[Bug c++/100629] New: Regression from 10 symbol mismatch between class definition and use with optimization

[Bug c++/100629] New: Regression from 10 symbol mismatch between class definition and use with optimization

[mckelvey at maskull dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100629

            Bug ID: 100629
           Summary: Regression from 10 symbol mismatch between class
                    definition and use with optimization
           Product: gcc
           Version: 11.1.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c++
          Assignee: unassigned at gcc dot gnu.org
          Reporter: mckelvey at maskull dot com
  Target Milestone: ---

Created attachment 50820
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=50820&action=edit
Source of Break class

On gcc-11/12 I get undefined symbols on link. Like this:

/usr/bin/ld: header_edit.o:header_edit.cc:(.text.startup+0x36a): undefined
reference to
`_ZN13PatternDriver12BreakPatternC1ERKNSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEERKNS_26ConflateIntegerScalarValueESB_RKNS_12TemplateEnumIL_ZNS_16complement_namesB5cxx11EEL_ZNS_14COMPLEMENTENUMEEEE'

>From main program header_edit.o:

_ZN13PatternDriver12BreakPatternC1ERKNSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEERKNS_26ConflateIntegerScalarValueESB_RKNS_12TemplateEnumIL_ZNS_16complement_namesB5cxx11EEL_ZNS_14COMPLEMENTENUMEEEE

from BreakPattern.o:

_ZN13PatternDriver12BreakPatternC1ERKNSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEERKNS_26ConflateIntegerScalarValueESB_RKNS_12TemplateEnumIL_ZNS_16complement_namesEEL_ZNS_14COMPLEMENTENUMEEEE

The problem is that the second string has the sequence B5cxx11. That sequence
goes away when I change the optimization level from level 3 to default. I don't
know how to reduce to a simple test case.

[Bug c++/100629] Regression from 10 symbol mismatch between class definition and use with optimization

[mckelvey at maskull dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100629

--- Comment #1 from James McKelvey <mckelvey at maskull dot com> ---
Created attachment 50821
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=50821&action=edit
Source of Main

[Bug c++/100629] Regression from 10 symbol mismatch between class definition and use with optimization

[mckelvey at maskull dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100629

--- Comment #2 from James McKelvey <mckelvey at maskull dot com> ---
Occurs with and without optimization. I'll compare errors between the two.

[Bug c++/100629] Regression from 10 symbol mismatch between class definition and use with optimization

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100629

--- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
The difference is
PatternDriver::TemplateEnum<PatternDriver::complement_names[abi:cxx11]
vs
PatternDriver::TemplateEnum<PatternDriver::complement_names

[Bug c++/100629] Regression from 10 symbol mismatch between class definition and use with optimization

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100629

--- Comment #4 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Andrew Pinski from comment #3)
> The difference is
> PatternDriver::TemplateEnum<PatternDriver::complement_names[abi:cxx11]
> vs
> PatternDriver::TemplateEnum<PatternDriver::complement_names

Can you attach the preprocessed source for both?

[Bug c++/100629] Regression from 10 symbol mismatch between class definition and use with optimization

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100629

--- Comment #5 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Andrew Pinski from comment #3)
> The difference is
> PatternDriver::TemplateEnum<PatternDriver::complement_names[abi:cxx11]
> vs
> PatternDriver::TemplateEnum<PatternDriver::complement_names

Sorry I just mean:
PatternDriver::complement_names[abi:cxx11]
vs
PatternDriver::complement_names

[Bug c++/100629] Regression from 10 symbol mismatch between class definition and use with optimization

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100629

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
     Ever confirmed|0                           |1
   Last reconfirmed|                            |2021-05-17
             Status|UNCONFIRMED                 |WAITING

--- Comment #6 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Andrew Pinski from comment #5)
> (In reply to Andrew Pinski from comment #3)
> > The difference is
> > PatternDriver::TemplateEnum<PatternDriver::complement_names[abi:cxx11]
> > vs
> > PatternDriver::TemplateEnum<PatternDriver::complement_names
> 
> Sorry I just mean:
> PatternDriver::complement_names[abi:cxx11]
> vs
> PatternDriver::complement_names

Are you sure there is no define for _GLIBCXX_USE_CXX11_ABI used somewhere?
See https://gcc.gnu.org/onlinedocs/libstdc++/manual/using_dual_abi.html also.

[Bug c++/100629] Regression from 10 symbol mismatch between class definition and use with optimization

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100629

--- Comment #7 from Jonathan Wakely <redi at gcc dot gnu.org> ---
(In reply to Andrew Pinski from comment #6)
> Are you sure there is no define for _GLIBCXX_USE_CXX11_ABI used somewhere?

