Re: GNAT ACT source tree

Re: GNAT ACT source tree

[Richard Kenner]

    I succeeded with bootstraping on x86-linux as well by plugging
    the ACT CVS sources plus files from gcc-head branch.

Good.  Thanks for the information.

    I needed to use 5.00 as base compiler to avoid failures (due to
    warnings) on the new "Unreferenced" pragma which is present in ACT
    sources, I assume ACT is using either 3.16 or 5.00 as baseline
    compiler. I assume either Warnings (Off) or Makefile trickery are
    needed if we want to preserve 3.2 GNAT as base build compiler.

Most ACT build machines will be using 3.15.  I was using a compiler of
approximately 5.00 vintage in my bootstrap.

    I needed a small patch to utils2.c which I did not found in gcc-32.dif:

Well, gcc-32.dif is a diff to GCC sources, so it wouldn't be there.

    --- gnat-head/utils2.c   
    +++ ./utils2.c
    +#include "output.h"

Peculiar that it isn't needed on Alpha, but I'll add that.

    Not counting tasking link failures due to this oversight, ACATS run
    showed 30 less failures than current CVS head. 

That's encouraging.  As far as I know, you are the first to do an ACATS
run with those sources.

Re: GNAT ACT source tree

[Laurent Guerby]

On Thu, 2003-04-17 at 13:36, Richard Kenner wrote:

>     --- gnat-head/utils2.c   
>     +++ ./utils2.c
>     +#include "output.h"
> 
> Peculiar that it isn't needed on Alpha, but I'll add that.

For reference the message was:

stage1/xgcc -Bstage1/ -B/home/guerby/work/gcc/install/install-2003-04-16-23-21-30/i686-pc-linux-gnu/bin/ -c-g -O2 -DIN_GCC   -W -Wall -Wwrite-strings -Wstrict-prototypes -Wmissing-prototypes    -Werror -fno-common   -
DHAVE_CONFIG_H    -I. -Iada -I/home/guerby/work/gcc/gcc/gcc -I/home/guerby/work/gcc/gcc/gcc/ada -I/home/guerby/work/gcc/gcc/gcc/config -I/home/guerby/work/gcc/gcc/gcc/../include /home/guerby/work/gcc/gcc/gcc/ada/utils2.c -o ada/utils2.o
/home/guerby/work/gcc/gcc/gcc/ada/utils2.c: In function `gnat_build_constructor':
/home/guerby/work/gcc/gcc/gcc/ada/utils2.c:1565: warning: implicit declaration of function `initializer_constant_valid_p'
make[2]: *** [ada/utils2.o] Error 1
make[2]: Leaving directory `/home/guerby/work/gcc/build/build-2003-04-16-23-21-30/gcc'
make[1]: *** [stage2_build] Error 2
make[1]: Leaving directory `/home/guerby/work/gcc/build/build-2003-04-16-23-21-30/gcc'
make: *** [bootstrap] Error 2
 
Not obvious how you don't get that on alpha, no target ifdef lurking
around.

-- 
Laurent Guerby <guerby@acm.org>

Re: GNAT ACT source tree

[Robert Dewar]

> 
> This was actually a transient error in our sources, which will be fixed
> with the next mirror.


Perhaps that is a good opportunity to emphasize that in general there is
no guarantee that the set of files available here will bootstrap or be
error free. This is an unmolested view of the current top-of-the-tree at
ACT, and such errors may occur from time to time. They will usually get
fixed right away, as we see them in our nightly builds.

Re: GNAT ACT source tree

[Arnaud Charlet]

> I needed to use 5.00 as base compiler to avoid
> failures (due to warnings) on the new "Unreferenced" pragma which
> is present in ACT sources, I assume ACT is using either 3.16 or 5.00 as
> baseline compiler. I assume either Warnings (Off) or Makefile trickery
> are needed if we want to preserve 3.2 GNAT as base build compiler.

This was actually a transient error in our sources, which will be fixed
with the next mirror.

Arno

Re: GNAT ACT source tree

[Laurent Guerby]

On Wed, 2003-04-16 at 14:13, Richard Kenner wrote:
>     We have a version of GNAT that should more or less compile with the
>     current GCC HEAD. 
> 
> Compiles and bootstraps (on Alpha Unix), but that's all the testing
> done so far.  There's no point in doing more until the merge from
> gcc-32.dif is done since those patches are needed to fix known bugs.

