Re: Building the Ada compiler... is http://gcc.gnu.org/install/build.html correct?

Re: Building the Ada compiler... is http://gcc.gnu.org/install/build.html correct?

[Frank Ecke]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Upon Nov 5, 2003, Rainer Orth spake unto us thuswise, saying:

> Currently I try to get it working on IRIX 6.2/6.5.  While the compiler
> proper bootstraps with the change suggested in PR middle-end/6552, make
> gnatlib_and_tools only builds the runtime library successfully.  The
> freshly built gnatmake prints

> __exc_unwind: Error: pc(0x1021075c) not in any .debug_frame table

> and just hangs.

I have experienced the same with GCC 3.3.3 and 3.3.4; only in my
case, the freshly built  gnatmake  doesn't  hang  but  causes  a
segmentation fault and produces a core dump:

../../gnatmake -c -I../rts -I. -I/usr/people/franke/install/srcdir/gcc/ada gnatchop --GCC="../../xgcc -B../../ -g      -gnatpg -gnata"
__exc_unwind: Error: pc(0x102aa700) not in any .debug_frame table
__exc_unwind: Error: pc(0x102aa70c) not in any .debug_frame table
__exc_unwind: Error: pc(0x102aa70c) not in any .debug_frame table
__exc_unwind: Error: pc(0x102aa70c) not in any .debug_frame table
__exc_unwind: Error: pc(0x102aa70c) not in any .debug_frame table
[snip]
__exc_unwind: Error: pc(0x102aa70c) not in any .debug_frame table
__exc_unwind: Error: pc(0x102aa70c) not in any .debug_frame table
make[2]: *** [../../gnatchop] Segmentation fault (core dumped)
make[2]: Leaving directory `/usr/people/franke/install/objdir/gcc/ada/tools'
make[1]: *** [gnattools2] Error 2
make[1]: Leaving directory `/usr/people/franke/install/objdir/gcc/ada'
make: *** [gnattools] Error 2
(glanzer 718) $

In  addition  to  that, I've found the following in the system's
log:

Jul 17 16:36:31 4A:glanzer unix: WARNING: Process gnatmake pid 2171 referenced bad addr 0xc0000fc00050fa38

For my little SGI, this is clearly out of reach, :-(

> This error is from the IRIX 6 libexc.so.  I have no idea
> what to do about this and haven't yet tried debugging what's going on
> here.

Has anyone been able to shed some light on this issue in the
meantime?


Regards,

Frank

- --

Frank Ecke <franke@euro-telematik.de>, see pgpkeys.mit.edu:11371
Fingerprint = 2CCD 7D50 E07E 557E EA2F  7454 068C B20A 20FF C83D

Bradley's Bromide:

If  computers  get  too  powerful,  we  can organize them into a
committee---that will do them in.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.1 (IRIX64)

iD8DBQFA+USbBoyyCiD/yD0RAkeJAJ43tNNb+aHcRNS+3QNNTlC1cfCDwACgge0P
t7i7x0AnfsusRohBtnCxVR0=
=XLFu
-----END PGP SIGNATURE-----

Re: Building the Ada compiler... is http://gcc.gnu.org/install/build.html correct?

[Rainer Orth]

Frank Ecke writes:

> > __exc_unwind: Error: pc(0x1021075c) not in any .debug_frame table
> 
> > and just hangs.
> 
> I have experienced the same with GCC 3.3.3 and 3.3.4; only in my
> case, the freshly built  gnatmake  doesn't  hang  but  causes  a
> segmentation fault and produces a core dump:
[...]
> Has anyone been able to shed some light on this issue in the
> meantime?

It doesn't occur in GCC 3.4 and up, and since nobody is doing any Ada work
on the 3.3 branch, I suggest you move to 3.4 instead.

	Rainer

-----------------------------------------------------------------------------
Rainer Orth, Faculty of Technology, Bielefeld University

Re: Building the Ada compiler... is http://gcc.gnu.org/install/build.html correct?

[Frank Ecke]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Rainer,

Upon Jul 19, 2004, ye spake unto me thuswise, saying:

> It  doesn't  occur  in GCC 3.4 and up, and since nobody is doing
> any Ada work on the 3.3  branch,  I  suggest  you  move  to  3.4
> instead.

