[Ada]: failure to build on Mandrake 8.1 / x86

[Ada]: failure to build on Mandrake 8.1 / x86

[guerby]

With both ACT GNAT 3.13p and GCC 2.96 20000731 + ALT GNAT 3.13p as
base compiler I get the same abort when first using the stage1
compiler on an Ada file.

I have current CVS with no modifications (I did a re-checkout to be
sure). A few days ago on the same system I was able to bootstrap with
a few local modification. Any idea of what I get wrong or is it the
tree?

-- 
Laurent Guerby <guerby@acm.org>

$ mkdir build; cd build
$ ../gcc/configure --prefix=$HOME/work/gcc/install-cvs --enable-languages=c,ada
$ make bootstrap
...
make[3]: Entering directory `/home/guerby/work/gcc/build/gcc/ada'
stage1/xgcc -Bstage1/ -B/home/guerby/work/gcc/install-cvs/i686-pc-linux-gnu/bin/ -c -g -O2   -W -Wall -gnatpg -gnata -I- -I. -I../../../gcc/gcc/ada ../../../gcc/gcc/ada/ada.ads
+===========================GNAT BUG DETECTED==============================+
| 5.00w (20010924) (i686-pc-linux-gnu) Gigi abort, Code=321                |
| Error detected at ../../../gcc/gcc/ada/ada.ads:19:1                      |

Re: [Ada]: failure to build on Mandrake 8.1 / x86

[guerby]

Should I (and/or other volunteers :) try do debug this, or is it to known
to ACT and the patch is waiting in the queue?

-- 
Laurent Guerby <guerby@acm.org>

With both ACT GNAT 3.13p and GCC 2.96 20000731 + ALT GNAT 3.13p as
base compiler I get the same abort when first using the stage1
compiler on an Ada file.

I have current CVS with no modifications (I did a re-checkout to be
sure). A few days ago on the same system I was able to bootstrap with
a few local modification. Any idea of what I get wrong or is it the
tree?

$ mkdir build; cd build
$ ../gcc/configure --prefix=$HOME/work/gcc/install-cvs --enable-languages=c,ada
$ make bootstrap
...
make[3]: Entering directory `/home/guerby/work/gcc/build/gcc/ada'
stage1/xgcc -Bstage1/ -B/home/guerby/work/gcc/install-cvs/i686-pc-linux-gnu/bin/ -c -g -O2   -W -Wall -gnatpg -gnata -I- -I. -I../../../gcc/gcc/ada ../../../gcc/gcc/ada/ada.ads
+===========================GNAT BUG DETECTED==============================+
| 5.00w (20010924) (i686-pc-linux-gnu) Gigi abort, Code=321                |
| Error detected at ../../../gcc/gcc/ada/ada.ads:19:1                      |

Re: [Ada]: failure to build on Mandrake 8.1 / x86

[Geert Bosch]

On Thu, 11 Oct 2001 guerby@acm.org wrote:

  Should I (and/or other volunteers :) try do debug this, or is it to known
  to ACT and the patch is waiting in the queue?

I am currently in the midst of catching up with a backlog of about 150 patches,
all of which are changes made since we contributed the Ada front end.
We know the front end is in good shape without these patches and with all
of them applied. I have bootstrapped with all my patches of yesterday
applied, and results were the same as before (comparison failure after 3rd 
stage).

I use ACT GNAT 3.14a1 (GCC 2.8.1 based) as base compiler.

  -Geert

Re: [Ada]: failure to build on Mandrake 8.1 / x86

[guerby]

With your lastest commits, I'm back to bootstrap land, thanks!

-- 
Laurent Guerby <guerby@acm.org>

Re: [Ada]: failure to build on Mandrake 8.1 / x86

[Geert Bosch]

On Fri, 12 Oct 2001 guerby@acm.org wrote:

  With your lastest commits, I'm back to bootstrap land, thanks!

That is strange, since these are automatically generated files,
and there are correct rules to update them.

  -Geert

Re: [Ada]: failure to build on Mandrake 8.1 / x86

[guerby]

> That is strange, since these are automatically generated files,
> and there are correct rules to update them.

But I'd say incorrect (or a least random) time stamps from CVS update.
The only way to get this right for CVS users is to force a generation
and to copy the generated files in the build dir iff they differ from
the $(srcdir) ones using move-if-change.

-- 
Laurent Guerby <guerby@acm.org>

Re: [Ada]: failure to build on Mandrake 8.1 / x86

[Richard Henderson]

On Fri, Oct 12, 2001 at 01:29:47AM +0200, guerby@acm.org wrote:
> > That is strange, since these are automatically generated files,
> > and there are correct rules to update them.
> 
> But I'd say incorrect (or a least random) time stamps from CVS update.

Generated files should be mentioned in ./contrib/gcc_update.


r~

Re: [Ada]: failure to build on Mandrake 8.1 / x86

[Geert Bosch]

On Thu, 11 Oct 2001, Richard Henderson wrote:

  On Fri, Oct 12, 2001 at 01:29:47AM +0200, guerby@acm.org wrote:
  > > That is strange, since these are automatically generated files,
  > > and there are correct rules to update them.
  > 
  > But I'd say incorrect (or a least random) time stamps from CVS update.
  