Both symbols are using std::__cxx11::basic_string so that means both files are
using the cxx11 ABI.

We need the preprocessed source. The files attached are useless because they
don't show the definition of PatternDriver::complement_names.

[Bug c++/100629] Regression from 10 symbol mismatch between class definition and use with optimization

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100629

--- Comment #8 from Jonathan Wakely <redi at gcc dot gnu.org> ---
I suspect that one file only sees a forward declaration of complement_names
(without the abi_tag) and the other file sees the definition (with the
abi_tag). The question is why the tag is on that type at all, but without
preprocessed sources we can only guess.

[Bug c++/100629] Regression from 10 symbol mismatch between class definition and use with optimization

[mckelvey at maskull dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100629

--- Comment #9 from James McKelvey <mckelvey at maskull dot com> ---
Created attachment 50823
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=50823&action=edit
Preprocessed source for BreakPattern.h

[Bug c++/100629] Regression from 10 symbol mismatch between class definition and use with optimization

[mckelvey at maskull dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100629

--- Comment #10 from James McKelvey <mckelvey at maskull dot com> ---
Created attachment 50824
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=50824&action=edit
Preprocessed source for main

[Bug c++/100629] Regression from 10 symbol mismatch between class definition and use with/without optimization

[mckelvey at maskull dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100629

--- Comment #11 from James McKelvey <mckelvey at maskull dot com> ---
I don't see _GLIBCXX_USE_CXX11_ABI anywhere in the code.

[Bug c++/100629] Regression from 10 symbol mismatch between class definition and use with/without optimization

[mckelvey at maskull dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100629

--- Comment #12 from James McKelvey <mckelvey at maskull dot com> ---
I don't see _GLIBCXX_USE_CXX11_ABI anywhere in the code.

[Bug c++/100629] Regression from 10 symbol mismatch between class definition and use with/without optimization

[mckelvey at maskull dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100629

James McKelvey <mckelvey at maskull dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |mckelvey at maskull dot com

--- Comment #13 from James McKelvey <mckelvey at maskull dot com> ---
Created attachment 50825
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=50825&action=edit
Preprocessed source for BreakPattern.cc

[Bug c++/100629] Regression from 10 symbol mismatch between class definition and use with/without optimization

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100629

--- Comment #14 from Jonathan Wakely <redi at gcc dot gnu.org> ---
(In reply to James McKelvey from comment #10)
> Created attachment 50824 [details]
> Preprocessed source for main

This is header_edit.cc right?

I don't see any definition of the missing constructor here. This one:


    BreakPattern(const pdstring& value,
                 const ConflateIntegerScalarValue& minimum,
                 const ConflateIntegerScalarValue& maximum,
                 const ComplementEnum& a_complement);

[Bug c++/100629] Regression from 10 symbol mismatch between class definition and use with/without optimization

[mckelvey at maskull dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100629

--- Comment #15 from James McKelvey <mckelvey at maskull dot com> ---
The Break's are actually functions that call the constructor. They look like
this:


inline BreakPattern

Break(const AbstractStringScalarValue& value,
      const ComplementEnum&            a_complement = noncomplement)
{
    return BreakPattern(value, a_complement);
}

there are 12 of them.

[Bug c++/100629] Regression from 10 symbol mismatch between class definition and use with/without optimization

[mckelvey at maskull dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100629

--- Comment #16 from James McKelvey <mckelvey at maskull dot com> ---
The Break's are actually functions that call the constructor. They look like
this:


inline BreakPattern

Break(const AbstractStringScalarValue& value,
      const ComplementEnum&            a_complement = noncomplement)
{
    return BreakPattern(value, a_complement);
}

there are 12 of them.

[Bug c++/100629] Regression from 10 symbol mismatch between class definition and use with/without optimization

[mckelvey at maskull dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100629

--- Comment #17 from James McKelvey <mckelvey at maskull dot com> ---
Sorry, that constructor is in the .cc file. Looks like the .cc contents weren't
picked up by the attempt to get the preprocessed source.


BreakPattern::

BreakPattern(const pdstring&                   value,
             const ConflateIntegerScalarValue& minimum,
             const ConflateIntegerScalarValue& maximum,
             const ComplementEnum&             a_complement)
    :
    AbsCharSetPattern(value,
                      a_complement,
                      std::max(minimum._get_minimum_match_length(), 1UL) -
1UL),
    _minimum(minimum._get_value()),
    _maximum(maximum._get_value())
{
}

[Bug c++/100629] Regression from 10 symbol mismatch between class definition and use with/without optimization

[mckelvey at maskull dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100629

--- Comment #18 from James McKelvey <mckelvey at maskull dot com> ---
If I build with -D_GLIBCXX_USE_CXX11_ABI=0 it all works. But with =1 I get the
mismatches.

