[Bug bootstrap/50822] [4.7 regression] Solaris/SPARC bootstrap failure with Sun as: invalid assembler for ios_init.cc

[Bug bootstrap/50822] [4.7 regression] Solaris/SPARC bootstrap failure with Sun as: invalid assembler for ios_init.cc

[ro at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50822

--- Comment #2 from Rainer Orth <ro at gcc dot gnu.org> 2011-10-21 13:38:33 UTC ---
Created attachment 25570
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=25570
assembler input

[Bug bootstrap/50822] [4.7 regression] Solaris/SPARC bootstrap failure with Sun as: invalid assembler for ios_init.cc

[ro at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50822

--- Comment #1 from Rainer Orth <ro at gcc dot gnu.org> 2011-10-21 13:37:47 UTC ---
Created attachment 25569
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=25569
preprocessed input

[Bug bootstrap/50822] New: [4.7 regression] Solaris/SPARC bootstrap failure with Sun as: invalid assembler for ios_init.cc

[ro at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50822

             Bug #: 50822
           Summary: [4.7 regression] Solaris/SPARC bootstrap failure with
                    Sun as: invalid assembler for ios_init.cc
    Classification: Unclassified
           Product: gcc
           Version: 4.7.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: bootstrap
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: ro@gcc.gnu.org
                CC: ebotcazou@gcc.gnu.org
              Host: sparc-sun-solaris2*
            Target: sparc-sun-solaris2*
             Build: sparc-sun-solaris2*


Between 20111014 and 20111021,Solaris/SPARC bootstrap with Sun as got broken
compiling the stage1 libstdc++:

> /var/gcc/regression/trunk/11-gcc/build/./gcc/xgcc -shared-libgcc -B/var/gcc/regression/trunk/11-gcc/build/./gcc -nostdinc++ -L/var/gcc/regression/trunk/11-gcc/build/sparc-sun-solaris2.11/libstdc++-v3/src -L/var/gcc/regression/trunk/11-gcc/build/sparc-sun-solaris2.11/libstdc++-v3/src/.libs -B/vol/gcc/sparc-sun-solaris2.11/bin/ -B/vol/gcc/sparc-sun-solaris2.11/lib/ -isystem /vol/gcc/sparc-sun-solaris2.11/include -isystem /vol/gcc/sparc-sun-solaris2.11/sys-include -I/var/gcc/regression/trunk/11-gcc/build/sparc-sun-solaris2.11/libstdc++-v3/include/sparc-sun-solaris2.11 -I/var/gcc/regression/trunk/11-gcc/build/sparc-sun-solaris2.11/libstdc++-v3/include -I/vol/gcc/src/hg/trunk/local/libstdc++-v3/libsupc++ -fno-implicit-templates -Wall -Wextra -Wwrite-strings -Wcast-qual -fdiagnostics-show-location=once -ffunction-sections -fdata-sections -frandom-seed=ios_init.lo -g -O2 -c /vol/gcc/src/hg/trunk/local/libstdc++-v3/src/ios_init.cc  -fPIC -DPIC -o .libs/ios_init.o -save-temps
/usr/bin/as: "ios_init.s", line 14123: error: symbol
"_ZN9__gnu_cxx13stdio_filebufIcSt11char_traitsIcEEC2EP6__FILESt13_Ios_Openmodej"
is used but not defined
/usr/bin/as: "ios_init.s", line 14123: error: symbol
"_ZN9__gnu_cxx13stdio_filebufIwSt11char_traitsIwEEC2EP6__FILESt13_Ios_Openmodej"
is used but not defined
/usr/bin/as: "ios_init.s": , approx line 14123: internal error:
evaluate_symbol_expression(): op 48?

In fact, both symbols are used, but not defined:

   
_ZN9__gnu_cxx13stdio_filebufIcSt11char_traitsIcEEC1EP6__FILESt13_Ios_Openmodej
=
_ZN9__gnu_cxx13stdio_filebufIcSt11char_traitsIcEEC2EP6__FILESt13_Ios_Openmodej
   
_ZN9__gnu_cxx13stdio_filebufIwSt11char_traitsIwEEC1EP6__FILESt13_Ios_Openmodej
=
_ZN9__gnu_cxx13stdio_filebufIwSt11char_traitsIwEEC2EP6__FILESt13_Ios_Openmodej

gas can cope with that, Sun as cannot.  I'll have to start a reghunt for the
culprit patch.

