Re: [PATCH][2.6.6-rc3] gcc-3.4.0 fixes

[Mikael Pettersson <mikpe@csd.uu.se>]

H. Peter Anvin writes:
 > Mikael Pettersson wrote:
 > > 
 > > You're assuming pointers have uniform representation.
 > > C makes no such guarantees, and machines _have_ had
 > > different types of representations in the past.
 >  >
 > 
 > By the way, I am not in any shape, way or form making that assumption - 
 > although that's presumably how it would be *implemented* in an architecture 
 > with sane pointers like, to the best of my knowledge ALL gcc targets.
 > 
 > (foo *)bar++;
 > 
 > ... should be implemented as ...
 > 
 > ({
 > 	foo *tmp1 = (foo *)bar;
 > 	tmp2 = tmp1 + 1;
 > 	bar = __typeof__(bar)tmp2;
 > 	tmp1;
 > })

I did an experiment with updating an unsigned short
via a cast-as-lvalue to unsigned char, and gcc did
in fact implement the temp + copy semantics you
describe above. That is,

	unsigned short x = 0xaafe;
	((unsigned char)x)++;

resulted in x == 0x00ff.

So cast-as-lvalue is at least reasonably correctly
implemented in gcc.

Whether it's useful and portable is a another matter :-)