[Bug target/51934] New: FAIL: g++.dg/torture/pr51344.C -O0 (test for excess errors) on powerpc*-*-*

[Bug target/51934] New: FAIL: g++.dg/torture/pr51344.C -O0 (test for excess errors) on powerpc*-*-*

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51934

             Bug #: 51934
           Summary: FAIL: g++.dg/torture/pr51344.C  -O0  (test for excess
                    errors) on powerpc*-*-*
    Classification: Unclassified
           Product: gcc
           Version: 4.7.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: dominiq@lps.ens.fr
                CC: ktietz@gcc.gnu.org, pthaugen@us.ibm.com
              Host: powerpc*-*-*
            Target: powerpc*-*-*
             Build: powerpc*-*-*


The test g++.dg/torture/pr51344.C fails on powerpc*-*-* (see
http://gcc.gnu.org/ml/gcc-testresults/2012-01/msg02019.html )

FAIL: g++.dg/torture/pr51344.C  -O0  (test for excess errors)
FAIL: g++.dg/torture/pr51344.C  -O1  (test for excess errors)
FAIL: g++.dg/torture/pr51344.C  -O2  (test for excess errors)
FAIL: g++.dg/torture/pr51344.C  -O3 -fomit-frame-pointer  (test for excess
errors)
FAIL: g++.dg/torture/pr51344.C  -O3 -g  (test for excess errors)
FAIL: g++.dg/torture/pr51344.C  -Os  (test for excess errors)
FAIL: g++.dg/torture/pr51344.C  -O2 -flto -fno-use-linker-plugin
-flto-partition=none  (test for excess errors)
FAIL: g++.dg/torture/pr51344.C  -O2 -flto -fuse-linker-plugin
-fno-fat-lto-objects  (test for excess errors)

with (at least on powerpc-apple-darwin9)

[karma] f90/bug% /opt/gcc/gcc4.7w/bin/g++ -c
/opt/gcc/work/gcc/testsuite/g++.dg/torture/pr51344.C
/opt/gcc/work/gcc/testsuite/g++.dg/torture/pr51344.C:7:58: warning: 'cdecl'
attribute directive ignored [-Wattributes]
/opt/gcc/work/gcc/testsuite/g++.dg/torture/pr51344.C:7:58: warning: 'cdecl'
attribute directive ignored [-Wattributes]

Is the attribute 'cdecl' supposed to exist on powerpc?

[Bug target/51934] FAIL: g++.dg/torture/pr51344.C -O0 (test for excess errors) on powerpc*-*-*

[mikpe at it dot uu.se]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51934

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

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

--- Comment #1 from Mikael Pettersson <mikpe at it dot uu.se> 2012-01-21 16:10:15 UTC ---
Also broken on 4.6 and 4.5 branches (don't know yet about 4.4, but the
offending change was backported to it too), see:

http://gcc.gnu.org/ml/gcc-testresults/2012-01/msg02026.html
http://gcc.gnu.org/ml/gcc-testresults/2012-01/msg01992.html
http://gcc.gnu.org/ml/gcc-testresults/2012-01/msg01945.html

[Bug target/51934] FAIL: g++.dg/torture/pr51344.C -O0 (test for excess errors) on powerpc*-*-*

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51934

--- Comment #2 from Dominique d'Humieres <dominiq at lps dot ens.fr> 2012-01-21 16:15:52 UTC ---
> Also broken on 4.6 and 4.5 branches ...

Mikael can you confirm that the failures are due to the warning "'cdecl'
attribute directive ignored"?

[Bug target/51934] FAIL: g++.dg/torture/pr51344.C -O0 (test for excess errors) on powerpc*-*-*

[mikpe at it dot uu.se]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51934

--- Comment #3 from Mikael Pettersson <mikpe at it dot uu.se> 2012-01-21 21:15:47 UTC ---
(In reply to comment #2)
> > Also broken on 4.6 and 4.5 branches ...
> 
> Mikael can you confirm that the failures are due to the warning "'cdecl'
> attribute directive ignored"?