  Rainer

[Bug bootstrap/50822] [4.7 regression] Solaris/SPARC bootstrap failure with Sun as: invalid assembler for ios_init.cc

[ro at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50822

Rainer Orth <ro at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |4.7.0

[Bug bootstrap/50822] [4.7 regression] Solaris/SPARC bootstrap failure with Sun as: invalid assembler for ios_init.cc

[ro at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50822

Rainer Orth <ro at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2011-10-25
                 CC|                            |hubicka at gcc dot gnu.org
     Ever Confirmed|0                           |1

--- Comment #3 from Rainer Orth <ro at gcc dot gnu.org> 2011-10-25 09:06:17 UTC ---
A reghunt identified this patch

2011-10-19  Jan Hubicka  <jh@suse.cz>

       * cgraphunit.c (handle_alias_pairs): Also handle wekref with destination
       declared.
       (output_weakrefs): New function.
       * varpool.c (varpool_create_variable_alias): Handle external aliases.

as the culprit.  This seems similar to PR lto/47333 where we also have
references
to undefined symbols if the assembler used doesn't support weakrefs.

In that PR, it was possible to recreate the bug with gas if one disabled
weakref support there (either by setting the autoconf cache variable or by
modifying auto-host.h after the configure run).  I'll check if this allows to
reproduce the problem here, too.

Jan, could you please look into this quickly?  This is a bootstrap failure on 
a primary platform.

Thanks.
  Rainer

[Bug bootstrap/50822] [4.7 regression] Solaris/SPARC bootstrap failure with Sun as: invalid assembler for ios_init.cc

[ro at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50822

--- Comment #4 from Rainer Orth <ro at gcc dot gnu.org> 2011-10-25 15:04:19 UTC ---
Even with .weakref disabled in gas, gcc emits

<symbol> = <nonexistant symbol>

which gas silently accepts, while Sun as does not.  How is this supposed to
work?

  Rainer

[Bug bootstrap/50822] [4.7 regression] Solaris/SPARC bootstrap failure with Sun as: invalid assembler for ios_init.cc

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50822

--- Comment #5 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-10-27 11:28:11 UTC ---
There is a dup for this bug ... (search for 'SUN as')

[Bug bootstrap/50822] [4.7 regression] Solaris/SPARC bootstrap failure with Sun as: invalid assembler for ios_init.cc

[ro at CeBiTec dot Uni-Bielefeld.DE]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50822

--- Comment #6 from ro at CeBiTec dot Uni-Bielefeld.DE <ro at CeBiTec dot Uni-Bielefeld.DE> 2011-10-27 11:34:47 UTC ---
> --- Comment #5 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-10-27 11:28:11 UTC ---
> There is a dup for this bug ... (search for 'SUN as')

Not exactly: that one (PR lto/47333) also FAILs with gas and weakref
support disabled during configure, while gas accepts the code when I
disable weakref support.  The construct used here isn't documented in
the gas manual and seems highly dubious to me.

    Rainer

[Bug bootstrap/50822] [4.7 regression] Solaris/SPARC bootstrap failure with Sun as: invalid assembler for ios_init.cc

[hubicka at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50822

--- Comment #7 from Jan Hubicka <hubicka at gcc dot gnu.org> 2011-11-07 15:47:08 UTC ---
Hi,
the problem seems to be due to fact that C++ FE output extern aliases.  This is
not quite documented and they was simply removed by the old alias pair handling
code (I believe by accident rather than by design), while the new code confuses
them with weakrefs and attempts to output into final assembly.

I think those aliases might potentially be useful for alias analysis and ==
folding - i.e. they provide knowledge that those two variables/functions must
be equivalent. So instead of convincing C++ FE to not output them it seems
better to make cgraph code to deal with them well. For 4.8 we may start using
the alias walkers in the alias analysis and folding to actually make use of
this.

I think it only imply need to ignore them when the functions are being output.

I am testing the attached patch that ought to solve the issue.

