[Bug ada/37109] can't canadian cross ada (if host=target, but host!=build)

[Bug ada/37109] can't canadian cross ada (if host=target, but host!=build)

[tg at mirbsd dot org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37109

Thorsten Glaser <tg at mirbsd dot org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |tg at mirbsd dot org

--- Comment #4 from Thorsten Glaser <tg at mirbsd dot org> 2011-04-29 23:00:59 UTC ---
Confirmed for gcc-4.4 (4.4.6-2)

[…]
/opt/crossada/bin/m68k-linux-gnu-gcc -c -g -O2  -gnatpg -gnata -gnatwns
-nostdinc -I- -I. -Iada -I../../src/gcc/a
da -I../../src/gcc/ada/gcc-interface ../../src/gcc/ada/butil.adb -o ada/butil.o
/opt/crossada/bin/m68k-linux-gnu-gcc -c -g -O2  -gnatpg -gnata -gnatwns
-nostdinc -I- -I. -Iada -I../../src/gcc/a
da -I../../src/gcc/ada/gcc-interface ../../src/gcc/ada/casing.adb -o
ada/casing.o
mkdir -p ada/bldtools/nmake_b
rm -f ada/bldtools/nmake_b/sinfo.ads ada/bldtools/nmake_b/nmake.adt
ada/bldtools/nmake_b/xnmake.adb ada/bldtools/
nmake_b/xutil.ads ada/bldtools/nmake_b/xutil.adb
cp -p ../../src/gcc/ada/sinfo.ads ../../src/gcc/ada/nmake.adt
../../src/gcc/ada/xnmake.adb ../../src/gcc/ada/xuti
l.ads ../../src/gcc/ada/xutil.adb ada/bldtools/nmake_b
(cd ada/bldtools/nmake_b && m68k-linux-gnu-gnatmake -q xnmake && ./xnmake -b
../../nmake.adb )
./xnmake[1]: syntax error: '(' unexpected
make[2]: *** [ada/nmake.adb] Error 1
make[2]: Leaving directory `/home/tg/Xg/gcc-4.4-4.4.6/b68k/gcc'
make[1]: *** [all-gcc] Error 2
make[1]: Leaving directory `/home/tg/Xg/gcc-4.4-4.4.6/b68k'
make: *** [all] Error 2

[Bug ada/37109] can't canadian cross ada (if host=target, but host!=build)

[tg at mirbsd dot org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37109

--- Comment #5 from Thorsten Glaser <tg at mirbsd dot org> 2011-04-29 23:09:28 UTC ---
> Confirmed for gcc-4.4 (4.4.6-2)

Building from Debian/amd64 by the way. Workaround is:
copy/paste the failing command line into another shell,
then edit it (remove the m68k-linux-gnu-) and run it,
then re-run make:

1|tg@zigo:~/Xg/gcc-4.4-4.4.6/b68k/gcc $ (cd ada/bldtools/nmake_s && gnatmake -q
xnmake && ./xnmake -s ../../nma>
x-gnu-gnatmake -q xtreeprs && ./xtreeprs ../../treeprs.ads )                   
                               <
/bin/mksh: ./xtreeprs: not executable: magic 7F45
1|tg@zigo:~/Xg/gcc-4.4-4.4.6/b68k/gcc $ (cd ada/bldtools/treeprs && gnatmake -q
xtreeprs && ./xtreeprs ../../tr>

[Bug ada/37109] can't canadian cross ada (if host=target, but host!=build)

[tg at mirbsd dot org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37109

--- Comment #6 from Thorsten Glaser <tg at mirbsd dot org> 2011-04-29 23:41:11 UTC ---
I also believe this in gnattools/Makefile (generated) to be wrong:

# For cross builds of gnattools,
# put the host RTS dir first in the PATH to hide the default runtime
# files that are among the sources
# FIXME: This should be done in configure.
RTS_DIR:=$(strip $(subst \,/,$(shell gnatls -v | grep adalib )))
gnattools-cross: $(GCC_DIR)/stamp-tools

See how it calls gnatls to find out the path to adalib, even though
it’s cross-compiling? From amd64 to m68k (LP64 → ILP32), this produces
errors later on; changing gnatls to m68k-linux-gnu-gnatls fixes them.

