[Bug c++/105997] New: A possible optimization bug

[Bug c++/105997] New: A possible optimization bug

[zhonghao at pku dot org.cn]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105997

            Bug ID: 105997
           Summary: A possible optimization bug
           Product: gcc
           Version: 10.1.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c++
          Assignee: unassigned at gcc dot gnu.org
          Reporter: zhonghao at pku dot org.cn
  Target Milestone: ---

ZynAddSubFX is a musical synthesizer. I find a strange code in this project:

https://github.com/zynaddsubfx/zynaddsubfx/blob/master/src/Misc/MiddleWare.cpp

#if defined(__GNUC__) && !defined(__clang__) && !defined(__INTEL_COMPILER)
#pragma GCC push_options
#pragma GCC optimize("O0")
#endif

void gcc_10_1_0_is_dumb(const std::vector<std::string> &files,
        const int N,
        char *types,
        rtosc_arg_t *args)
{
        types[N] = 0;
        for(int i=0; i<N; ++i) {
            args[i].s = files[i].c_str();
            types[i]  = 's';
        }
}

#if defined(__GNUC__) && !defined(__clang__) && !defined(__INTEL_COMPILER)
#pragma GCC pop_options
#endif


I checked its commit, and found a message that complained an optimization bug
in gcc:

https://github.com/zynaddsubfx/zynaddsubfx/commit/a1abae354e826802f8b6f990cae225d6fb06b2ac

"Disable gcc optimizations for a specific function

Due to a gcc opt bug a piece of code was put in a separate function to
prevent gcc from optimizing it. Later versions of gcc inlined the
function and the bug reappeared. So now we explicitly tell gcc to not
optimize it."

Is it real a GCC bug?

[Bug c++/105997] A possible optimization bug

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105997

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|---                         |INVALID
             Status|UNCONFIRMED                 |RESOLVED

--- Comment #1 from Jonathan Wakely <redi at gcc dot gnu.org> ---
This is not a valid bug report.

https://gcc.gnu.org/bugs

[Bug c++/105997] A possible optimization bug

[schwab@linux-m68k.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105997

--- Comment #2 from Andreas Schwab <schwab@linux-m68k.org> ---
This line is definitely a bug in ZynAddSubFX:

1154:        bool is_dir = fn->d_type & DT_DIR;

d_type is not a bitfield.

[Bug c++/105997] A possible optimization bug

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105997

--- Comment #3 from Jonathan Wakely <redi at gcc dot gnu.org> ---
If anybody wants to turn it into a proper bug report, the original commit (and
description) is at
https://github.com/zynaddsubfx/zynaddsubfx/commit/22a5f8da76c16c47c3814eedaec8f9fe0ff4248b

[Bug c++/105997] A possible optimization bug

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105997

--- Comment #4 from Jonathan Wakely <redi at gcc dot gnu.org> ---
(In reply to Andreas Schwab from comment #2)
> This line is definitely a bug in ZynAddSubFX:
> 
> 1154:        bool is_dir = fn->d_type & DT_DIR;
> 
> d_type is not a bitfield.

Not all systems even have a dirent::d_type field (POSIX doesn't require it).

[Bug c++/105997] A possible optimization bug

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105997

--- Comment #5 from Jonathan Wakely <redi at gcc dot gnu.org> ---
(In reply to Jonathan Wakely from comment #3)
> If anybody wants to turn it into a proper bug report, the original commit
> (and description) is at
> https://github.com/zynaddsubfx/zynaddsubfx/commit/
> 22a5f8da76c16c47c3814eedaec8f9fe0ff4248b

"Symptoms of the failed optimization include a partial copy of values
with some corrupted or turned into null pointers."

Sounds likely to be a lifetime issue with the std::string values e.g. trying to
read them after the containing vector has been destroyed.

But somebody has to provide a proper bug report to be sure. Please stop opening
bug reports that are just a link to comments in github. If you want to
investigate them yourself, great. But "Somebody said GCC broke their code" is
not a valid bug report.

[Bug c++/105997] A possible optimization bug

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105997

--- Comment #6 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Also people who assume the compiler is broken ask them to try
-fsanitize=address and other sanitizers really. Gcc does is more "sensitive" to
undefined behavior and more bugs in the user code does happen.