I  did,  but  3.4.1 brought no real alleviation.  For one thing,
it's much  harder  now  to  get  going  since  3.4  expects  the
bootstrap  compiler  to be Ada-aware already (3.3 allowed you to
have two different bootstrap gcc's: one for compiling C and  one
for  compiling Ada).  Staring from scratch, as i did, i only had
gcc 3.1 (C, C++, *no* Ada) and a binary distribution of GNAT 3.4
at  my  disposal.  The latter came with a gnatgcc and could thus
handle Ada.

   The  problem  now  was  that both gcc and gnatgcc were needed
during  the  bootstrap,  but  i  found  myself  unable  to  tell
configure  that  gcc  is supposed to be used for C (and C++) and
that gnatgcc is for Ada.  Because as soon as Ada  needed  to  be
compiled,  gcc  would  whine, and i had to restart the bootstrap
with

   $ make bootstrap CC=gnatgcc

This  went  fine  until  the  bootstrap went back to compiling C
because now gnatgcc would complain (actually, it did  compile  a
few  C  files,  but  bailed  out  soon after that).  So i had to
restart the bootstrap once again, this time with

   $ make bootstrap CC=gcc

This went fine until it was time to again compile some Ada ...

   It  worked  in  the  end,  after  saying,  alternately, "make
bootstrap CC=gnatgcc" or "make bootstrap  CC=gcc"  a  couple  of
times,  but  it  would  be  nice  if it was possible, during the
configuration step, to state that during the bootstrap, compiler
A  is  to  be used to compile C (and C++) and that compiler B is
deemed to be used whenever Ada code needs to be compiled.

   After the bootstrap, at stage 3, the following happened

stage2/xgcc -Bstage2/ -B/usr/local2/mips-sgi-irix6.2/bin/ -c -g -O2      -gnatpg -gnata -I- -I. -Iada -I../../srcdir/gcc/ada ../../srcdir/gcc/ada/ada.ads -o ada/ada.o
xgcc: Internal error: Segmentation fault (program gnat1)
Please submit a full bug report.
See <URL:http://gcc.gnu.org/bugs.html> for instructions.
make[2]: *** [ada/ada.o] Error 1
make[2]: Leaving directory `/usr/people/franke/install/objdir/gcc'
make[1]: *** [stage3_build] Error 2
make[1]: Leaving directory `/usr/people/franke/install/objdir/gcc'
make: *** [bootstrap] Error 2

The   problem   is   known   to   occur   in   3.4.0   as   well
(http://gcc.gnu.org/ml/gcc-bugs/2004-01/msg01929.html).      Any
ideas?


Regards,

Frank

- --

Frank Ecke <franke@euro-telematik.de>, see pgpkeys.mit.edu:11371
Fingerprint = 2CCD 7D50 E07E 557E EA2F  7454 068C B20A 20FF C83D

The  following statement is not true.  The previous statement is
true.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.1 (IRIX64)

iD8DBQFBHkbBBoyyCiD/yD0RAiDSAJ9IH/cD1Og61JU4uFIHRbgHbtDqiwCgnsIl
Mk+KAa064br5QLqs/Xt8M/M=
=Q3Pz
-----END PGP SIGNATURE-----

Re: Building the Ada compiler... is http://gcc.gnu.org/install/build.html correct?

[Toon Moene]

Arnaud Charlet wrote:

 > Someone else wrote:

>>>Any news on having a documented (in the Ada install documentation)
>>>collection of contributed bootstrap binaries (now including C compilers as
>>>well as gnat1 and gnatbind)

> It appears that we were too optimistic about providing binaries, and
> that this requires too much work/set up on our side.

Hmmm, I just did:

apt-get install gnat-3.3

on my Debian testing system (powerpc-unknown-linux-gnu) and it Just Worked.

The only reason you see no Ada test suite results from me is that for 
some reason that I do not understand the test suite isn't run.

Cheers,

-- 
Toon Moene - mailto:toon@moene.indiv.nluug.nl - phoneto: +31 346 214290
Saturnushof 14, 3738 XG  Maartensdijk, The Netherlands
Maintainer, GNU Fortran 77: http://gcc.gnu.org/onlinedocs/g77_news.html
GNU Fortran 95: http://gcc.gnu.org/fortran/ (under construction)

Re: Building the Ada compiler... is http://gcc.gnu.org/install/build.html correct?