Index: cgraphunit.c
===================================================================
--- cgraphunit.c        (revision 181033)
+++ cgraphunit.c        (working copy)
@@ -2101,13 +2101,15 @@ output_weakrefs (void)
   struct varpool_node *vnode;
   for (node = cgraph_nodes; node; node = node->next)
     if (node->alias && DECL_EXTERNAL (node->decl)
-        && !TREE_ASM_WRITTEN (node->decl))
+        && !TREE_ASM_WRITTEN (node->decl)
+       && lookup_attribute ("weakref", DECL_ATTRIBUTES (node->decl)))
       assemble_alias (node->decl,
                      node->thunk.alias ? DECL_ASSEMBLER_NAME
(node->thunk.alias)
                      : get_alias_symbol (node->decl));
   for (vnode = varpool_nodes; vnode; vnode = vnode->next)
     if (vnode->alias && DECL_EXTERNAL (vnode->decl)
-        && !TREE_ASM_WRITTEN (vnode->decl))
+        && !TREE_ASM_WRITTEN (vnode->decl)
+       && lookup_attribute ("weakref", DECL_ATTRIBUTES (vnode->decl)))
       assemble_alias (vnode->decl,
                      vnode->alias_of ? DECL_ASSEMBLER_NAME (vnode->alias_of)
                      : get_alias_symbol (vnode->decl));

[Bug bootstrap/50822] [4.7 regression] Solaris/SPARC bootstrap failure with Sun as: invalid assembler for ios_init.cc

[ro at CeBiTec dot Uni-Bielefeld.DE]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50822

--- Comment #8 from ro at CeBiTec dot Uni-Bielefeld.DE <ro at CeBiTec dot Uni-Bielefeld.DE> 2011-11-07 17:54:00 UTC ---
> I am testing the attached patch that ought to solve the issue.

A sparc-sun-solaris2.11 bootstrap with Sun as is now into stage3, so
this looks quite good so far.  I'll let you know if the bootstrap
reveals any unforseen problems.

Thanks a lot.

    Rainer

[Bug bootstrap/50822] [4.7 regression] Solaris/SPARC bootstrap failure with Sun as: invalid assembler for ios_init.cc

[ro at CeBiTec dot Uni-Bielefeld.DE]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50822

--- Comment #9 from ro at CeBiTec dot Uni-Bielefeld.DE <ro at CeBiTec dot Uni-Bielefeld.DE> 2011-11-08 12:49:18 UTC ---
The bootstrap has now completed, and although there are a couple of new
failures compared to the last successful bootstrap as of r179988, AFAICS
they are unrelated to your patch.  So unless your testing finds other
problems, we're good to go.

Thanks.
        Rainer

[Bug bootstrap/50822] [4.7 regression] Solaris/SPARC bootstrap failure with Sun as: invalid assembler for ios_init.cc

[hubicka at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50822

--- Comment #10 from Jan Hubicka <hubicka at gcc dot gnu.org> 2011-11-09 12:38:23 UTC ---
thanks, the patch ought to be safe WRT weakref targets, so I will commit it
today.

[Bug bootstrap/50822] [4.7 regression] Solaris/SPARC bootstrap failure with Sun as: invalid assembler for ios_init.cc

[hubicka at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50822

--- Comment #11 from Jan Hubicka <hubicka at gcc dot gnu.org> 2011-11-12 13:19:59 UTC ---
Author: hubicka
Date: Sat Nov 12 13:19:55 2011
New Revision: 181311

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=181311
Log:

    PR bootstrap/50822
    * cgraphunit.c (output_weakrefs): Output really only weakrefs.

Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/cgraphunit.c

[Bug bootstrap/50822] [4.7 regression] Solaris/SPARC bootstrap failure with Sun as: invalid assembler for ios_init.cc

[ro at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50822

Rainer Orth <ro at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
                URL|                            |http://gcc.gnu.org/ml/gcc-p
                   |                            |atches/2011-11/msg01583.htm
                   |                            |l
         Resolution|                            |FIXED

--- Comment #12 from Rainer Orth <ro at gcc dot gnu.org> 2011-11-23 15:51:06 UTC ---
Fixed for 4.7.0 by Jan's patch.