[Bug ada/37109] can't canadian cross ada (if host=target, but host!=build)

[mikpe at it dot uu.se]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37109

Mikael Pettersson <mikpe at it dot uu.se> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |mikpe at it dot uu.se

--- Comment #7 from Mikael Pettersson <mikpe at it dot uu.se> 2011-04-30 12:40:50 UTC ---
(In reply to comment #5)
> > Confirmed for gcc-4.4 (4.4.6-2)
> 
> Building from Debian/amd64 by the way. Workaround is:
> copy/paste the failing command line into another shell,
> then edit it (remove the m68k-linux-gnu-) and run it,
> then re-run make:
> 
> 1|tg@zigo:~/Xg/gcc-4.4-4.4.6/b68k/gcc $ (cd ada/bldtools/nmake_s && gnatmake -q
> xnmake && ./xnmake -s ../../nma>
> x-gnu-gnatmake -q xtreeprs && ./xtreeprs ../../treeprs.ads )                   
>                                <
> /bin/mksh: ./xtreeprs: not executable: magic 7F45
> 1|tg@zigo:~/Xg/gcc-4.4-4.4.6/b68k/gcc $ (cd ada/bldtools/treeprs && gnatmake -q
> xtreeprs && ./xtreeprs ../../tr>

So you're trying to bootstrap gnat on m68k-linux.  Excellent.  I've tried to do
that myself recently, following my bootstrap of gnat on arm-linux-gnueabi a few
years ago.  The cross-compilation issues are real but can be worked around
manually.  The more serious problem I found is that the final gnat executables
segfault on the m68k target.  Even a trivial 'gnat' w/o options which should
just print a few lines of help text segfaults.  So I fear there's a runtime
issue or miscompilation somewhere.

[Bug ada/37109] can't canadian cross ada (if host=target, but host!=build)

[tg at mirbsd dot org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37109

--- Comment #8 from Thorsten Glaser <tg at mirbsd dot org> 2011-04-30 13:36:17 UTC ---
(In reply to comment #7)

> manually.  The more serious problem I found is that the final gnat executables
> segfault on the m68k target.

Even with no argument, yes. It’s in:

hostparm.ads:   Normalized_CWD : constant String := "." & Direct_Separator;

This in end calls memcpy with a length of about 2 GiB. I had thought this
to be an issue of bootstrapping from amd64, which platform did you use?

[Bug ada/37109] can't canadian cross ada (if host=target, but host!=build)

[mikpe at it dot uu.se]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37109

--- Comment #9 from Mikael Pettersson <mikpe at it dot uu.se> 2011-04-30 13:49:37 UTC ---
(In reply to comment #8)
> (In reply to comment #7)
> 
> > manually.  The more serious problem I found is that the final gnat executables
> > segfault on the m68k target.
> 
> Even with no argument, yes. It’s in:
> 
> hostparm.ads:   Normalized_CWD : constant String := "." & Direct_Separator;
> 
> This in end calls memcpy with a length of about 2 GiB. I had thought this
> to be an issue of bootstrapping from amd64, which platform did you use?

i686-linux.

[Bug ada/37109] can't canadian cross ada (if host=target, but host!=build)

[tg at mirbsd dot org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37109

--- Comment #10 from Thorsten Glaser <tg at mirbsd dot org> 2011-04-30 13:56:23 UTC ---
(In reply to comment #9)
> (In reply to comment #8)

> > to be an issue of bootstrapping from amd64, which platform did you use?
> 
> i686-linux.

Ok, then I can continue with trying from amd64. *THANKS*!  That saves me
quite a lot of additional effort for nothing.

At a first glance, there’s a lot of system-linux-*.ads but none for m68k.
Should I have a look in that area, or is that not an issue at all?

(I know almost nothing about Ada, except it’s written with a capital A,
like Lovelace.)

[Bug ada/37109] can't canadian cross ada (if host=target, but host!=build)

[mikpe at it dot uu.se]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37109

--- Comment #11 from Mikael Pettersson <mikpe at it dot uu.se> 2011-04-30 15:09:32 UTC ---
(In reply to comment #10)
> At a first glance, there’s a lot of system-linux-*.ads but none for m68k.
> Should I have a look in that area, or is that not an issue at all?

It will be an issue, but I got the segfaults even with such a file, so I think
the problem is more fundamental than that.

What you should do is make system-linux-m68k.ads a clone of e.g.
system-linux-ppc.ads (both big-endian 32-bitters), set the ZCX variables to
false, and disable stack check probes.  Then update gcc-interface/Makefile.in
with a case that matches m68k-linux and does what other linux targets do, but
selecting system-linux-m68k.ads and setting EH_MECHANISM= (the empty string). 
This enables SJLJ EH which is simpler and adequate for initial bootstrap.  Once
gnat works on m68k you can try enabling ZCX EH (which will need dwarf2
unwinding).

(I can't post a patch due to delays at the FSF wrt my employer's disclaimer.)

[Bug ada/37109] can't canadian cross ada (if host=target, but host!=build)

[tg at mirbsd dot org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37109

--- Comment #12 from Thorsten Glaser <tg at mirbsd dot org> 2011-04-30 15:20:08 UTC ---
(In reply to comment #11)

> It will be an issue, but I got the segfaults even with such a file, so I think
> the problem is more fundamental than that.

OK, I’ll try.

> What you should do is make system-linux-m68k.ads a clone of e.g.

I’ve cloned system.ads and made changes while looking at i386, mipseb, armeb
and macppc – let’s see how it goes from there.

