apr-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Paul Querna <c...@force-elite.com>
Subject Re: svn commit: r178386 - /apr/apr/trunk/build/apr_hints.m4
Date Wed, 25 May 2005 07:10:41 GMT
Joe Orton wrote:
> On Tue, May 24, 2005 at 11:28:50PM -0700, Paul Querna wrote:
> 
>>jorton@apache.org wrote:
>>
>>>Author: jorton
>>>Date: Tue May 24 23:24:19 2005
>>>New Revision: 178386
>>>
>>>URL: http://svn.apache.org/viewcvs?rev=178386&view=rev
>>>Log:
>>>* build/apr_hints.m4 (APR_PRELOAD): Prevent use of poll() on Darwin.
>>
>>I believe this should only apply to Darwin <8.0 or Mac OSX <10.4.0.
>>
>>poll() is known to be broken on the older versions, but it appears to
>>work correctly in 10.4.
> 
> 
> Al Begley @ Apple does not concur:
> 
> http://issues.apache.org/bugzilla/show_bug.cgi?id=34332#c15
> 
> neither does the fact that everywhere I keep reading that poll() is
> "deprecated" on Darwin and is implemented as a wrapper for select() in
> libc.  Or did that change in 10.4?

As far as I know, it changed in 10.4.  The man page no longer mentions
depreciation, and it seems to work correctly.  It would be nice to get
an 'official' word out of someone @apple.com.

>>This could all be mute if we added a KQueue backend for apr_poll().  I
>>had the code to do it at one time, I will look at adding it later this week.
> 
> 
> apr_poll() is not used by apr_wait_for_io_or_timeout().
> 

Yep. It uses poll() directly, or a pollset.  Maybe we should just
preload WAITIO_USES_POLL to 0 on darwin, instead of killing poll completely?

> But making apr_poll() use a kqueue-like interface is a really bad idea
> anyway, as I've argued before.  apr_poll() should be efficient to use
> occasionally or for small numbers of fds; and will be necessarily faster
> than apr_pollset_* in those cases since it requires just one system call
> - no overhead for setting up and tearing down the kqueue/epoll/... 
> interface.

Perhaps, but if they could be cached, and re-used, it is faster if we
will be doing it multiple times.  On FreeBSD, I commonly saw kqueue to
have about 20% lower latency than poll().

-Paul

Mime
View raw message