https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92493
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |jakub at gcc dot gnu.org
--- Comment #1 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
This is caused by the:
278098 msebor if ((RECORD_OR_UNION_TYPE_P (xtype)
278098 msebor && field_at_offset (xtype, *off, &index))
278098 msebor || (TREE_CODE (xtype) == ARRAY_TYPE
278098 msebor && TREE_CODE (TREE_TYPE (xtype)) ==
ARRAY_TYPE
278098 msebor && array_elt_at_offset (xtype, *off,
&index)))
278098 msebor {
278098 msebor *fldoff += index;
278098 msebor *off -= index;
278098 msebor fldoff = NULL;
278098 msebor }
278098 msebor }
278098 msebor
278098 msebor return get_origin_and_offset (x, fldoff, NULL);
get_origin_and_offset assumes that fldoff is never NULL.
If the intent that writes to *fldoff from the recursive call should be ignored,
then perhaps it can do say:
index = 0;
fldoff = &index;
instead of fldoff = NULL; or perhaps
else if (idx < HOST_WIDE_INT_MAX)
*fldoff += idx * int_size_in_bytes (eltype);
else
*fldoff = idx;
should be
else if (fldoff)
{
if (idx < HOST_WIDE_INT_MAX)
*fldoff += idx * int_size_in_bytes (eltype);
else
*fldoff = idx;
}
But, it isn't obvious what this function is meant for at all, so while either
one would fix the ICE, it is unclear what the right thing is.