[Rainer Orth]

Arnaud Charlet writes:

> If this is still felt as a necessity, we would be glad to host the
> binaries contributed by other people on the libre site, although I
> guess this could also be done on the gcc.gnu.org site directly (whichever
> way is more convenient).

I think hosting them on gcc.gnu.org is the obvious solution for FSF
GCC/GNAT, but that's for the maintainers of that machine and the SC to
decide.

> Given that there have been a couple of releases with Ada included, and
> that more people have a pre-existing Ada set up, this issue may not be 
> as important as it used to be.

It is still quite hard to get there on non-GNU/Linux platforms: you have to
start bootstrapping from old GNAT 3.x binaries (like GNAT 3.11p for IRIX
5.3 to get to IRIX 6.[25], or GNAT 3.12p for Tru64 UNIX V4.0D to get to
V4.0F/V5.1B).  This gets especially convoluted when the GCC included in
those distributions miscompiles GCC 3.3/3.4 ;-(

Solaris (both SPARC and x86) is easier, fortunately, since ACT provides
reasonably recent binary distributions for SPARC and the x86 binary
distribution for Solaris 2.6 works reasonably well on newer releases due to
Sun's guarantee of binary compatibility.

I still think it shouldn't be upon the casual (or not so casual) end user
to go through all the necessary contortions.

> The Ada front-end is not at a stage where binary compatibility can be ensured
> for lots of various reasons, so I think the above concern is premature.
> Also, the provided binaries will only be provided to help
> people bootstrap gnat, they do not have to be binary compatible to
> build gnat properly.

Ok, I'll formally submit the patch shortly, together with test results.

> > * Unlike the other backends, GNAT isn't currently relocatable: if GCC/GNAT
> >   is built with the default prefix (/usr/local), but installed with make
> >   DESTDIR=<somedir>, I get the following error:
> 
> You simply need to set the environment variable GNAT_ROOT to <somedir>/usr/local

I consider this a workaround only, not a solution.  I'll submit a PR for
this issue.

	Rainer

-----------------------------------------------------------------------------
Rainer Orth, Faculty of Technology, Bielefeld University

Re: Building the Ada compiler... is http://gcc.gnu.org/install/build.html correct?

[Arnaud Charlet]

> > Any news on having a documented (in the Ada install documentation)
> > collection of contributed bootstrap binaries (now including C compilers as
> > well as gnat1 and gnatbind)

It appears that we were too optimistic about providing binaries, and
that this requires too much work/set up on our side.
     
If this is still felt as a necessity, we would be glad to host the
binaries contributed by other people on the libre site, although I
guess this could also be done on the gcc.gnu.org site directly (whichever
way is more convenient).

Given that there have been a couple of releases with Ada included, and
that more people have a pre-existing Ada set up, this issue may not be 
as important as it used to be.

> * What are the consequences of the above workaround on IRIX?  It seems to
>   break binary compatibility, so if the underlying middle-end bug is fixed
>   and the workaround reverted, we have a problem.

The Ada front-end is not at a stage where binary compatibility can be ensured
for lots of various reasons, so I think the above concern is premature.
Also, the provided binaries will only be provided to help
people bootstrap gnat, they do not have to be binary compatible to
build gnat properly.

> * Unlike the other backends, GNAT isn't currently relocatable: if GCC/GNAT
>   is built with the default prefix (/usr/local), but installed with make
>   DESTDIR=<somedir>, I get the following error:

You simply need to set the environment variable GNAT_ROOT to <somedir>/usr/local

Arno

Re: Building the Ada compiler... is http://gcc.gnu.org/install/build.html correct?

[Rainer Orth]

Joseph S. Myers writes:

> Any news on having a documented (in the Ada install documentation)
> collection of contributed bootstrap binaries (now including C compilers as
> well as gnat1 and gnatbind), as proposed by Robert Dewar in
> <http://gcc.gnu.org/ml/gcc/2001-02/msg01233.html> (and repeated in various
> subsequent messages)?

