apr-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "James Mansion" <ja...@wgold.demon.co.uk>
Subject RE: [WIN32] alternative apr_pollset implementation proposal
Date Sat, 30 Apr 2005 10:43:36 GMT
Bill,

I think there are any number of cleaner approaches than emulating the
stupid poll and select semantics - mostly looking more like aio with
some sort of completion event, so the kernel just gets on with things.

But isn't the problem at hand that a scalable implementation is needed
under the existing poll interface?

Personally I would suggest making legacy apps that use that interface as
scalable as possible across platforms, and then adding a further interface
that can unify IO in the more object-oriented manner you suggest.  Maybe
at some point poll can be deprecated - but I'd be surprised if that was ever
acceptable.

Which is odd in a way really, because in every application I've ever
written, I've only ever really run app logic when IOs complete, or
when some event has caused me to generate and queue data to send.

It may be that once buffer space was so tight that data to send was
generated only when the kernel might accept it, and input buffer
space was similarly scarce, but I doubt that is the case now even for
embedded developers - or at least the ones who might use APR.  Those
doing IP with a PIC may see things differently.

James




Mime
View raw message