[Bug c++/55325] New: [4.8 Regression]: g++.dg/cpp0x/constexpr-complex.C excess errors

[Bug c++/55325] New: [4.8 Regression]: g++.dg/cpp0x/constexpr-complex.C excess errors

[hp at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55325

             Bug #: 55325
           Summary: [4.8 Regression]: g++.dg/cpp0x/constexpr-complex.C
                    excess errors
    Classification: Unclassified
           Product: gcc
           Version: 4.8.0
            Status: UNCONFIRMED
          Keywords: wrong-code
          Severity: normal
          Priority: P3
         Component: c++
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: hp@gcc.gnu.org
                CC: emsr@gcc.gnu.org
              Host: x86_64-unknown-linux-gnu
            Target: cris-axis-elf, sh4-linux, i686-linux, x86_64-linux


This test previously passed, now it fails.
A patch in the revision range (last_known_working:first_known_failing)
193374:193385
exposed or caused this regression.  Since then it fails as follows:

Running /tmp/hpautotest-gcc1/gcc/gcc/testsuite/g++.dg/dg.exp ...
...
FAIL: g++.dg/cpp0x/constexpr-complex.C (test for excess errors)


In g++.log for cris-elf:
Executing on host:
/tmp/hpautotest-gcc1/cris-elf/gccobj/gcc/testsuite/g++/../../g++
-B/tmp/hpautotest-gcc1/cris-elf/gccobj/gcc/testsuite/g++/../../
/tmp/hpautotest-gcc1/gcc/gcc/testsuite/g++.dg/cpp0x/constexpr-complex.C 
-fno-diagnostics-show-caret  -nostdinc++
-I/tmp/hpautotest-gcc1/cris-elf/gccobj/cris-elf/libstdc++-v3/include/cris-elf
-I/tmp/hpautotest-gcc1/cris-elf/gccobj/cris-elf/libstdc++-v3/include
-I/tmp/hpautotest-gcc1/gcc/libstdc++-v3/libsupc++
-I/tmp/hpautotest-gcc1/gcc/libstdc++-v3/include/backward
-I/tmp/hpautotest-gcc1/gcc/libstdc++-v3/testsuite/util -fmessage-length=0 
-std=c++0x  -S   -isystem
/tmp/hpautotest-gcc1/cris-elf/gccobj/cris-elf/./newlib/targ-include -isystem
/tmp/hpautotest-gcc1/gcc/newlib/libc/include  -o constexpr-complex.s   
(timeout = 300)
/tmp/hpautotest-gcc1/gcc/gcc/testsuite/g++.dg/cpp0x/constexpr-complex.C: In
constructor 'constexpr complex::complex(float, float)':
/tmp/hpautotest-gcc1/gcc/gcc/testsuite/g++.dg/cpp0x/constexpr-complex.C:12:26:
error: unable to find numeric literal operator 'operator"" fi'

Looks like the regression is universal:
sh4-linux: <http://gcc.gnu.org/ml/gcc-testresults/2012-11/msg01164.html>
i686-linux: <http://gcc.gnu.org/ml/gcc-testresults/2012-11/msg01171.html>
x86_64-linux: <http://gcc.gnu.org/ml/gcc-testresults/2012-11/msg01173.html>

Author of suspect patch in revision range CC:ed.

[Bug c++/55325] [4.8 Regression]: g++.dg/cpp0x/constexpr-complex.C excess errors

[paolo.carlini at oracle dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55325

Paolo Carlini <paolo.carlini at oracle dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2012-11-14
     Ever Confirmed|0                           |1

--- Comment #1 from Paolo Carlini <paolo.carlini at oracle dot com> 2012-11-14 16:18:05 UTC ---
I also see failing g++.dg/parse/template23.C and for this one too the problem
seems related to the recent changes for PR54413.

[Bug c++/55325] [4.8 Regression]: g++.dg/cpp0x/constexpr-complex.C excess errors

[paolo.carlini at oracle dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55325

Paolo Carlini <paolo.carlini at oracle dot com> changed:

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

--- Comment #2 from Paolo Carlini <paolo.carlini at oracle dot com> 2012-11-14 16:19:54 UTC ---
*** Bug 55326 has been marked as a duplicate of this bug. ***

[Bug c++/55325] [4.8 Regression]: g++.dg/cpp0x/constexpr-complex.C excess errors

[hp at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55325

--- Comment #3 from Hans-Peter Nilsson <hp at gcc dot gnu.org> 2012-11-14 16:59:29 UTC ---
(In reply to comment #1)
> I also see failing g++.dg/parse/template23.C and for this one too the problem
> seems related to the recent changes for PR54413.

Right, thanks, I was just about to analyze that one.  Speaking of that commit,
some of the new tests fail for me:
FAIL: g++.dg/cpp0x/gnu_fext-numeric-literals.C (test for excess errors)
FAIL: g++.dg/cpp0x/std_fext-numeric-literals.C (test for excess errors)

and in g++.log:
Excess errors:
/tmp/hpautotest-gcc1/gcc/gcc/testsuite/g++.dg/cpp0x/gnu_fext-numeric-literals.C:94:3:
error: unsupported non-standard suffix on floating constant
/tmp/hpautotest-gcc1/gcc/gcc/testsuite/g++.dg/cpp0x/gnu_fext-numeric-literals.C:95:3:
error: unsupported non-standard suffix on floating constant
/tmp/hpautotest-gcc1/gcc/gcc/testsuite/g++.dg/cpp0x/gnu_fext-numeric-literals.C:96:3:
error: unsupported non-standard suffix on floating constant
/tmp/hpautotest-gcc1/gcc/gcc/testsuite/g++.dg/cpp0x/gnu_fext-numeric-literals.C:97:3:
error: unsupported non-standard sufffix on floating constant
maybe related, maybe material for a new PR.

[Bug c++/55325] [4.8 Regression]: g++.dg/cpp0x/constexpr-complex.C excess errors

[3dw4rd at verizon dot net]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55325

--- Comment #4 from Ed Smith-Rowland <3dw4rd at verizon dot net> 2012-11-14 20:20:27 UTC ---
OK, g++.dg/cpp0x/constexpr-complex.C will fail with the patch to control GNU
literal parsing.  i.e.  this behavior in intended.

The purpose of the test is to check that constexpr works with C99 complex
literals.  So we could fix the test in either one of two equivalent ways:

Invoke with gnu++0x:
// { dg-options -std=gnu++0x }

Invoke with new flag:
// { dg-options -std=c++0x -fext-numeric-literals }

Any preference on which one?

I'll look at the other fails.

[Bug c++/55325] [4.8 Regression]: g++.dg/cpp0x/constexpr-complex.C excess errors

[paolo.carlini at oracle dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55325

--- Comment #5 from Paolo Carlini <paolo.carlini at oracle dot com> 2012-11-14 20:36:37 UTC ---
If you ask me, I have I slight preference for the latter, because isn't always
obvious what gnu++* includes beyond c++*. But Jason will tell you, and the fix
is straightforward anyway.

[Bug c++/55325] [4.8 Regression]: g++.dg/cpp0x/constexpr-complex.C excess errors

[3dw4rd at verizon dot net]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55325

--- Comment #6 from Ed Smith-Rowland <3dw4rd at verizon dot net> 2012-11-14 21:52:14 UTC ---
(In reply to comment #4)
> 
> Invoke with gnu++0x:
> // { dg-options -std=gnu++0x }
> 
> Invoke with new flag:
> // { dg-options -std=c++0x -fext-numeric-literals }
> 
> I'll look at the other fails.

Similarly for template23.C.  We'll just set the correct flag in the testcase.

I'm still looking at the errors in comment 3.

[Bug c++/55325] [4.8 Regression]: g++.dg/cpp0x/constexpr-complex.C excess errors

[3dw4rd at verizon dot net]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55325

--- Comment #7 from Ed Smith-Rowland <3dw4rd at verizon dot net> 2012-11-14 22:18:09 UTC ---
(In reply to comment #3)
> (In reply to comment #1)
> > I also see failing g++.dg/parse/template23.C and for this one too the problem
> > seems related to the recent changes for PR54413.
> 
> Right, thanks, I was just about to analyze that one.  Speaking of that commit,
> some of the new tests fail for me:
> FAIL: g++.dg/cpp0x/gnu_fext-numeric-literals.C (test for excess errors)
> FAIL: g++.dg/cpp0x/std_fext-numeric-literals.C (test for excess errors)
> 
> and in g++.log:
> Excess errors:
> /tmp/hpautotest-gcc1/gcc/gcc/testsuite/g++.dg/cpp0x/gnu_fext-numeric-literals.C:94:3:
> ...
> maybe related, maybe material for a new PR.

I'm not getting this on x86_64-linux at least.  What target are you on?

Ed

[Bug c++/55325] [4.8 Regression]: g++.dg/cpp0x/constexpr-complex.C excess errors

[paolo.carlini at oracle dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55325

--- Comment #8 from Paolo Carlini <paolo.carlini at oracle dot com> 2012-11-14 22:28:19 UTC ---
The latter seems indeed a target issue, as you can read here Hans-Peter
reported it for cris-elf, but have a look to the testresults mailing list to
double check. Personally, I would recommend applying asap the (obvious, IMHO)
fixes for the first two regression to remove noise from the results of, eg,
people testing on Linux.

[Bug c++/55325] [4.8 Regression]: g++.dg/cpp0x/constexpr-complex.C excess errors

[paolo.carlini at oracle dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55325

--- Comment #9 from Paolo Carlini <paolo.carlini at oracle dot com> 2012-11-14 22:30:51 UTC ---
In fact, however, a Linux target like s390x is also affected:

  http://gcc.gnu.org/ml/gcc-testresults/2012-11/msg01187.html

[Bug c++/55325] [4.8 Regression]: g++.dg/cpp0x/constexpr-complex.C excess errors

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55325

--- Comment #10 from Uros Bizjak <ubizjak at gmail dot com> 2012-11-14 22:42:45 UTC ---
(In reply to comment #7)
> > Right, thanks, I was just about to analyze that one.  Speaking of that commit,
> > some of the new tests fail for me:
> > FAIL: g++.dg/cpp0x/gnu_fext-numeric-literals.C (test for excess errors)
> > FAIL: g++.dg/cpp0x/std_fext-numeric-literals.C (test for excess errors)
> > 
> > and in g++.log:
> > Excess errors:
> > /tmp/hpautotest-gcc1/gcc/gcc/testsuite/g++.dg/cpp0x/gnu_fext-numeric-literals.C:94:3:
> > ...
> > maybe related, maybe material for a new PR.
> 
> I'm not getting this on x86_64-linux at least.  What target are you on?

Probably due to the fact that x86_64 defines TARGET_C_MODE_FOR_SUFFIX that
declares support for non-standard 'q' and 'w' suffixes. There is a small group
of targets that declare this define:

config/aarch64/aarch64.c:#undef TARGET_C_MODE_FOR_SUFFIX
config/aarch64/aarch64.c:#define TARGET_C_MODE_FOR_SUFFIX
aarch64_c_mode_for_suffix
config/ia64/ia64.c:#undef TARGET_C_MODE_FOR_SUFFIX
config/ia64/ia64.c:#define TARGET_C_MODE_FOR_SUFFIX ia64_c_mode_for_suffix
config/i386/i386.c.orig:#undef TARGET_C_MODE_FOR_SUFFIX
config/i386/i386.c.orig:#define TARGET_C_MODE_FOR_SUFFIX ix86_c_mode_for_suffix
config/i386/i386.c:#undef TARGET_C_MODE_FOR_SUFFIX
config/i386/i386.c:#define TARGET_C_MODE_FOR_SUFFIX ix86_c_mode_for_suffix
config/pa/pa.c:#undef TARGET_C_MODE_FOR_SUFFIX
config/pa/pa.c:#define TARGET_C_MODE_FOR_SUFFIX pa_c_mode_for_suffix

Others are out of luck with 'q' and 'w' suffixes.

[Bug c++/55325] [4.8 Regression]: g++.dg/cpp0x/constexpr-complex.C excess errors

[hp at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55325

--- Comment #11 from Hans-Peter Nilsson <hp at gcc dot gnu.org> 2012-11-15 02:39:57 UTC ---
(In reply to comment #7)
> (In reply to comment #3)
> > Excess errors:
> > /tmp/hpautotest-gcc1/gcc/gcc/testsuite/g++.dg/cpp0x/gnu_fext-numeric-literals.C:94:3:
> > ...
> > maybe related, maybe material for a new PR.
> 
> I'm not getting this on x86_64-linux at least.  What target are you on?

I guess it's no longer interesting seeing the later comments, in particular
comment #10, but it's cris-axis-elf (obvious by elimination, see listed
targets).

Since TARGET_C_MODE_FOR_SUFFIX doesn't have a sane default, this part of the
test should IMHO be parametrized by say an testsuite "effective-target".

Or should the default for TARGET_C_MODE_FOR_SUFFIX changed to something sane?
By "sane" I mean the mode for "long double", (i.e. DFmode by default) for 'q'
and 'w'.  (BTW, I hate that I can't grep for TARGET_C_MODE_FOR_SUFFIX in the
sources and get sane results other than the target hits.)

[Bug c++/55325] [4.8 Regression]: g++.dg/cpp0x/constexpr-complex.C excess errors

[jason at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55325

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> 2012-11-15 02:57:01 UTC ---
(In reply to comment #5)
> If you ask me, I have I slight preference for the latter, because isn't always
> obvious what gnu++* includes beyond c++*. But Jason will tell you, and the fix
> is straightforward anyway.

gnu++ is mostly extra keywords.  I've fixed the testcases.

[Bug c++/55325] [4.8 Regression]: g++.dg/cpp0x/constexpr-complex.C excess errors

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55325

--- Comment #13 from Uros Bizjak <ubizjak at gmail dot com> 2012-11-15 08:26:11 UTC ---
(In reply to comment #10)

> Probably due to the fact that x86_64 defines TARGET_C_MODE_FOR_SUFFIX that
> declares support for non-standard 'q' and 'w' suffixes. There is a small group
> of targets that declare this define:

Patch for this problem is at [1].

[1] http://gcc.gnu.org/ml/gcc-patches/2012-11/msg01198.html

[Bug c++/55325] [4.8 Regression]: g++.dg/cpp0x/constexpr-complex.C excess errors

[hp at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55325

--- Comment #14 from Hans-Peter Nilsson <hp at gcc dot gnu.org> 2012-11-15 08:44:05 UTC ---
(In reply to comment #13)
> [1] http://gcc.gnu.org/ml/gcc-patches/2012-11/msg01198.html

Thanks.  Note that pa handles 'q'.

[Bug c++/55325] [4.8 Regression]: g++.dg/cpp0x/constexpr-complex.C excess errors

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55325

--- Comment #15 from Uros Bizjak <ubizjak at gmail dot com> 2012-11-15 16:23:51 UTC ---
(In reply to comment #14)
> (In reply to comment #13)
> > [1] http://gcc.gnu.org/ml/gcc-patches/2012-11/msg01198.html
> 
> Thanks.  Note that pa handles 'q'.

So, let's auto-detect support for non-standard suffixes in patch v2 [1].

[1] http://gcc.gnu.org/ml/gcc-patches/2012-11/msg01263.html

[Bug c++/55325] [4.8 Regression]: g++.dg/cpp0x/constexpr-complex.C excess errors

[hp at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55325

--- Comment #16 from Hans-Peter Nilsson <hp at gcc dot gnu.org> 2012-11-15 20:06:32 UTC ---
(In reply to comment #15)
> So, let's auto-detect support for non-standard suffixes in patch v2 [1].
> 
> [1] http://gcc.gnu.org/ml/gcc-patches/2012-11/msg01263.html

JFTR, I see there's a v3:
http://gcc.gnu.org/ml/gcc-patches/2012-11/msg01291.html
Nice. Thanks.