Yes, the 4.6 failures are caused by that cdecl attribute ignored warning.

BTW, 4.6 fails on sparc64-linux in the same way.

[Bug target/51934] FAIL: g++.dg/torture/pr51344.C -O0 (test for excess errors) on powerpc*-*-*

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51934

Dominique d'Humieres <dominiq at lps dot ens.fr> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2012-01-21
     Ever Confirmed|0                           |1

--- Comment #4 from Dominique d'Humieres <dominiq at lps dot ens.fr> 2012-01-21 22:22:50 UTC ---
> Yes, the 4.6 failures are caused by that cdecl attribute ignored warning.
>
> BTW, 4.6 fails on sparc64-linux in the same way.

Thanks, setting the PR to NEW.

[Bug target/51934] FAIL: g++.dg/torture/pr51344.C -O0 (test for excess errors) on powerpc*-*-*

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51934

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jakub at gcc dot gnu.org

--- Comment #5 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-01-21 22:35:58 UTC ---
The test just should use some attribute that is common to all targets, like
__attribute__((noinline)) or similar, unless it didn't fail with that attribute
before the fix.  If it didn't, then the test needs to be limited to targets
that support it.

[Bug target/51934] FAIL: g++.dg/torture/pr51344.C -O0 (test for excess errors) on powerpc*-*-*

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51934

--- Comment #6 from Dominique d'Humieres <dominiq at lps dot ens.fr> 2012-01-21 23:15:31 UTC ---
> The test just should use some attribute that is common to all targets, like
> __attribute__((noinline)) or similar, unless it didn't fail with that attribute
> before the fix.  If it didn't, then the test needs to be limited to targets
> that support it.

The attribute format seems to do the trick: it hangs on 4.6.2 and compiles with
recent trunk.

--- ../_clean/gcc/testsuite/g++.dg/torture/pr51344.C    2012-01-20
11:40:19.000000000 +0100
+++ gcc/testsuite/g++.dg/torture/pr51344.C    2012-01-22 00:13:16.000000000
+0100
@@ -4,7 +4,7 @@ class A;
 template <class T>
 class B
 {
-  friend __attribute__((cdecl)) A& operator >>(A& a, B& b)
+  friend __attribute__((format)) A& operator >>(A& a, B& b)
   {
     return a;
   }

[Bug target/51934] FAIL: g++.dg/torture/pr51344.C -O0 (test for excess errors) on powerpc*-*-*

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51934

--- Comment #7 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-01-21 23:30:25 UTC ---
You mean errors out?  format attribute must have 3 arguments.
Try leaf, or nothrow etc. attributes instead, format is a bad idea for a method
that isn't printf/scanf like.

[Bug target/51934] FAIL: g++.dg/torture/pr51344.C -O0 (test for excess errors) on powerpc*-*-*

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51934

--- Comment #8 from Dominique d'Humieres <dominiq at lps dot ens.fr> 2012-01-21 23:58:19 UTC ---
> You mean errors out?  format attribute must have 3 arguments.
> Try leaf, or nothrow etc. attributes instead, format is a bad idea for a method
> that isn't printf/scanf like.

Well, format was the first attribute I have found with your requirements:
common to all platforms, hangs before the commit and passes after it (aligned,
cold, flatten, malloc, noclone, noinline,  nothrow, pure don't work; ifunc ,
no_instrument_function, noreturn, no_split_stack give warnings or errors).
nonnull meets also the above criteria, but is also supposed to use arguments.
If someone has more patience than me, be my guest!

Now if its use without arguments should yields an error, it is another bug.

