Re: Recent Ada ACATS regressions

Re: Recent Ada ACATS regressions

[Robert Dewar]

nothing to share on the arm project yet, something may appear later.

Re: Recent Ada ACATS regressions

[Laurent Guerby]

> The MIPS port requires Laurent's fix in PR8344.
> Basically the file gcc/ada/final.c conflicts in the build with
> the file gcc/final.c.  His fix was to renamed the one in the
> ada directory to adafinal.c.

If anyone could at last approve this one (for 3.2 and HEAD) or clearly 
state what's wrong
with it (please no argument about rewriting the whole ada/Makefile 
thanks :),
that would be great. FYI here is the non blank / non comment
content of the ada/final.c file, in extenso:


void
__gnat_finalize ()
{
}



-- 
Laurent Guerby <guerby@acm.org>

Re: Recent Ada ACATS regressions

[Joel Sherrill]

Robert Dewar wrote:
> 
> We are now finally close to a beta release of the GCC 3 based technology for
> our customers. What that means is that we finally feel we have something in
> moderately reasonable shape at least on some targets (Windows, Solaris,
> GNU/Linux are all in pretty good shape).
> 
> That was a critical point for us to get to internally, now it makes sense
> to resyncrhonize the trees. We have also made some internal improvements
> to the format of the files (in particular removal of internal version
> numbers) which should make the merge easier in future.


FWIW we were getting pretty good results on the sparc/erc32 and
powerpc/psim with GNAT/RTEMS.  We have also spot checked 
i386 and mips.  The MIPS port requires Laurent's fix in PR8344.
Basically the file gcc/ada/final.c conflicts in the build with
the file gcc/final.c.  His fix was to renamed the one in the
ada directory to adafinal.c.

Any news on the arm or m68k to share?  Dr. Dewar hinted at some
arm progress a while back. :)

-- 
Joel Sherrill, Ph.D.             Director of Research & Development
joel@OARcorp.com                 On-Line Applications Research
Ask me about RTEMS: a free RTOS  Huntsville AL 35805
Support Available                (256) 722-9985

RE: Recent Ada ACATS regressions

[Jerry van Dijk]

Arnaud Charlet writes:

> ACT's "internal" tree is based on the GCC 3.2 branch, and this is were we
> have stabilized things and focused our efforts (this is what Robert's
> message explained), since working on the GCC HEAD has proven to be
> impractical
> for getting a stable compiler, at least in our experience and for the kind
> of quality our customers need.

I am happy to see progress. However will a 3.2 merge also benifit a possible
3.2.2 release ? On mingw we had to leave Ada out of the 3.2.1 release as the
compiler was not yet in a really usable shape. It would be great to add it
back to a 3.2.2 release.

--  Jerry van Dijk   | email: jdijk@acm.org
--  Leiden, Holland  | web:   users.ncrvnet.nl/gmvdijk

Re: Recent Ada ACATS regressions

[Robert Dewar]

> However, with my free software home developper hat on, it seems
> to me that this implies that ACT spends ressources
> on 3.2 (reasonable) and so has very little ressource left for HEAD.

That's certainly been true up to now, since it has been a very big effort
to get a realiable GNAT working on GCC 3.x. But now we are almost there,
we definitely will shift resources to HEAD.

> Would it be possible to ship diffs to volunteers between 3.2 CVS and
> ACT and/or HEAD and some ACT person tree if it makes more sense?
> (You might want to put them on ACT Europe libre site or something
> like that.)

We will definitely investigate this possibility. 

Re: Recent Ada ACATS regressions

[Laurent Guerby]

Arnaud Charlet wrote:
 > I think it is appropriate to add some clarification.
 >
 > ACT's "internal" tree is based on the GCC 3.2 branch, and this is were we
 > have stabilized things and focused our efforts (this is what Robert's
 > message explained), since working on the GCC HEAD has proven to be 
impractical
 > for getting a stable compiler, at least in our experience and for the 
kind
 > of quality our customers need.

Thanks for the clarification Arnaud!

 > The new failures you are reporting are related to changes in
 > the HEAD that are new uncovered failures, or most likely require 
changes in
 > gigi.
 >
 > ACT does *not* currently have any GCC 3.3 or 3.4 version that works or
 > has been tested (as explained by Richard's message), so this is not a 
simple
 > matter of merging, since we do not have anything to merge on the HEAD at
 > this point.

With my ACT customer hat on, I agree that it is unreasonable to release
anything to customers from HEAD :).