> system-linux-ppc.ads (both big-endian 32-bitters), set the ZCX variables to
> false, and disable stack check probes.  Then update gcc-interface/Makefile.in
> with a case that matches m68k-linux and does what other linux targets do, but
> selecting system-linux-m68k.ads and setting EH_MECHANISM= (the empty string). 
> This enables SJLJ EH which is simpler and adequate for initial bootstrap.  Once
> gnat works on m68k you can try enabling ZCX EH (which will need dwarf2
> unwinding).

Will do if that fails. Thanks for the ideas.

> (I can't post a patch due to delays at the FSF wrt my employer's disclaimer.)

That’s luckily no problem for me, I have papers with the FSF standing,
even explicitly for GCC already.

[Bug ada/37109] can't canadian cross ada (if host=target, but host!=build)

[tg at mirbsd dot org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37109

--- Comment #13 from Thorsten Glaser <tg at mirbsd dot org> 2011-04-30 15:31:06 UTC ---
Ah well. I found out that Ada doesn’t compile if you have a style error õÕ
and that “make clean” doesn’t clean all stampfiles… got a crosscompiler now,
let’s see whether it can cross-build a native compiler and that then works.

Should we open up another bugreport for _that_ (port GNAT to GNU/Linux/m68k)
though, since it doesn’t really belong here?

[Bug ada/37109] can't canadian cross ada (if host=target, but host!=build)

[mikpe at it dot uu.se]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37109

--- Comment #14 from Mikael Pettersson <mikpe at it dot uu.se> 2011-04-30 15:40:54 UTC ---
(In reply to comment #13)
> Should we open up another bugreport for _that_ (port GNAT to GNU/Linux/m68k)
> though, since it doesn’t really belong here?

Yes.

[Bug ada/37109] can't canadian cross ada (if host=target, but host!=build)

[tg at mirbsd dot org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37109

--- Comment #15 from Thorsten Glaser <tg at mirbsd dot org> 2011-04-30 15:48:47 UTC ---
(In reply to comment #14)
> (In reply to comment #13)
> > Should we open up another bugreport for _that_ (port GNAT to GNU/Linux/m68k)
> > though, since it doesn’t really belong here?
> 
> Yes.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48835

[Bug ada/37109] can't canadian cross ada (if host=target, but host!=build)

[ebotcazou at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=37109

Eric Botcazou <ebotcazou at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
                 CC|                            |ebotcazou at gcc dot gnu.org
         Resolution|---                         |FIXED
   Target Milestone|---                         |5.0

--- Comment #16 from Eric Botcazou <ebotcazou at gcc dot gnu.org> ---
Presumably.

[Bug ada/37109] can't canadian cross ada (if host=target, but host!=build)

[christian dot joensson at gmail dot com]

------- Comment #3 from christian dot joensson at gmail dot com  2009-05-13 10:25 -------
sorry, I meant gnatbind instead of gnatlib...


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37109

[Bug ada/37109] can't canadian cross ada (if host=target, but host!=build)

[christian dot joensson at gmail dot com]

------- Comment #2 from christian dot joensson at gmail dot com  2009-05-13 10:24 -------
what is the current situation concerning gnatlib? is that passed explicitly
from stage to stage?


-- 

christian dot joensson at gmail dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |christian dot joensson at
                   |                            |gmail dot com


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37109

[Bug ada/37109] can't canadian cross ada (if host=target, but host!=build)

[charlet at gcc dot gnu dot org]

------- Comment #1 from charlet at gcc dot gnu dot org  2008-08-13 09:28 -------
This was broken by the following change AFAICT (rev 121082):

<<
2007-01-23  Richard Guenther  <rguenther@suse.de>

        PR bootstrap/30541
        * Makefile.def (flags_to_pass): Add GNATBIND and GNATMAKE.
        * Makefile.tpl (GNATBIND): Substitute it.
        (GNATMAKE): Likewise.
        (POSTSTAGE1_FLAGS_TO_PASS): Pass staged GNATBIND instead
        of STAGE_PREFIX.
        * Makefile.in: Regenerate.
        * configure: Regenerate.

        config/
        * acx.m4 (ACX_PROG_GNAT): Check for gnatmake.

        libada/
        * Makefile.in (STAGE_PREFIX): Do not define.
        (FLAGS_TO_PASS): Do not pass STAGE_PREFIX.

        gnattools/
        * Makefile.in (STAGE_PREFIX): Do not define.

        ada/
        * Make-lang.in: Replace invocations of gnatmake with $(GNATMAKE).
        (gnatboot2): Pass staged GNATMAKE instead of STAGE_PREFIX.
        (gnatboot3): Likewise.
        (GNATBIND): Do not define.
        * Makefile.in (GNATBIND): Do not define.
>>

Before that, we were using 'gnatmake' instead of '$(GNATMAKE)', which is what's
causing troubles here.

Arno


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37109