[Bug target/51934] FAIL: g++.dg/torture/pr51344.C -O0 (test for excess errors) on powerpc*-*-*

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51934

Dominique d'Humieres <dominiq at lps dot ens.fr> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |Greta.Yorsh at arm dot com

--- Comment #9 from Dominique d'Humieres <dominiq at lps dot ens.fr> 2012-01-23 12:36:59 UTC ---
*** Bug 51963 has been marked as a duplicate of this bug. ***

[Bug target/51934] FAIL: g++.dg/torture/pr51344.C -O0 (test for excess errors) due to "cdecl attribute ignored" warning

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51934

Dominique d'Humieres <dominiq at lps dot ens.fr> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Target|powerpc*-*-*                |powerpc*-*-* arm-none-eabi
            Summary|FAIL:                       |FAIL:
                   |g++.dg/torture/pr51344.C    |g++.dg/torture/pr51344.C
                   |-O0  (test for excess       |-O0  (test for excess
                   |errors) on powerpc*-*-*     |errors)  due to "cdecl
                   |                            |attribute ignored" warning

--- Comment #10 from Dominique d'Humieres <dominiq at lps dot ens.fr> 2012-01-23 12:40:36 UTC ---
Changed the summary.

[Bug target/51934] FAIL: g++.dg/torture/pr51344.C -O0 (test for excess errors) due to "cdecl attribute ignored" warning

[jason at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51934

--- Comment #11 from Jason Merrill <jason at gcc dot gnu.org> 2012-01-23 14:43:30 UTC ---
Author: jason
Date: Mon Jan 23 14:43:25 2012
New Revision: 183434

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=183434
Log:
    PR target/51934
    * g++.dg/torture/pr51344.C: Use noreturn instead of cdecl.

Modified:
    trunk/gcc/testsuite/ChangeLog
    trunk/gcc/testsuite/g++.dg/torture/pr51344.C

[Bug target/51934] FAIL: g++.dg/torture/pr51344.C -O0 (test for excess errors) due to "cdecl attribute ignored" warning

[jason at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51934

Jason Merrill <jason at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
                 CC|                            |jason at gcc dot gnu.org
         Resolution|                            |FIXED

--- Comment #12 from Jason Merrill <jason at gcc dot gnu.org> 2012-01-23 14:50:00 UTC ---
Should be fixed now.

[Bug target/51934] FAIL: g++.dg/torture/pr51344.C -O0 (test for excess errors) due to "cdecl attribute ignored" warning

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51934

--- Comment #13 from Dominique d'Humieres <dominiq at lps dot ens.fr> 2012-01-23 14:55:08 UTC ---
> Log:
>     PR target/51934
>     * g++.dg/torture/pr51344.C: Use noreturn instead of cdecl.
>
> Should be fixed now.

The use of 'noreturn' yields a warning with g++ 4.6.2 and not an infinite loop:

[macbook] f90/bug% time g++-fsf-4.6 -c pr51344_db.C
pr51344_db.C: In function 'A& operator>>(A&, B<T>&)':
pr51344_db.C:9:12: warning: function declared 'noreturn' has a 'return'
statement [enabled by default]
0.012u 0.024s 0:00.93 3.2%    0+0k 3+13io 0pf+0w

so IMO the test is no longer testing what has been fixed (see comment #8 for a
list of what I have tried).

[Bug target/51934] FAIL: g++.dg/torture/pr51344.C -O0 (test for excess errors) due to "cdecl attribute ignored" warning

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51934

--- Comment #14 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-01-23 15:10:42 UTC ---
Well, if it hangs before the fix with the noreturn attribute, then it is
trivial to replace the return a; with for (;;);

[Bug target/51934] FAIL: g++.dg/torture/pr51344.C -O0 (test for excess errors) due to "cdecl attribute ignored" warning

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51934

--- Comment #15 from Dominique d'Humieres <dominiq at lps dot ens.fr> 2012-01-23 15:19:37 UTC ---
> Well, if it hangs before the fix with the noreturn attribute, then it is
> trivial to replace the return a; with for (;;);

