Really weird things happening in Guile/GDB

[Mikael Djurfeldt]

Hi Jim,

I've experienced something really weird while running and debugging
Guile in GDB.  I thought maybe I could run this by you since you've
been working on GDB.

It all started with Guile getting hung during one of the first garbage
collections.  It turned out that it got into an infinity loop in
scm_i_dbl2big due to this function being passed a NAN from
scm_gc_stats.

When tracking down where this NAN got created, it was at the statement
marked below within gc.c:gc_end_stats.  I could set a watchpoint on
the double scm_gc_cells_marked_acc, and could see
scm_gc_cells_marked_acc change from 0 to -nan(0x8000000000000) when
adding the difference between the ulong scm_gc_cells_swept=0x6f58 and
the ulong scm_gc_cells_collected=0x350d at the line marked with =>.

static void
gc_end_stats ()
{
  unsigned long t = scm_c_get_internal_run_time ();
  scm_gc_time_taken += (t - t_before_gc);
  scm_gc_sweep_time_taken += (t - t_before_sweep);
  ++scm_gc_times;

=>scm_gc_cells_marked_acc += scm_gc_cells_swept - scm_gc_cells_collected;
  scm_gc_cells_swept_acc += scm_gc_cells_swept;
}

Of course I expected a compiler bug (am using gcc version 2.95.4
20010902 (Debian prerelease)), so I looked at the assembler output,
which looks completely OK:

----------------------------------------------------------------------
gc_end_stats:
.LBB14:
        pushl %ebp
        movl %esp,%ebp
        subl $28,%esp
        pushl %edi
        pushl %esi
        pushl %ebx
        call .L316
.L316:
        popl %ebx
        addl $_GLOBAL_OFFSET_TABLE_+[.-.L316],%ebx
.stabn 68,0,875,.LM135-gc_end_stats
.LM135:
        call address@hidden
        movl %eax,%ecx
.stabn 68,0,876,.LM136-gc_end_stats
.LM136:
        movl address@hidden(%ebx),%eax
        movl %ecx,%edx
        subl address@hidden(%ebx),%edx
        addl %edx,(%eax)
.stabn 68,0,877,.LM137-gc_end_stats
.LM137:
        movl address@hidden(%ebx),%eax
        subl address@hidden(%ebx),%ecx
        addl %ecx,(%eax)
.stabn 68,0,878,.LM138-gc_end_stats
.LM138:
        movl address@hidden(%ebx),%eax
        incl (%eax)
.stabn 68,0,884,.LM139-gc_end_stats
.LM139:
        movl address@hidden(%ebx),%esi
        movl address@hidden(%ebx),%edi
        movl (%edi),%eax
        movl (%esi),%edx
        subl %eax,%edx
        movl %edx,%eax
        xorl %edx,%edx
        movl %eax,-8(%ebp)
        movl %edx,-4(%ebp)
        fildll -8(%ebp)
        movl address@hidden(%ebx),%ecx
        faddl (%ecx)
.stabn 68,0,885,.LM140-gc_end_stats
.LM140:
        addl $-12,%esp
        leal address@hidden(%ebx),%eax
.stabn 68,0,884,.LM141-gc_end_stats
.LM141:
        fstpl (%ecx)
.stabn 68,0,885,.LM142-gc_end_stats
.LM142:
        pushl (%edi)
        pushl (%esi)
        pushl 4(%ecx)
        pushl (%ecx)
        pushl %eax
        call address@hidden
.stabn 68,0,887,.LM143-gc_end_stats
.LM143:
        movl (%esi),%eax
        xorl %edx,%edx
        movl %eax,-8(%ebp)
        movl %edx,-4(%ebp)
        fildll -8(%ebp)
        movl address@hidden(%ebx),%ecx
        faddl (%ecx)
        fstpl (%ecx)
        leal -40(%ebp),%esp
        popl %ebx
        popl %esi
        popl %edi
        leave
        ret
.stabn 68,0,888,.LM144-gc_end_stats
.LM144:
.LBE14:
.Lfe13:
        .size    gc_end_stats,.Lfe13-gc_end_stats
.stabs "t:r(0,5)",64,0,875,1
.stabn 192,0,0,.LBB14-gc_end_stats
.stabn 224,0,0,.LBE14-gc_end_stats
.Lscope12:
.stabs "",36,0,0,.Lscope12-gc_end_stats
----------------------------------------------------------------------

What on Earth might be going on???

Mikael (who was debugging using GNU gdb 5.0.90-cvs (MI_OUT) on an
i686-pc-linux-gnu)