I've already built GCC 3.3.2 (C and Ada only), with --disable-nls
--disable-multilib and no additional configure options for

* Solaris 8/SPARC (sparc-sun-solaris2.8)

* Solaris 9/x86 (i386-pc-solaris2.9)

* Tru64 UNIX V4.0F (alpha-dec-osf4.0f)

* Tru64 UNIX V5.1B (alpha-dec-osf5.1b)

Currently I try to get it working on IRIX 6.2/6.5.  While the compiler
proper bootstraps with the change suggested in PR middle-end/6552, make
gnatlib_and_tools only builds the runtime library successfully.  The
freshly built gnatmake prints

__exc_unwind: Error: pc(0x1021075c) not in any .debug_frame table

and just hangs.  This error is from the IRIX 6 libexc.so.  I have no idea
what to do about this and haven't yet tried debugging what's going on
here.  Instead, I'm getting close to have C and Ada binaries from current
mainline for

* IRIX 6.2 (mips-sgi-irix6.2)

* IRIX 6.5 (mips-sgi-irix6.5)

While I would like to have a common base version for all platforms, the
mainline binaries will have to do if no solution for 3.3.x comes up.

There are some open questions:

* What are the consequences of the above workaround on IRIX?  It seems to
  break binary compatibility, so if the underlying middle-end bug is fixed
  and the workaround reverted, we have a problem.

* I don't really want to host the resulting binary distributions.  They
  should be on either gcc.gnu.org or libre.act-europe.fr.

* Unlike the other backends, GNAT isn't currently relocatable: if GCC/GNAT
  is built with the default prefix (/usr/local), but installed with make
  DESTDIR=<somedir>, I get the following error:

$ <somedir>/usr/local/bin/gcc -c conftest.adb    
fatal error, run-time library not installed correctly
cannot locate file system.ads
compilation abandoned

  While this can be worked around using the .gnat_wrapper script
  distributed with ACT's binary distributions, this shouldn't be necessary
  in the first place.

	Rainer

-----------------------------------------------------------------------------
Rainer Orth, Faculty of Technology, Bielefeld University

Re: Building the Ada compiler... is http://gcc.gnu.org/install/build.html correct?

[Joseph S. Myers]

On Wed, 5 Nov 2003, Rainer Orth wrote:

> As I've indicated before, the current support for building with different
> Ada and C compilers should be removed in favor of requiring a gcc that
> includes GNAT if Ada support is desired.

Any news on having a documented (in the Ada install documentation)
collection of contributed bootstrap binaries (now including C compilers as
well as gnat1 and gnatbind), as proposed by Robert Dewar in
<http://gcc.gnu.org/ml/gcc/2001-02/msg01233.html> (and repeated in various
subsequent messages)?

-- 
Joseph S. Myers
jsm@polyomino.org.uk

Re: Building the Ada compiler... is http://gcc.gnu.org/install/build.html correct?

[Rainer Orth]

Christian Joensson <christian@j-son.org> writes:

> Just a ping here... is the information concerning building the Ada
> compiler in http://gcc.gnu.org/install/build.html correct?
> 
> More specifically, is
[...]
>          cd srcdir/gcc/ada
>          touch treeprs.ads [es]info.h nmake.ad[bs]

This part should be unnecessary.  gcc_update takes care of this when
building from CVS.

Besides, there are other errors:

* It isn't necessary to specify --enable-languages: if you have an
  Ada-capable gcc as a bootstrap compiler, the ada backend will be built
  automatically.

* Nowadays gnatmake *is* required since it is used to build xgnatug.

* While make gnatlib_and_tools doesn't work in parallel, parallel bootstrap
  itself works fine.

As I've indicated before, the current support for building with different
Ada and C compilers should be removed in favor of requiring a gcc that
includes GNAT if Ada support is desired.

	Rainer

Building the Ada compiler... is http://gcc.gnu.org/install/build.html correct?

[Christian Joensson]

Just a ping here... is the information concerning building the Ada
compiler in http://gcc.gnu.org/install/build.html correct?

More specifically, is

         cd objdir
         srcdir/configure --enable-languages=c,ada
         cd srcdir/gcc/ada
         touch treeprs.ads [es]info.h nmake.ad[bs]
         cd objdir
         make bootstrap
         cd gcc
         make gnatlib_and_tools
         cd ..
     
still required?

Cheers,

/ChJ