It does not hang:

macbook] f90/bug% cat pr51344_db.C
/* { dg-do compile } */
class A;

template <class T>
class B
{
  friend __attribute__((noreturn)) A& operator >>(A& a, B& b)
  {
    for (;;);
    /* return a; */
  }
};
[macbook] f90/bug% time g++-fsf-4.6 -c pr51344_db.C
0.010u 0.015s 0:00.05 40.0%    0+0k 0+4io 0pf+0w

while

[macbook] f90/bug% cat pr51344_db.C
/* { dg-do compile } */
class A;

template <class T>
class B
{
  friend __attribute__((format)) A& operator >>(A& a, B& b)
  {
    for (;;);
    /* return a; */
  }
};
macbook] f90/bug% time g++-fsf-4.6 -c pr51344_db.C
^C0.002u 0.005s 0:37.44 0.0%    0+0k 0+1io 0pf+0w

^
| interrupted

[Bug target/51934] FAIL: g++.dg/torture/pr51344.C -O0 (test for excess errors) due to "cdecl attribute ignored" warning

[jason at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51934

--- Comment #16 from Jason Merrill <jason at gcc dot gnu.org> 2012-01-23 15:30:58 UTC ---
Author: jason
Date: Mon Jan 23 15:30:48 2012
New Revision: 183436

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=183436
Log:
    PR target/51934
    * g++.dg/torture/pr51344.C: Limit to x86.

Modified:
    trunk/gcc/testsuite/ChangeLog
    trunk/gcc/testsuite/g++.dg/torture/pr51344.C

[Bug target/51934] FAIL: g++.dg/torture/pr51344.C -O0 (test for excess errors) due to "cdecl attribute ignored" warning

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51934

--- Comment #17 from Dominique d'Humieres <dominiq at lps dot ens.fr> 2012-01-23 15:50:16 UTC ---
> g++.dg/torture/pr51344.C: Limit to x86.

Note that using 'format' instead of 'cdecl' hangs also on
powerpc-apple-darwin9:

[karma] f90/bug% time g++-fsf-4.6 pr51344_db.C
^C0.002u 0.006s 1:35.05 0.0%    0+0k 2+7io 46pf+0w
 ^
 | interrupted
[karma] f90/bug% time /opt/gcc/gcc4.7w/bin/g++ -c pr51344_db.C
0.036u 0.071s 0:01.50 6.6%    0+0k 16+10io 792pf+0w
[karma] f90/bug% /opt/gcc/gcc4.7w/bin/g++ -v
Using built-in specs.
COLLECT_GCC=/opt/gcc/gcc4.7w/bin/g++
COLLECT_LTO_WRAPPER=/opt/gcc/gcc4.7w/libexec/gcc/powerpc-apple-darwin9.8.0/4.7.0/lto-wrapper
Target: powerpc-apple-darwin9.8.0
Configured with: ../work/configure --prefix=/opt/gcc/gcc4.7w
--enable-languages=c,c++,fortran,objc,obj-c++,java,ada,lto --with-gmp=/opt/mp
--with-system-zlib --with-cloog=/opt/mp --enable-lto --enable-cloog-backend=isl
Thread model: posix
gcc version 4.7.0 20120121 (experimental) [trunk revision 183381p3] (GCC) 

If 'format' is invalid in this context, g++ should probably emit an error
instead of hanging.

[Bug target/51934] FAIL: g++.dg/torture/pr51344.C -O0 (test for excess errors) due to "cdecl attribute ignored" warning

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51934

--- Comment #18 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-01-23 15:58:01 UTC ---
I bet you'd have to instantiate the template to see the errors.
Nevertheless, it would be a bad idea to add __attribute__((format)) to the
testcase and expect that we don't error on that, perhaps one day we'd error on
it even when not instantiated.

[Bug target/51934] FAIL: g++.dg/torture/pr51344.C -O0 (test for excess errors) due to "cdecl attribute ignored" warning

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51934

--- Comment #19 from Dominique d'Humieres <dominiq at lps dot ens.fr> 2012-01-23 22:05:44 UTC ---
> g++.dg/torture/pr51344.C: Limit to x86.
>
> -/* { dg-do compile } */
> +/* { dg-do compile { target { i?86-*-* && ilp32 } } } */

Any reason to limit the test to i?86 and 32 bit mode? Why not on x86_64 and 64
bit mode?
Is it related to the failures seen at
http://gcc.gnu.org/ml/gcc-testresults/2012-01/msg02281.html for instance?

