Joel C. Salomon
5/16/2011 7:51:00 PM
On Monday, May 16, 2011 3:10:22 PM UTC-4, Seebs wrote:
> On 2011-05-16, Joel C. Salomon <joelcs...@gmail.com> wrote:
> > Digression: I've been wondering whether it's possible for a
> >compiler for a bit-addressable machine to get away with tightly-packing
> >_Bools and still somehow get away with lying that sizeof(_Bool)==1.
> >The only way I can think to detect this would be something like
>
> > bool bits[16] = {0,0,0,0, 0,0,0,0, 1,1,1,1, 1,1,1,1};
> > bool bp = bits;
> > char *cp = (char *)bits;
> > assert((void *)bp == (void *)cp);
>
> > -- but is that even mandated?
>
> Hmm. Here's my thought. In such a case, consider the two objects
> bits[3] and bits[5]. Obviously, sizeof(bits[3]) can't be less than 1,
> and CHAR_BIT can't be less than 8, so we know for sure that there are
> 8 consecutive bits which are part of bits[3]. Same for bits[5]. But
> their storage spaces, it seems to me, clearly overlap. (If not, look
> at bits[4], which will overlap one of them.)
I thought I was missing something obvious; thanks.
I guess the way to "properly" permit this optimization is to use "register _Bool" for individual variables, and bit-fields instead of arrays. (At least post-TC2, see N1356.)
--Joel