Strange backtrace from GDB 7.5.91

Strange backtrace from GDB 7.5.91

[Eli Zaretskii]

I saw today a very strange backtrace, while debugging with GDB 7.5.91,
which I got by stepping through relocate_gdb_directory:

  (top-gdb) s
  relocate_path (flag=1, initial=0x71a51b "d:/usr/share/gdb",
      progname=0x260ed0 "d:\\usr\\eli\\utils\\gdb-7.5.91\\gdb\\gdb.exe")
      at main.c:124
  124       dir = relocate_path (gdb_program_name, initial, flag);
  (top-gdb) bt
  #0  relocate_path (flag=1, initial=0x71a51b "d:/usr/share/gdb",
      progname=0x260ed0 "d:\\usr\\eli\\utils\\gdb-7.5.91\\gdb\\gdb.exe")
      at main.c:124
  #1  relocate_gdb_directory (
      initial=initial@entry=0x71a51b "d:/usr/share/gdb", flag=flag@entry=1)
      at main.c:124
  #2  0x004f0b1d in captured_main (data=data@entry=0x572ff10) at main.c:399

But relocate_path's argument list is this:

  relocate_path (const char *progname, const char *initial, int flag)

i.e., the FLAG argument is the last one, not the first one.

After one more "step", relocate_path appears twice in the backtrace,
once with correct order of arguments, the other one with an incorrect
order:

  (top-gdb) s
  relocate_path (initial=0x71a51b "d:/usr/share/gdb",
      progname=0x260ed0 "d:\\usr\\eli\\utils\\gdb-7.5.91\\gdb\\gdb.exe",
      flag=<optimized out>) at main.c:109
  109         return make_relative_prefix (progname, BINDIR, initial);
  (top-gdb) bt
  #0  relocate_path (initial=0x71a51b "d:/usr/share/gdb",
      progname=0x260ed0 "d:\\usr\\eli\\utils\\gdb-7.5.91\\gdb\\gdb.exe",
      flag=<optimized out>) at main.c:109
  #1  relocate_path (flag=1, initial=0x71a51b "d:/usr/share/gdb",
      progname=0x260ed0 "d:\\usr\\eli\\utils\\gdb-7.5.91\\gdb\\gdb.exe")
      at main.c:120
  #2  relocate_gdb_directory (
      initial=initial@entry=0x71a51b "d:/usr/share/gdb", flag=flag@entry=1)
      at main.c:124
  #3  0x004f0b1d in captured_main (data=data@entry=0x572ff10) at main.c:399

Is this due to inlining or something?  Is this a bug?

The compiler was GCC 4.7.2, if that matters, and the compilation
switches were "-g3 -O2".  MinGW GCC produced DWARF-2 debug info.

Re: Strange backtrace from GDB 7.5.91

[Jan Kratochvil]

On Wed, 20 Mar 2013 18:45:02 +0100, Eli Zaretskii wrote:
>   #0  relocate_path (flag=1, initial=0x71a51b "d:/usr/share/gdb",
>       progname=0x260ed0 "d:\\usr\\eli\\utils\\gdb-7.5.91\\gdb\\gdb.exe")
>       at main.c:124
>   #1  relocate_gdb_directory (
>       initial=initial@entry=0x71a51b "d:/usr/share/gdb", flag=flag@entry=1)
>       at main.c:124
[...]
> But relocate_path's argument list is this:
> 
>   relocate_path (const char *progname, const char *initial, int flag)
> 
> i.e., the FLAG argument is the last one, not the first one.

It needs a fix in GDB:
	Bug 15223 - reversed order of inlined function parameters 
	http://sourceware.org/bugzilla/show_bug.cgi?id=15223


>   #0  relocate_path (initial=0x71a51b "d:/usr/share/gdb",
>       progname=0x260ed0 "d:\\usr\\eli\\utils\\gdb-7.5.91\\gdb\\gdb.exe",
>       flag=<optimized out>) at main.c:109
>   #1  relocate_path (flag=1, initial=0x71a51b "d:/usr/share/gdb",
>       progname=0x260ed0 "d:\\usr\\eli\\utils\\gdb-7.5.91\\gdb\\gdb.exe")
>       at main.c:120
>   #2  relocate_gdb_directory (
>       initial=initial@entry=0x71a51b "d:/usr/share/gdb", flag=flag@entry=1)
>       at main.c:124
>   #3  0x004f0b1d in captured_main (data=data@entry=0x572ff10) at main.c:399
> 
> Is this due to inlining or something?  Is this a bug?

Probably all the frames #0..#2 are inlined, see "info frame".  It is not a bug
a single function is there multiple times, GCC does so.


Jan

Re: Strange backtrace from GDB 7.5.91

[Eli Zaretskii]

> Date: Wed, 20 Mar 2013 18:57:40 +0100
> From: Jan Kratochvil <jan.kratochvil@redhat.com>
> Cc: gdb@sourceware.org
> 
> > But relocate_path's argument list is this:
> > 
> >   relocate_path (const char *progname, const char *initial, int flag)
> > 
> > i.e., the FLAG argument is the last one, not the first one.
> 
> It needs a fix in GDB:
> 	Bug 15223 - reversed order of inlined function parameters 
> 	http://sourceware.org/bugzilla/show_bug.cgi?id=15223

Thanks.

Any chance of fixing this before 7.6 is released?

Re: Strange backtrace from GDB 7.5.91

[Jan Kratochvil]

On Wed, 20 Mar 2013 19:04:09 +0100, Eli Zaretskii wrote:
> > It needs a fix in GDB:
> > 	Bug 15223 - reversed order of inlined function parameters 
> > 	http://sourceware.org/bugzilla/show_bug.cgi?id=15223
> 
> Thanks.
> 
> Any chance of fixing this before 7.6 is released?

I do not think so, it is not a regression and I do not even think such change
is safe so late in the release process.


Jan