  Generated files should be mentioned in ./contrib/gcc_update.
  
Ah. I submitted a patch for gcc_release, as requested, but I did not
get feedback on that yet. As I understand it now, the processing for
updating the autogenerated sources should go in gcc_update.
I'll make that change and resubmit the patches for gcc_release and 
gcc_update.
  
  -Geert

Re: [Ada]: failure to build on Mandrake 8.1 / x86

[Joseph S. Myers]

On Thu, 11 Oct 2001, Geert Bosch wrote:

> Ah. I submitted a patch for gcc_release, as requested, but I did not
> get feedback on that yet. As I understand it now, the processing for
> updating the autogenerated sources should go in gcc_update.
> I'll make that change and resubmit the patches for gcc_release and 
> gcc_update.

Generated files *not in CVS* should be generated from gcc_release.  The
normal method is for a bootstrap to put them in the source directory, but
if building puts them in the build directory instead then special
treatment is needed (and is currently present for the message catalogs).
If they are only generated in maintainer-mode, then the configure in
gcc_release would need to configure for maintainer-mode.

Generated files *in CVS* should be updated in CVS whenever their source
files are updated, and should be listed in gcc_update to keep the
timestamps in order.

Your patch to gcc_release also added generation of Ada tarballs.  This
should be done at the precise point at which snapshots move from the 3.0
branch to the mainline (and the scripts should be set up to generate a
separate Chill tarball as well, not because that is useful but so it
doesn't go in the core tarball).  I think it is about time to move
snapshots back to the mainline, but that should be approved by Mark and
the first mainline snapshot run manually.

-- 
Joseph S. Myers
jsm28@cam.ac.uk

Re: [Ada]: failure to build on Mandrake 8.1 / x86

[Geert Bosch]

On Fri, 12 Oct 2001, Joseph S. Myers wrote:
  
  Generated files *not in CVS* should be generated from gcc_release.  The
  normal method is for a bootstrap to put them in the source directory, but
  if building puts them in the build directory instead then special
  treatment is needed (and is currently present for the message catalogs).

I get conflicting messages on two points, could you answer these:
  * Should the autogenerated files be checked in or not?
  * Should the files be put in the source directory or in the build directory?

I'd like to do things in a way that fits with current, or better: desired,
GCC practice, but would prevent from switching between different approaches
all the time.

  -Geert

Re: [Ada]: failure to build on Mandrake 8.1 / x86

[Joseph S. Myers]

On Thu, 11 Oct 2001, Geert Bosch wrote:

> I get conflicting messages on two points, could you answer these:
>   * Should the autogenerated files be checked in or not?

They should be checked in iff they are not removed by make
maintainer-clean, which should be iff they are needed to run configure and
build the compiler in the normal way, given maintainer tools (which may
include Perl, a full GNAT installation, etc.) installed.

>   * Should the files be put in the source directory or in the build directory?

The trend seems to be for them to go in the build directory.  If generated
files go in the build directory, but need maintainer tools which users
aren't required to have installed to be built (such as Perl and Bison),
then the release script needs to be set up to copy them to the source
directory after building the compiler, and the Makefiles need to be set up
to find them there for users of releases who don't have the maintainer
tools installed.  The simpler approach, currently used for most such
files, is for them to go in the source directory, and so automatically end
up in release tarballs which are built from the source directory after a
bootstrap has been run in a separate build directory.  In this case
(generated in source directory but not checked in), the files should be
listed in .cvsignore.

-- 
Joseph S. Myers
jsm28@cam.ac.uk

Re: [Ada]: failure to build on Mandrake 8.1 / x86

[Zack Weinberg]

On Thu, Oct 11, 2001 at 09:22:26PM -0400, Geert Bosch wrote:
> On Fri, 12 Oct 2001, Joseph S. Myers wrote:
>   
>   Generated files *not in CVS* should be generated from gcc_release.  The
>   normal method is for a bootstrap to put them in the source directory, but
>   if building puts them in the build directory instead then special
>   treatment is needed (and is currently present for the message catalogs).
> 
> I get conflicting messages on two points, could you answer these:
>   * Should the autogenerated files be checked in or not?
>   * Should the files be put in the source directory or in the build directory?

At present, the generated files require fairly exotic tools (i.e. a
complete prior GNAT installation) to create, so they should remain in
CVS.  That unfortunately means they need to be generated into the
source directory, or people will forget to copy new versions into the
source directory when they make changes to the original files.

Is it possible to enforce checking in fresh derived files with the
CVS commit script?

zw

Re: [Ada]: failure to build on Mandrake 8.1 / x86

[Florian Weimer]

<guerby@acm.org> writes:

> Should I (and/or other volunteers :) try do debug this, or is it to
> known to ACT and the patch is waiting in the queue?

The community has probably caused this, so the community should sort
it out. ;-)

Have you tried to build the compileer with CC=gnatgcc?  The result
could tell us in which direction to look.

(BTW, IIRC, the email address "From: <guerby@acm.org>" is invalid
according to RFC 822, the phrase before an addr-spec cannot be empty.)