However, with my free software home developper hat on, it seems
to me that this implies that ACT spends ressources
on 3.2 (reasonable) and so has very little ressource left for HEAD.

Would it be possible to ship diffs to volunteers between 3.2 CVS and
ACT and/or HEAD and some ACT person tree if it makes more sense?
(You might want to put them on ACT Europe libre site or something
like that.)

People outside ACT might not be able to help that much, but at least
we can help pinpoint some amount of HEAD regressions on a per patch
basis with some confidence that it's not wasted effort.

The ACATS test prototype is used to my knowledge on a significant
number of platforms now.

-- 
Laurent Guerby <guerby@acm.org>

Re: Recent Ada ACATS regressions

[Arnaud Charlet]

I think it is appropriate to add some clarification.

ACT's "internal" tree is based on the GCC 3.2 branch, and this is were we
have stabilized things and focused our efforts (this is what Robert's
message explained), since working on the GCC HEAD has proven to be impractical
for getting a stable compiler, at least in our experience and for the kind
of quality our customers need.

The new failures you are reporting are related to changes in
the HEAD that are new uncovered failures, or most likely require changes in
gigi.

ACT does *not* currently have any GCC 3.3 or 3.4 version that works or
has been tested (as explained by Richard's message), so this is not a simple
matter of merging, since we do not have anything to merge on the HEAD at
this point.

Arno

RE: Recent Ada ACATS regressions

[Jerry van Dijk]

Laurent Guerby writes:

> Any progress on ACT side on the merge? For volunteers I assume
> it is a total waste of time to chase such problems since
> with high probability they've been fixed in the ACT tree.

Which is exactly my problem too. I report mingw ACATS (and other
failures) on GNATS, but as ACT might already have private fixes
for them, I am not going to look any further into them.

--  Jerry van Dijk   | email: jdijk@acm.org
--  Leiden, Holland  | web:   users.ncrvnet.nl/gmvdijk

Re: Recent Ada ACATS regressions

[Robert Dewar]

We are now finally close to a beta release of the GCC 3 based technology for
our customers. What that means is that we finally feel we have something in
moderately reasonable shape at least on some targets (Windows, Solaris, 
GNU/Linux are all in pretty good shape). 

That was a critical point for us to get to internally, now it makes sense
to resyncrhonize the trees. We have also made some internal improvements
to the format of the files (in particular removal of internal version
numbers) which should make the merge easier in future.

Re: Recent Ada ACATS regressions

[Richard Kenner]

    Any progress on ACT side on the merge? 

I've started to work on getting the current ACT sources to work with
top-of-tree GCC, which is the first step.

    For volunteers I assume it is a total waste of time to chase such
    problems since with high probability they've been fixed in the ACT tree.

I agree.

Recent Ada ACATS regressions

[Laurent Guerby]

A dozen tests ICE, they weren't on 20030118.

c37010a 
c37010b 
c371003 
c37213f 
c37213j
c37213k 
c37213l 
c37215h 
c37312a 
c41301a
c41304b 
c48005a

Small testcase below. On the other side, three tests.

cxa4026 
cxa4027 
cxa4030

are now passing instead of failing during execution, these
are the three which started failing after the BIB merge,
reported  at <http://gcc.gnu.org/ml/gcc/2002-12/msg00969.html>
(which also has a small test case).

Any progress on ACT side on the merge? For volunteers I assume
it is a total waste of time to chase such problems since
with high probability they've been fixed in the ACT tree.

After the merge, what about uploading the daily ACT tree / GCC tree
diff to a few volunteers to assist with the synchronisation?

-- 
Laurent Guerby <guerby@acm.org>

$ gcc -c t1.ads
+===========================GNAT BUG DETECTED==============================+
| 3.4 20030126 (experimental) (i686-pc-linux-gnu) GCC error:               |
| in tree_low_cst, at tree.c:3205                                          |
| Error detected at t1.ads:3:4                                             |
| Please submit a bug report; see http://gcc.gnu.org/bugs.html.            |
| Include the entire contents of this bug box in the report.               |
| Include the exact gcc or gnatmake command that you entered.              |
| Also include sources listed below in gnatchop format                     |
| concatenated together with no headers between files.                     |
+==========================================================================+
$ cat t1.ads
package T1 is

    type T (A : Integer) is record
       case A is
          when 1 => A1 : Integer;
          when others => null;
       end case;
    end record;

end T1;