Re: [liblo] TCP

On Tue, Dec 23, 2008 at 8:27 PM, Stephen Sinclair <rad...@gm...> wrote:
> On Tue, Dec 23, 2008 at 4:46 PM, Steve Harris
> <S.W...@ec...> wrote:
>> On 23 Dec 2008, at 19:03, Stephen Sinclair wrote:

>>> Maybe Steve H. can tell me, but I'm not actually sure what advantage
>>> poll() has over select(), actually.
>>
>> Not really my field, but I used to work without shouting distance of a
>> load of network research people, I asked them and they said "use
>> poll(), it's better" or words to that effect :)
>>
>> I vaguely remember that select is only efficient with a small number
>> of sockets/FDs.
>
>
> Hmm.. :)  Interesting, I personally haven't used poll before at all.
> I looked into it, as I got curious.  It looks like select() is
> basically limited in number of sockets because it uses a bitmask.
>
> Looks like it wasn't necessary before, but if I'm to maintain a list
> of sockets it works well.  I can probably save some coding by using an
> array instead of a linked list and pass it in directly.  (I was trying
> to avoid shuffling array items around, but now that I think about it,
> it's probably a lot more efficient to shuffle an array when
> connections are removed, than to build it from a list every time
> lo_server_recv() is called.)
>
> I think for portability I might change it to use autoconf to detect
> poll() and use it, instead of just assuming it is there on non-Windows
> systems.

I generally use poll() because it comes from the POSIX standard.
Someone recently pointed out to me that POSIX now specifies select()
as well.  That was originally a BSD interface (IIRC).

I think both are probably OK for most purposes.  Poll still seems
slightly better because it handles processes with >32 open files.
-- 
 joq