On 01/13/2012 06:05 PM, Paolo Bonzini wrote:
> I think it's not entirely correct because the cast in QTAILQ_PREV and
> QTAILQ_FOREACH_REVERSE does not look like valid ANSI C. No matter how
> hard I look I admit I cannot figure out how it works, but anyway I
> suspect it can be changed to ANSI C using typeof if one was bitten by
> it. So removing QCIRCLEQ is not a bad idea anyway.
Ah, got it. Here are two ways to rewrite it using typeof (not exactly
ANSI C of course, but perhaps somewhat more aliasing friendly).
#define QTAILQ_PREV(elm, next) \
(*(((__typeof__((elm)->next) *)((elm)->next.tqe_prev))->tqe_prev))
#define Q_TAILQ_PREV(tqe) \
((__typeof__(tqe) *)((tqe).tqe_prev))
#define QTAILQ_PREV(elm, next) \
(Q_TAILQ_PREV(*Q_TAILQ_PREV((elm)->next))->tqe_next)
It's treating the TAILQ as a half-linear (forwards), half-circular
(backwards) list. Doing two backwards accesses and one forwards
access ensures that the last access is always on a pointer rather
than a pointer-to-pointer. Clever.
Paolo