The test passes for both -m32 and -m64 on x86_64-apple-darwin10 with

/* { dg-do compile { target { i?86-*-* x86_64-*-* } } } */

[Bug target/51934] FAIL: g++.dg/torture/pr51344.C -O0 (test for excess errors) due to "cdecl attribute ignored" warning

[jason at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51934

--- Comment #20 from Jason Merrill <jason at gcc dot gnu.org> 2012-01-24 02:45:28 UTC ---
cdecl only applies to the 32-bit ABI.  ix86_handle_cconv_attribute normally
warns about using it in 64-bit mode, I wonder why it doesn't on darwin?

[Bug target/51934] FAIL: g++.dg/torture/pr51344.C -O0 (test for excess errors) due to "cdecl attribute ignored" warning

[danglin at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51934

John David Anglin <danglin at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|RESOLVED                    |REOPENED
         Resolution|FIXED                       |

--- Comment #21 from John David Anglin <danglin at gcc dot gnu.org> 2012-01-25 14:57:39 UTC ---
Patch also needs to be applied to 4.6 branch.

[Bug target/51934] FAIL: g++.dg/torture/pr51344.C -O0 (test for excess errors) due to "cdecl attribute ignored" warning

[jason at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51934

--- Comment #22 from Jason Merrill <jason at gcc dot gnu.org> 2012-01-25 16:33:56 UTC ---
Author: jason
Date: Wed Jan 25 16:33:50 2012
New Revision: 183526

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=183526
Log:
    PR target/51934
    * g++.dg/torture/pr51344.C: Limit to x86.

Modified:
    branches/gcc-4_6-branch/gcc/testsuite/ChangeLog
    branches/gcc-4_6-branch/gcc/testsuite/g++.dg/torture/pr51344.C

[Bug target/51934] FAIL: g++.dg/torture/pr51344.C -O0 (test for excess errors) due to "cdecl attribute ignored" warning

[danglin at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51934

John David Anglin <danglin at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|REOPENED                    |RESOLVED
         Resolution|                            |FIXED

--- Comment #23 from John David Anglin <danglin at gcc dot gnu.org> 2012-01-25 19:08:41 UTC ---
Resolved.

[Bug target/51934] FAIL: g++.dg/torture/pr51344.C -O0 (test for excess errors) due to "cdecl attribute ignored" warning

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51934

--- Comment #24 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-03-06 08:26:03 UTC ---
Author: jakub
Date: Tue Mar  6 08:25:51 2012
New Revision: 184976

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=184976
Log:
    Backported from 4.6 branch
    2012-01-25  Jason Merrill  <jason@redhat.com>

    PR target/51934
    * g++.dg/torture/pr51344.C: Limit to x86.

Modified:
    branches/gcc-4_5-branch/gcc/testsuite/ChangeLog
    branches/gcc-4_5-branch/gcc/testsuite/g++.dg/torture/pr51344.C

[Bug target/51934] FAIL: g++.dg/torture/pr51344.C -O0 (test for excess errors) due to "cdecl attribute ignored" warning

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51934

--- Comment #25 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-03-06 08:26:32 UTC ---
Author: jakub
Date: Tue Mar  6 08:26:22 2012
New Revision: 184977

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=184977
Log:
    Backported from 4.6 branch
    2012-01-25  Jason Merrill  <jason@redhat.com>

    PR target/51934
    * g++.dg/torture/pr51344.C: Limit to x86.

Modified:
    branches/gcc-4_4-branch/gcc/testsuite/ChangeLog
    branches/gcc-4_4-branch/gcc/testsuite/g++.dg/torture/pr51344.C