On Mon, Jun 27, 2005 at 05:11:10PM -0400, Andy Dougherty wrote:
> Well, I think there are already way too many pointer casts and related
> games in the source. Perhaps more to the point, not all casts are going
> to work.
Well, there is that. :-)
In this case, as I understand it, we're looking at one pointer where
some data are above the pointer and other data are below it. (I'm
basing this on my memory of a design discussion in Austria; if I'm
mistaken, then ignore the rest.)
Given "struct below" and "struct above", this should be portable C:
struct group {
struct below b;
struct above a;
};
struct group *g = malloc(sizeof *g) /* OR WHATEVER METHOD OF ALLOC */
/* now take an 'above' pointer, so access to 'a' is fast & easy */
struct above *a = &g->a; /* addr is (char*)g + offsetof(struct
group, a) */
/* now make macros that easily let us point from the 'above' to the
'below' */
#define GROUP(a) (struct group *)((char *)(a) - offsetof(struct group,
a))
#define BELOW(a) (&GROUP(a)->b)
struct below *b = B(a);
Look, Ma! No unions! And no worries about alignment, either.
(Of course if this were C++ we could just do multiple inheritance and
let the static_cast<> template handle it. But whatever. :-))
> In particular, the last time I checked with gcc on SPARC, I got over 7,000
> warnings of the form 'cast increases required alignment of target type'.
Then gcc should just shut up, IMO. When I cast from void* or char* to
whatever*, it's my lookout as programmer if that's an error, and
there's nothing about it that deserves a compiler warning ... mostly
because there are some things in C that require it, and the compiler
can't possibly distinguish the good from the bogus. It's like warning
about '=' in boolean expressions. Yeah, I know. :-P
> More importantly, though I wonder if that particular micro-optimization
> is important to do at this point in parrot's development.
Probably not. It's more important pedagogically. We[*] are learning
about C by running into what doesn't work.
[*] not the editorial 'we' this time
--
Chip Salzenberg <[EMAIL PROTECTED]>
