[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: ITP: gvd

Hi,

First a disclaimer: I worked nearly two years on GNAT for Ada Core
Technologies and ACT Europe (summer, part time and full-time in the
1995-1998 range), I no longer work for them since march 1998, I'm now
working for one of their client in Paris (BNP Paribas, a Bank) and
we're paying for support for various platforms including Linux (we
have our home-grown x86 GNU/Linux cluster ;-).

I have a x86 box with Red Hat 6.1, 7.0 and Mandrake 7.1, I installed
those to help checking the GNAT RPMs portability, and I'll install
debian soon too ;-).

Christopher C. Chimelis <chris@debian.org> wrote (about GNAT on Alpha GNU/Linux):
> I've been trying to get up the energy to do this for Alpha.  Any tips on
> where to start would be appreciated, though, since the docs for porting it
> are very lacking...

In another message:
> [...] Unfortunately, I don't have Tru64 installed on any of my Alphas (no
> license) :-(  I may have access to many Alphas running Tru64 soon, though,
> so perhaps I'll wait and see what happens with that first.

You can either cross-compile from x86, from Tru64, or may be there's
some kind of binary compatibility between GNU/Linux and Tru64, I don't
know. 

My understanding of the cross process (Richard Kenner explained it to
me a while ago), is that you configure for a cross target, you make
with "-S" to produce a whole bunch of ".s" files, then on the target
you assemble and link the whole thing (ftp stuff if no NFS), you
install what you have somewhere and then you try a full native build
on the target. 

IIRC there are special targets in the GNAT Makefile to do so.

The compiler itself depends on small non problematic parts of the
runtime.  The last step is to adjust target dependant things in the
Makefile to get tasking working (should be very near x86 GNU/Linux
with glibc threads, may be a few constants and Makefile hacks to
recognize the target name).

I have recently "packaged" the executable part of the official Ada 95
validation suite "ACATS" (announced on GCC and on RTEMS lists, the
RTEMS guyes mirrored the package, see emails after my signature).

I'm willing to help anyone that wants to work on porting/testing GNAT.

> I saw the threads on the gcc list about incorporating GNAT into gcc
> (finally) and that sounds like it'll work out, although I don't know how
> long it will take (seems like ACT is having a hard time letting go of the
> control that they currently have).

David Starner <dvdeug@x8b4e516e.dhcp.okstate.edu> wrote:
> This message came across comp.lang.ada at one point. I don't know if it
> will help much - you might try emailing the author, who works for ACT.
> Alternately, GNAT should be part of GCC Any Day Now, and at that point,
> gcc@gcc.gnu.org will probably be an excellent place to get assistance.

Right. 

<RUMOUR> 

Since Richard Kenner seems to be tackling VMS and DOS issues, which I
guess are at the end of his queue ;-), I assume RSN is the time
frame. My bet (from recent discussion on gcc) is that they are
checking performance issues. Also I think there will be another 2.8.1
based release, GNAT 3.14, and this will probably be the last one
before the move to GCC CVS.

</RUMOUR>

Aaron Lehmann <aaronl@vitelus.com> wrote:
> FWIW, the specs of the professional ($$$) version describe it as
> compatible with Tru64 Unix and VMS, so I bet it either already has
> internal Alpha support in the free version or could easilly be ported
> to Alpha.

ACT no longer publically distributes binaries for platforms other than
Windows/x86, GNU/Linux/x86 and Solaris/SPARC. However as a customer I
have access to all binaries including those for Tru64 4.0f, and the
license is GPL, so no legal problem.  I will inform ACT anyway, they
might already have looked at the issue, and if so I'm sure they will
be helpful.

I'm out until 1800 Paris time, so do not expect immediate replies.

-- 
Laurent Guerby <guerby@acm.org>

======================================================================

$Id$
README.ACATS4GNAT for version 0.2, 20Nov2000

To run ACATS 2.3 for GNU Ada:

$ cd acats4gnat-0.1/work
$ time ../acats4gnat/run_all.sh >& run.log &

It will take 20 minutes to download 6.1M of ACATS on 56K modem (it
only happens the first time), and then compile and run for 45 minutes
on a P2 350MHz. You need 60M of disk space in the work directory.

Once it is finished, you should see for GNAT 3.13p/ACT:

$ grep 'acats4gnat result' run.log 
acats4gnat results cz 3 /       4
acats4gnat results a 75 /      75
acats4gnat results c2 35 /      35
acats4gnat results c3 348 /     349
acats4gnat results c4 338 /     338
acats4gnat results c5 95 /      95
acats4gnat results c6 81 /      81
acats4gnat results c7 48 /      50
acats4gnat results c8 140 /     140
acats4gnat results c9 254 /     255
acats4gnat results ca 73 /      74
acats4gnat results cb 43 /      43
acats4gnat results cc 117 /     117
acats4gnat results cd 172 /     172
acats4gnat results ce 268 /     268
acats4gnat results cxa 85 /      85
acats4gnat results cxb 29 /      30
acats4gnat results cxc 16 /      16
acats4gnat results cxd 39 /      39
acats4gnat results cxe 1 /       1
acats4gnat results cxf 20 /      20
acats4gnat results cxg 29 /      29
acats4gnat results cxh 4 /       4
acats4gnat results d 4 /       4
acats4gnat results e 11 /      11
acats4gnat results l 26 /      26

2361 tests are built and run, 14 could be made to run but are not
packaged yet, 2 are not run because they cause major disruption,
see acats4gnat/norun.lst for details.

