Re: [patch 14/22] pollfs: pollable futex

[Davi Arnaut <davi@haxent.com.br>]

Andrew Morton wrote:
> On Sun, 6 May 2007 00:50:47 -0700 "Ulrich Drepper" <drepper@gmail.com> wrote:
> 
>>> I really do not understand your point. You're too smart to not appreciate
>>> the beauty and the simmetry of objects that responds to a common interface
>>> (our files, win32 handles), and that fits our existing kernel infrastructure.
>> You're blinded by this symmetry.  Not everything that looks like a
>> good fit is a good idea.  This is one case.  Get over it, poll is not
>> powerful enough to serve as the unifying event mechanism.
> 
> What is your position on the timerfd/signalfd/etc patches?
> 
> Seems to me that if we were to have fancy new event-delivery machinery
> like kevent then the timerfd/signalfd work is heading in the other
> direction and ultimately would prove to have been unneeded?

IMHO, I thought we had already gone down the *fd road with inotify,
posix message queue, and _hundred_ others file objects with poll methods.

I also think that inotify+(e)poll proves how well the fd/epoll model
fits together, scales, and that a new fancy event-delivery machinery is
not necessary. And it makes me wonder why I hadn't followed its "watch"
approach for futexes:

futex_init(); // Davide's anon fd
futex_add_watch(int fd, void *addr, int val, uint32_t mask);
futex_rm_watch(int fd, uint32_t wd);

Anyway, this unifying event machinery can be built, if needed, in user
space by libevent and others.

--
Davi Arnaut