Re: GNAT and GCC 3.0

[dewar@gnat.com (Robert Dewar)]

Ronald Cole <ronald@forte-intl.com> wrote in message news:<m3g08sil5p.fsf@yakisoba.forte-intl.com>...

> Yes, gnat will be in gcc-3, but Mr. Dewar cautions that 
> it is "highly experimental"; which I interpret to mean 
> that he doesn't consider it as reliable as the gcc-2.8 
> based gnat.  I'm reasonable certain that
> that wouldn't be the case today had Mr. Dewar really 
> opened up gnat five years ago and allowed a lot more eyes 
> on his work-in-progress.

That would not have helped at all. All the problems are
in gcc, not in gnat. The first task was to take all the
fixes that we had published openly in the GCC 2.8 sources
and fit them into 3.x. Richard Kenner did this work on
behalf of ACT as you will remember, and it involved many
thousands of lines of fairly complex patches.

We then had to steadily fix GCC 3 problems to get GNAT to
work correctly, and it is not a matter of us "not considering" the
current version to be less reliable, it
is that we *know* it is less reliable. How do we know? --
because tests in our test suite based on customer proprietary code are
failing, and of course only we can
work on these problems, and we are doing so (yes, once we
understand the problem, we can create a small cut down
example, but once we understand the problem, we can easily
fix it anyway).

We also have the problem of continued patches to GCC breaking things.
Hopefully soon, more people will be
building GNAT to test their changes, and that will help.
But it is fairly frequent that people check in changes
that break C bootstraps, so that problem won't go away
completely. You have to be careful in picking up the
snap shot sources to get a good set, and that means
watching the gcc mailing list. In general I would advise
anyone interested in picking up GNAT snapshots from 
gnu.org to subscribe to the relevant lists, and get a
feel for what is going on.

Originally we had wanted to delay submitting the sources
till the test suites showed them to be clean, but we decided to submit
them earlier, because they are in reasonable shape for many simple
programs, and because
then people can figure out build procedures etc (most of
the useful discussions so far have been with respect to
build procedures on the gcc list).

Certainly Ron Cole was free, as was anyone else, to get
3.13 running on 3.1 (the issues of getting 3.13 running
are identical to those of getting later versions to run,
and as above, the problems are in gcc, not in gnat). In
fact several people *did* try to get 3.13 running on GCC
3, but no one came near to succeeding, which did  not surprise us,
since it was a huge amount of work, and indeed
the work is still not complete.

Ron, please feel free to set to work and fix any remaining
problems you see :-)

Oh, and by the way ACT now has 35 full time people, "Mr.
Dewar" does not make all the decisions, such decisions
are made by the appropriate groups of people in the company, and
represent a consensus position within ACT.

Robert Dewar