The know failures are detailed in acats4gnat/BUGS.

Note that GNAT 3.13p/ALT fails a few more tests, because of lack of full
Annex D support, problem with select then abort on the main thread
with LinuxThreads, and of code generation problems with -O2 -fPIC on
Ada.Strings.Maps which cause 2 cxa tests to fail (all reported to
ACT).

Laurent Guerby <guerby@acm.org>

======================================================================

Mailing-List: contact rtems-users-help@oarcorp.com; run by ezmlm
Sender: joel@mail.club-internet.fr
Date: Tue, 21 Nov 2000 08:36:56 -0600
From: Joel Sherrill <joel.sherrill@oarcorp.com>
Reply-To: joel.sherrill@oarcorp.com
Organization: OAR Corporation
X-Accept-Language: en
To: rtems-users@oarcorp.com
Subject: GNAT/RTEMS and ACATS Suite
Content-Type: text/plain; charset=us-ascii
X-Mmail: \Recent                                 
X-M-Uid: 0058307.974818250


This is a SIGNIFICANT opportunity to contribute to the
GNU community in general and Ada community in particular.

Laurent Guerby <guerby@acm.org> has been working on a set of
scripts to run the ACATS (formerly ACVC) on GNAT.   This is
in anticipation of the merger of GNAT into the main
gcc.gnu.org CVS repository.  I think I am speaking for both
Laurent and myself when I say that we would love to see 
this suite run on GNAT/RTEMS.  I have made the scripts
available at:

ftp://ftp.oarcorp.com/pub/rtems/snapshots/contrib/acats

The full ACATS includes 1000's of tests.  I think Laurent
has focused on the executable portion of them.  This is
a VERY serious test of the Ada run-time and a very good
exercise of the POSIX tasking model.  I used the ACVC to
find problems in the RTEMS POSIX implementation and debug
it when it was first implemented.

Laurent wrote (the first?) ACVC scripts for ACT.  I used
his scripts to run the validation suite on the SPARC/ERC32
when we received the official validation certificate.

I would encourage others to play with this. :)  Eventually
GNAT will be merged into the main gcc source repository
and it will be then possible to build GNAT for any RTEMS 
target.  If there is a simulator, it will be easy to 
run the full suite automatically.

FYI .. simulators out there and RTEMS status in relation:

  + sparc, powerpc, c3x/c4x - suitable for ACATS
  + i960, h8, arm - missing timer source
  + m68000/cpu32 (BSCV) - suitable but BSP not complete
  + i386 (plex86) - suitable but needs attention
  + MIPS - suitable but no BSP
  + SH - suitable but no BSP

I would encourage the GNAT/RTEMS folks to rally around
and have an impact.  We can be THE GNAT-cross community.

-- 
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

======================================================================

Mailing-List: contact rtems-users-help@oarcorp.com; run by ezmlm
Date: Wed, 22 Nov 2000 00:56:03 +0100
X-Authentication-Warning: ulmo: guerby set sender to guerby@acm.org using -f
From: Laurent Guerby <guerby@acm.org>
To: rtems-users@oarcorp.com
In-reply-to: <3A1A8888.8AAA9C7A@OARcorp.com> (message from Joel Sherrill on
	Tue, 21 Nov 2000 08:36:56 -0600)
Subject: Re: GNAT/RTEMS and ACATS Suite
Reply-to: guerby@acm.org
X-Mmail: \Recent                                 
X-M-Uid: 0058510.974851072

Just a few notes on the ACATS scripts made available:

- The acats4gnat script are not designed to do an official validation,
just as a way to get all executable tests to run as easily as
possible, and for this purpose I've tried to keep them as simple as
possible. Non executable tests are not even looked at.

- Adaptation to a cross environment should be easy since the script
uses the regular "gnatmake" command, and there is one line to launch
the test executable. I haven't tried to do so yet.

- The scripts ACT uses (and Joel mentions) are far more complicated
because they're addressing the process of a real Ada validation, and I
worked on them while working for ACT, but I believe Gary Dismukes
wrote the initial version of them.

- ACT has kindly contributed their ACATS configuration files which I
hacked a bit.

- I worked for ACT, but I no longer do, at my new work we're client of
ACT services in the non-embedded work.

- We're all waiting for the GNAT sources integration into GCC CVS
repository, which should solve a lot of problems with GNAT/RTEMS.

- When this happens I'll try to provide an integrated GCC/Ada testing
driver, and GNAT/RTEMS support is high on the list ;-).

- Feel free to contact me through this list or at <guerby@acm.org> if
you need help, answers, have suspicious looking failures or have
contribution to make. I read my personal email from 2000 to 2300 Paris
time, if I don't answer after a week, please resend, this is a
processing mistake on my side ;-).

-- 
Laurent Guerby <guerby@acm.org>
Reply to: