[Bug debug/46580] New: -fcompare-debug failure with -gstabs -g due to different symbol mangling

[Bug debug/46580] New: -fcompare-debug failure with -gstabs -g due to different symbol mangling

[zsojka at seznam dot cz]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46580

           Summary: -fcompare-debug failure with -gstabs -g due to
                    different symbol mangling
           Product: gcc
           Version: 4.6.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: debug
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: zsojka@seznam.cz


Created attachment 22469
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=22469
reduced testcase

Compiler output:
$ gcc -gstabs -g -fcompare-debug testcase.C        
gcc: error: testcase.C: -fcompare-debug failure

Differences:
$ diff testcase.*gkd
3c3
< ;; Function (_ZN1SC2EU8__vectorf) 
---
> ;; Function (_ZN1SC1EU8__vectorf) 

Both are demangled by c++filt to "S::S(float __vector)"

Tested revisions:
r166936 - fail
4.5 r166509 - fail

[Bug debug/46580] -fcompare-debug failure with -gstabs -g due to different symbol mangling

[zsojka at seznam dot cz]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46580

--- Comment #1 from Zdenek Sojka <zsojka at seznam dot cz> 2010-11-20 16:05:51 UTC ---
Created attachment 22470
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=22470
another testcase

$ gcc -gstabs -g -fcompare-debug testcase2.C
gcc: error: testcase2.C: -fcompare-debug failure (length)
$ diff testcase2.*gkd
3c3
< ;; Function (_ZN1f1fEv) 
---
> ;; Function (_ZN3._01fEv) 

First one demangles into "f::f()", second into "._0::f()"

[Bug debug/46580] -fcompare-debug failure with -gstabs -g due to different symbol mangling

[zsojka at seznam dot cz]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46580

Zdenek Sojka <zsojka at seznam dot cz> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |aoliva at gcc dot gnu.org
      Known to fail|                            |4.2.4, 4.3.5, 4.4.6

--- Comment #2 from Zdenek Sojka <zsojka at seznam dot cz> 2010-11-20 16:29:57 UTC ---
For the second testcase, all 4.2.4, 4.3.5, 4.4.6, 4.5.2 and 4.6.0 generate
different symbol names for "-gstabs -g" case, compared to "" case.
gcc 4.0.4 and 4.1.2 don't generate any code at all (?)
gcc 3.3.6 and 3.4.6 generate same symbol names, _ZN1f1fEv. I don't know if
that's by luck or not, so I am not adding them to the "Known to work" field.

The function can be renamed to something else than "f()", the problem persists.


First testcase seems to fail only in 4.5 and 4.6.

This probably isn't typical -fcompare-debug failure.

[Bug c++/46580] -fcompare-debug failure with -gstabs -g due to different symbol mangling

[aoliva at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46580

Alexandre Oliva <aoliva at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2010.11.21 09:23:01
          Component|debug                       |c++
     Ever Confirmed|0                           |1

--- Comment #3 from Alexandre Oliva <aoliva at gcc dot gnu.org> 2010-11-21 09:23:01 UTC ---
Indeed, not a typical -fcompare-debug error.

I looked a bit into the second testcase.  The problem is that finish_struct
calls debughooks->type_decl through finish_struct_1 and
rest_of_type_compilation, and within dbxout_type_decl we compute the
assembler_name for the type and nested decls.  Later on, when we override the
class name when processing the typedef name for the still anonymous class, the
assembler names are not touched, and remain with incorrect name mangling.

I can't decide whether this is a problem in the C++ front end, that should call
rest_of_type_compilation only at the end of the typedef declaration, or in
dbxout, that should refrain from resolving assembler names of anonymous types
before the end of compilation, if that's at all possible.  I don't know much
about stabs, so I'll leave this to someone else who does.

Reassigning to c++ in the hope that some g++ expert will chime in.

[Bug c++/46580] -fcompare-debug failure with -gstabs -g due to different symbol mangling

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=46580

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
      Known to work|5.3.0, 5.5.0, 6.1.0, 7.1.0  |

--- Comment #4 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
The first testcase is fixed in GCC 5.3.0 and GCC 6+.
The second testcase is fixed in GCC 10+.

[Bug c++/46580] -fcompare-debug failure with -gstabs -g due to different symbol mangling

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=46580

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|---                         |FIXED
   Target Milestone|---                         |10.0

--- Comment #5 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
The first testcase was most likely fixed by r5-9033 (r6-555 + r6-5094) which
changed where the alias managing happened.

The second testcase was most likely fixed by r10-707 which changed how the
TYPE_NAME is done for these kind of types.