[Bug c++/100629] Regression from 10 symbol mismatch between class definition and use with/without optimization

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100629

Jonathan Wakely <redi at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|WAITING                     |UNCONFIRMED
     Ever confirmed|1                           |0

--- Comment #19 from Jonathan Wakely <redi at gcc dot gnu.org> ---
(In reply to James McKelvey from comment #17)
> Sorry, that constructor is in the .cc file.

Ah yes, sorry, I see it now.

I tried to reproduce the issue with this code, but the symbol is always mangled
consistently (with and without MAIN defined):

#include <vector>
#include <string>

class CString { };

extern const std::vector<std::string> complement_names;
extern const CString COMPLEMENTENUM;


template <const std::vector<std::string>& names,
          const CString& title>
class TemplateEnum
{
};

class ConflateIntegerScalarValue { };

typedef TemplateEnum<complement_names, COMPLEMENTENUM> ComplementEnum;

struct BreakPattern
{
    BreakPattern(const std::string&,
                        const ConflateIntegerScalarValue&,
                        const ConflateIntegerScalarValue&,
                        const ComplementEnum&);
};

#ifdef MAIN
int main()
{
  BreakPattern b("", {}, {}, {});
}
#else
BreakPattern::
BreakPattern(const std::string&,
             const ConflateIntegerScalarValue&,
             const ConflateIntegerScalarValue&,
             const ComplementEnum&)
{
}
#endif

[Bug c++/100629] Regression from 10 symbol mismatch between class definition and use with/without optimization

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100629

Jonathan Wakely <redi at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
     Ever confirmed|0                           |1
             Status|UNCONFIRMED                 |WAITING

--- Comment #20 from Jonathan Wakely <redi at gcc dot gnu.org> ---
Please also provide all the gcc options you're using, as requested at
https://gcc.gnu.org/bugs/

[Bug c++/100629] Regression from 10 symbol mismatch between class definition and use with/without optimization

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100629

Jonathan Wakely <redi at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |link-failure
      Known to fail|                            |11.1.1, 12.0
             Status|WAITING                     |NEW

--- Comment #21 from Jonathan Wakely <redi at gcc dot gnu.org> ---
(In reply to James McKelvey from comment #10)
> Created attachment 50824 [details]
> Preprocessed source for main

OK, I can confirm it using this file, at -O2 or -O3.

[Bug c++/100629] Regression from 10 symbol mismatch between class definition and use with/without optimization

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100629

--- Comment #22 from Jonathan Wakely <redi at gcc dot gnu.org> ---
This doesn't seem to be a compiler change, because I see the same mangling
compiling the .ii with GCC 10 (after adjusting the .ii to remove the uses of
the new intrinsics from GCC 11). So possibly a library change.

[Bug c++/100629] Regression from 10 symbol mismatch between class definition and use with/without optimization

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100629

--- Comment #23 from Jonathan Wakely <redi at gcc dot gnu.org> ---
OK, the bad mangling happens with -O1 -fdevirtualize and not with -O1 alone.

[Bug c++/100629] Regression from 10 symbol mismatch between class definition and use with/without optimization

[mckelvey at maskull dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100629

--- Comment #24 from James McKelvey <mckelvey at maskull dot com> ---
Running Cygwin 3.2.0(0.340/5/3) under Windows 10.

Using built-in specs.
COLLECT_GCC=/usr/local/bin/g++
COLLECT_LTO_WRAPPER=/usr/local/libexec/gcc/x86_64-pc-cygwin/11.1.1/lto-wrapper.exe
Target: x86_64-pc-cygwin
Configured with: ./configure --enable-languages=c,c++
Thread model: single
Supported LTO compression algorithms: zlib
gcc version 11.1.1 20210515 (GCC)

Compiler options compiling project:

/usr/local/bin/g++ -std=c++0x -c  -O3 -DNDEBUG     -DUSE_INTL=1 -DUSE_MUTEX=1 
-D_FORTIFY_SOURCE=1 -pedantic-errors -Werror -fno-common -Wall -Wold-style-cast
-Wsign-promo -Wpointer-arith -Wundef -Wwrite-strings -Winvalid-pch
-Woverloaded-virtual -Wcast-qual -Wextra -Wredundant-decls -Wshadow
-Wcast-align=strict -Wcomment -fstrict-aliasing -Winit-self
-Wmissing-include-dirs -Wswitch-default -Wswitch-enum -Wlogical-op -Wconversion
-Wsign-conversion -Wmissing-declarations -Wdeprecated -ftree-switch-conversion
-Wuninitialized -Wparentheses -Wimplicit-fallthrough=5 -Wformat-nonliteral
-Wformat-truncation=2 -Wformat-signedness -Wdouble-promotion -Wformat=2
-Wformat-overflow=2 -Wformat-security -Wnull-dereference -Wmain
-Wmultistatement-macros -Wsequence-point -Wswitch-default -Wswitch-enum
-Wunused-value -Wstrict-overflow=2 -Warith-conversion -Wfloat-conversion
-Wduplicated-cond -Wunsafe-loop-optimizations -Wreturn-type -Wunused-parameter
-Wmaybe-uninitialized -Wstrict-aliasing -Wsuggest-attribute=noreturn
-Wsuggest-attribute=format -Wsuggest-attribute=malloc
-Wmissing-format-attribute -Wmissing-noreturn -Walloc-zero -Walloca
-Wtrampolines -Wcast-function-type -Wlogical-op -Wpacked -Wredundant-decls
-Wunused -Wmismatched-dealloc -Wmismatched-new-delete -Wfree-nonheap-object
-Wexceptions -MMD      -fimplicit-templates

For linking:

/usr/local/bin/g++ -std=c++0x  -O3 -DNDEBUG     -DUSE_INTL=1 -DUSE_MUTEX=1 
-D_FORTIFY_SOURCE=1 -pedantic-errors -Werror -fno-common -Wall -Wold-style-cast
-Wsign-promo -Wpointer-arith -Wundef -Wwrite-strings -Winvalid-pch
-Woverloaded-virtual -Wcast-qual -Wextra -Wredundant-decls -Wshadow
-Wcast-align=strict -Wcomment -fstrict-aliasing -Winit-self
-Wmissing-include-dirs -Wswitch-default -Wswitch-enum -Wlogical-op -Wconversion
-Wsign-conversion -Wmissing-declarations -Wdeprecated -ftree-switch-conversion
-Wuninitialized -Wparentheses -Wimplicit-fallthrough=5 -Wformat-nonliteral
-Wformat-truncation=2 -Wformat-signedness -Wdouble-promotion -Wformat=2
-Wformat-overflow=2 -Wformat-security -Wnull-dereference -Wmain
-Wmultistatement-macros -Wsequence-point -Wswitch-default -Wswitch-enum
-Wunused-value -Wstrict-overflow=2 -Warith-conversion -Wfloat-conversion
-Wduplicated-cond -Wunsafe-loop-optimizations -Wreturn-type -Wunused-parameter
-Wmaybe-uninitialized -Wstrict-aliasing -Wsuggest-attribute=noreturn
-Wsuggest-attribute=format -Wsuggest-attribute=malloc
-Wmissing-format-attribute -Wmissing-noreturn -Walloc-zero -Walloca
-Wtrampolines -Wcast-function-type -Wlogical-op -Wpacked -Wredundant-decls
-Wunused -Wmismatched-dealloc -Wmismatched-new-delete -Wfree-nonheap-object
-Wexceptions -Winline -MMD      -fimplicit-templates  -Wl,-warn-common -L.. -o
header_edit.exe header_edit.o -lPatternDriver -lintl -lpthread

Yes I know that most warnings have no effect at link time.

[Bug c++/100629] Regression from 10 symbol mismatch between class definition and use with/without optimization

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100629

--- Comment #25 from Jonathan Wakely <redi at gcc dot gnu.org> ---
Created attachment 50830
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=50830&action=edit
auto-reduced testcase

This produces different mangling depending on -fdevirtualize

[Bug c++/100629] Regression from 10 symbol mismatch between class definition and use with/without optimization

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100629

--- Comment #26 from Jonathan Wakely <redi at gcc dot gnu.org> ---
But I can't confirm this as a regression, because the same behaviour goes all
the way back to GCC 5.1 when the abi_tag was introduced.

[Bug c++/100629] Regression from 10 symbol mismatch between class definition and use with/without optimization

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100629

--- Comment #27 from Jonathan Wakely <redi at gcc dot gnu.org> ---
Created attachment 50831
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=50831&action=edit
further reduced

[Bug c++/100629] Regression from 10 symbol mismatch between class definition and use with/without optimization

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100629

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           See Also|                            |https://gcc.gnu.org/bugzill
                   |                            |a/show_bug.cgi?id=69621
           Keywords|                            |ABI, wrong-code

--- Comment #28 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Seems related to PR 69621.