Bug or compile issue?

[Charles Day]

I'm seeing some CRIT messages on my Mac, which is running the latest code
from SVN trunk. I've used gdb to try to figure out what's happening, and it
seems a bit weird, like a possible compile or other platform-specific issue.
The CRIT messages look like:
* 03:28:15  CRIT <gnc.ledger>         [gnc_split_register_get_trans_split()]
bad row
* 03:28:15  CRIT <gnc.register.gnome>
gnucash_sheet_cursor_set_from_table: assertion `gnucash_sheet_cell_valid
(sheet, v_loc)' failed

The specific line that seems to fail is line 1471 of
src/register/register-core/table-allgui.c:
  vloc = *virt_loc;

This statement is supposed to copy a struct:

(gdb) whatis vloc
type = VirtualLocation
(gdb) whatis *virt_loc
type = VirtualLocation
(gdb) ptype VirtualLocation
type = struct _VirtualLocation {
    VirtualCellLocation vcell_loc;
    int phys_row_offset;
    int phys_col_offset;
}
(gdb) ptype VirtualCellLocation
type = struct _VirtualCellLocation {
    int virt_row;
    int virt_col;
}

However, the struct only seems to be copied properly *sometimes*.  For
example:
(gdb) next
1471  vloc = *virt_loc;
(gdb) print *virt_loc
$82 = {
  vcell_loc = {
    virt_row = 2,
    virt_col = 0
  },
  phys_row_offset = 0,
  phys_col_offset = 0
}
(gdb) next
1473  vcell = gnc_table_get_virtual_cell (table, vloc.vcell_loc);
(gdb) print vloc
$90 = {
  vcell_loc = {
    virt_row = 2,
    virt_col = 0
  },
  phys_row_offset = -1073749880,
  phys_col_offset = 720727
}

So part of the struct copied OK (virt_row, virt_col) and the rest seems to
be garbage (phys_row_offset, phys_col_offset). Even stepping forward a few
more times doesn't change the result. Any ideas how this could be happening?
Some kind of optimization I should turn off perhaps?

I don't think it is a gdb problem, as I still get the CRIT messages without
gdb.

Cheers,
Charles