I succeeded with bootstraping on x86-linux as well by plugging
the ACT CVS sources plus files from gcc-head branch.

I needed to use 5.00 as base compiler to avoid
failures (due to warnings) on the new "Unreferenced" pragma which
is present in ACT sources, I assume ACT is using either 3.16 or 5.00 as
baseline compiler. I assume either Warnings (Off) or Makefile trickery
are needed if we want to preserve 3.2 GNAT as base build compiler.

I needed a small patch to utils2.c which I did not
found in gcc-32.dif:

--- gnat-head/utils2.c   
+++ ./utils2.c
@@ -30,6 +30,7 @@
 #include "tm.h"
 #include "tree.h"
 #include "flags.h"
+#include "output.h"
 #include "ada.h"
 #include "types.h"
 #include "atree.h"

I also removed gnatmem and -lgmem related stuff (list
of object files not including memtrack.o and -laddr2line
not being on my machine) in order to build
gnatlib_and_tools

I found afterwards I should have included this patch from gcc-32.dif
in order to be able to link tasking programs:

2002-09-17  Vincent Celier <celier@gnat.com>
 
        * configure.in: Add gnat to the list of thread packages
 
2002-09-12  Vincent Celier <celier@gnat.com>
 
        * gthr-gnat.h, gthr-gnat.c: new sources for implementation of
        --enable-threads=gnat.
        * Makefile.in: Add gthr-gnat.c to LIB2ADDEH.
        * configure: Add gnat to the list of thread packages.
        * config/t-linux: Add gthr-gnat.c to LIB2ADDEH and LIB2ADDEHDEP

Not counting tasking link failures due to this oversight, ACATS run
showed 30 less failures than current CVS head. 

I'll probably be off-net this week-end.

-- 
Laurent Guerby <guerby@acm.org>

Re: GNAT ACT source tree

[Geert Bosch]

On Wednesday, Apr 16, 2003, at 06:42 America/New_York, Laurent Guerby 
wrote:
> I also noticed there are some RCS
> tags in the sources so one must be careful not to reintroduce
> them when merging into the FSF tree.

I will remove these from the ACT tree as well as apply the other
header changes required.

   -Geert

Re: GNAT ACT source tree

[Richard Kenner]

    We have a version of GNAT that should more or less compile with the
    current GCC HEAD. 

Compiles and bootstraps (on Alpha Unix), but that's all the testing
done so far.  There's no point in doing more until the merge from
gcc-32.dif is done since those patches are needed to fix known bugs.

Re: GNAT ACT source tree

[Richard Kenner]

    I noticed that some patches from gcc-32.dif have been recently commited
    by Richard Kenner to the mainline. 

Right.  I'm working on going through all of them, but it will take time.

Re: GNAT ACT source tree

[Arnaud Charlet]

> Thanks! CVS worked for me. What is the best way to proceed in order
> to help the merge process? Is the ACT tree tested with mainline GCC?

Not currently.

We have a version of GNAT that should more or less compile with the
current GCC HEAD. to get it, you need to get the file on the "gcc-head"
branch.

Note that only a few files (the gigi files mainly) are branched, so you
need to get the other files from the HEAD. cvs update -f is your friend.

Arno

GNAT ACT source tree

[Laurent Guerby]

> cvs -d :pserver:anoncvs@libre.act-europe.fr:/anoncvs login
> empty password
> cvs -d :pserver:anoncvs@libre.act-europe.fr:/anoncvs co gnat


Thanks! CVS worked for me. What is the best way to proceed in order
to help the merge process? Is the ACT tree tested with mainline GCC?

I noticed that some patches from gcc-32.dif have been recently commited
by Richard Kenner to the mainline. I also noticed there are some RCS
tags in the sources so one must be careful not to reintroduce
them when merging into the FSF tree.

I have not succeeded yet in having tasking working with Red Hat 9
and NPTL, so I haven't tested full ACATS recently.

-- 
Laurent Guerby <guerby@acm.org>