On Mar 9, 2006, at 2:22 PM, Mike Orr wrote:

> On 3/9/06, David Binger  wrote:
>>
>> On Mar 9, 2006, at 10:30 AM, Graham Fawcett wrote:
>>
>>> On 3/9/2006 10:19 AM, Graham Fawcett wrote:
>>>
>>>> On 3/8/2006 8:59 PM, David Binger wrote:
>>>>
>>>>> There isn't any difference between QP and Quixote with respect
>>>>> to compatibility.  The scgi server used by both uses passfd, which
>>>>> does not run on Windows.
>>>>
>>>> But the protocol doesn't require *forking*, right? It's an
>>>> implementation detail. Writing a less-efficient but stable-and-
>>>> semantically-equivalent SCGI implementation for win32 shouldn't be
>>>> a total show-stopper, just a fun weekend challenge for
>>>> someone. :-).
>>>
>>>
>>> Oops, replace 'forking' with 'passfd'.  Still true, n'est-ce pas?
>>
>>
>> Right.  Any volunteers?
>
> This would be a synchronious server then or a multithreaded server?

The server in qp dispatches connections to a pool of subprocesses,
using exactly the same pattern as Quixote's scgi server.
If there is a way to get similar behavior on Windows, that would
be my personal preference.

I think we could make QP work with a multithreaded server if we
modify the set_publisher()/get_publisher() functions to store
and maintain a separate Publisher instance for each thread.

I think I'll go ahead and do that anyway to make it easier to try.

It would also be cool if we could make QP and Durus twisted-friendly.
Would that be difficult?