------- Comment #18 from matz at gcc dot gnu dot org 2008-01-09 10:32 ------- It seems that this should work too: bitsize is 8 in this example IIRC, so copy_mode would be byte_mode, which is what we need here. Additionally using word_mode when we can seems worthwhile too.
One thing would bother me, though: If TYPE_ALIGN(type) is very funny, then bitsize is too, so that mode_for_size could return BLKmode, so word_mode would be chosen to copy, which might then trip over the same problem. But for that TYPE_ALIGN must be so funny like 3 bits, i.e. something for which no INT mode exists. I believe it's impossible that TYPE_ALIGN is neither a multiple of word_mode nor a number for which mode_for_size ever returns BLKmode (i.e. either it returns BLKmode, but then only for very large aligns, which are multiples of word_mode, or it will return something != BLKmode, which we then can use). IOW: it's impossible that TYPE_ALIGN ever is something like 3*BITS_PER_UNIT. So, with that "argumentation" I think this would work, except for my probably theoretical fear from comment #9, that the mode argument for extract_bit_field is only a suggestion. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=31309
