Re: [Qemu-devel] [PATCH v3 02/10] main-loop: switch to g_poll() on POSIX hosts

[Paolo Bonzini]

Il 04/02/2013 18:51, Fabien Chouteau ha scritto:
> On 02/04/2013 06:35 PM, Paolo Bonzini wrote:
>> Il 04/02/2013 16:07, Stefan Hajnoczi ha scritto:
>>>>> No idea why the windows version has been mixing g_poll() and select().
>>>>> I'd hope that this series kills select() for uniformity's sake, but the
>>>>> 00/10 subject and the commit msg for this patch indicate otherwise.
>>> This is why I CCed Fabien.  I left the g_poll-followed-by-select
>>> behavior.  It may be to do with Windows treating sockets different from
>>> other objects.  Paolo may know the answer, too.
>>
>> Windows uses g_poll for HANDLEs and select for sockets.  All sockets are
>> configured to trigger an event handle when there is a change in the
>> socket availability, so g_poll exits and select is reexecuted.
> 
> Yes, the reason for this select() call is that we know when an event
> occurred on one socket, but we don't know what kind of event and which
> socket.
> 
> Would it be different if we had a HANDLE for each socket, instead of
> one for all?

Nope, I think you cannot assign different events for read and write.

Paolo