifying the
remaining problems and fixing them (as is normal, the
issue is not so much fixing the problems, as identifying
why some giant test case is not working -- and these test
cases are mostly proprietary code, so this is something
we have to work on).

As we work out the necessary fixes, they will eventually
be reflected in the sources at gnu.org, and eventually we
may be able to bring our internal tree into very close
correspondence with the one at gnu.org (right now our
primary internal tree is still 2.8.1 based, since that
is still the only fully reliable version).

> but you wouldn't cut loose with your precious
> "wave front" sources back then.

It's not a matter of precious, but in our judgment, it would
have been quite unhelpful to release random sets of sources
which had not been fully field tested. The sources at gnu.org are of
quite a different character, and it is important to understand that,
unlike the commercial releases of GNAT, they have not successfully
passed our
internal quality testing.

The good thing about the sources being at gnu.org is that
it allows the general community to contribute more effectively, but it
is important to understand that the
consequence is that there can be multiple versions of 
GNAT around from this source and you need to be definitely
aware of what you are getting and what you can expect from it in terms
of quality and reliability.

ACT will continue to release commercial versions of GNAT
that have been through all our quality procedures, and
at least for a limited set of targets, will follow along
with corresponding public releases, to be available on our
libre site at www.act-europe.fr

Eventually, we will get the ACATS test suite set up as
part of the gnu.org material, as well as a selected set
of ACT tests that we wrote and which can be distributed.
That will help ensure that the gnu.org version stays
coherent. We also will move towards a situation where
people will be able to build GNAT to check that their
gcc patches do not upset its integrity, but of course
there is no guarantee that any given days snapshots
are fully reliable.

Robert Dewar