Re: GCC endless loop for store_bit_field with alignment 0

[David Edelsohn <dje at watson dot ibm dot com>]

	In my previous note I mis-typed my proposed change to
expmed.c:store_split_bit_field().  It should have been:

  if (GET_CODE (op0) == REG || GET_CODE (op0) == SUBREG)
    unit = BITS_PER_WORD;
  else
    unit = MAX (MIN (align, BITS_PER_WORD), BITS_PER_UNIT);


However, that still does not solve the problem because
store_split_bit_field() calls store_fixed_bit_field() which calls
get_best_mode() with alignment 0.  get_best_mode() returns VOIDmode
because the alignment is 0, so store_fixed_bit_field() performs:


     if (mode == VOIDmode)
        {
          /* The only way this should occur is if the field spans word
             boundaries.  */
          store_split_bit_field (op0,
                                 bitsize, bitpos + offset * BITS_PER_UNIT,
                                 value, struct_align);

which places GCC into infinite recursion between store_fixed_bit_field()
and store_split_bit_field().  (This same recursion occurs in the original
case, but with bitsize equal to 0, store_split_bit_field() succeeds in
moving zero bits.)

	Basically store_fixed_bit_field() and store_split_bit_field()
combination are not robust in the face of zero alignment.  And, I do not
understand why the alignment of this object is zero.

	GCC probably should abort much earlier in the case of zero
alignment as a self-check, but I am not sure how high up the call chain.
expmed.c:store_bit_field()?  expr.c:store_field()?
expr.c:expand_assignment()? 

Thanks, David

	P.S. This is relative to the 2000-08-08 tree.