Re: [patch] futex API cleanups, futex-api-cleanup-2.5.69-A2

[Linus Torvalds]

On Mon, 19 May 2003, Ingo Molnar wrote:
> 
>  - start the phasing out of FUTEX_FD. This i believe is quite unclean and
>    unrobust, because it attaches a new concept (futexes) to a very old
>    (polling) concept. We want futex support in kernel-AIO, not in the
>    polling APIs. AFAIK only NGPT uses FUTEX_FD.

This sounds like a bad idea. 

Expecting "select()" and "poll()" to go away, and calling them "unclean
and unrobust" is just silly and stupid. There are a hell of a lot more 
programs using select-loops out there than there are AIO versions, and I'd 
argue that AIO is likely to be the much more "unrobust" solution, and 
probably doesn't even scale any better than using epoll.

In fact, it's hard to see any real advantages of aio over a sane polling
loop, as long as the polling doesn't have some O(n) overhead (in other
words, as long as you use epoll).

So stop pushing your own agenda and broken morals down other peoples 
throats, and re-do this patch properly.

		Linus