Re: Ping^2: [patch] convert some call ABI macros to hooks, apply to sh, add bitfield swapper

[Richard Henderson <rth at redhat dot com>]

On Fri, Aug 29, 2003 at 08:08:10PM -0400, DJ Delorie wrote:
> > +   if ((* targetm.ms_bitfield_layout_p) (rli->t)
> > +       && (* targetm.reverse_bitfield_layout_p) (rli->t)
> > +       && ! TYPE_PACKED (rli->t))
> > +       reverse_bitfield_layout (rli);
> > 
> > Why would ms_bitfield_layout_p or TYPE_PACKED have anything
> > necessarily to do with reverse_bitfield_layout_p?  Seems like
> > these tests, if needed, should be pushed down into the 
> > reverse_bitfield_layout_p predicate.
> 
> Consider this case:
> 
> 	struct {
> 	  int i:4
> 	  char j:4
> 	} foo;
> 
> The layout normally goes like this (little endian):
> 
> 	i = bits 0..3 of SI at byte offset 0
> 	j = bits 4..7 of QI at byte offset 0
> 
> Now, asking gcc to put the bits at the other end of "the" underlying
> type becomes nonsensical, because there is no *one* type underlying.
> In order to even consider the reversal, there must be just one
> underlying type for each grouping of bitfields, and the conditionals I
> specified (in theory) do that.  IMHO gcc cannot honor a request to
> reverse them (reverse_bitfield_layout_p) unless those conditions are
> met, so there's no point letting the target decide.

And ms_bitfield_layout_p determines this?  What about 
!PCC_BITFIELD_TYPE_MATTERS?

Anyway, all of this needs to be a huge comment before that bit of code.


r~