[usocket-devel] Release 0.4.0

[Chun Tian (binghe)]

Hi, Douglas

Thanks very much for participate our discussion!

> Hello Chun,
>
> I will supply a patch to implement all these soon.

I think no patch will be better, if exist (even internal) functions in  
SCL can be used directly. This is quite different from the UDP support  
(most important interface functions lost), and USOCKET doesn't want to  
support future SCL versions only, exist versions will be supported if  
possible.

P.S. Of course you can export or add any functions in SCL future  
versions freely, we'll try to follow your steps:)

>
>
> The 'ext:connect-to-inet-socket timeout has been included for some  
> time, and this
> is the timeout for the connection attempt alone and does not affect  
> the returned
> socket.

Yes, I think that's the meaning of SOCKET-CONNECT's TIMEOUT keyword,  
just the timeout for the connection attempt alone.

In LispWorks, the COMM:OPEN-TCP-STREAM has three timeout-related  
keywords: TIMEOUT, READ-TIMEOUT, and WRITE-TIMEOUT. USOCKET's SOCKET- 
CONNECT just use the first, that's just good.

I think USOCKET should add SOCKET-OPTION and (SETF SOCKET-OPTION)  
interfaces for get/set the other socket options for post-created  
sockets, options which I can tell at this time will be:

  * SO_KEEPALIVE
  * SO_REUSEADDR
  * SO_RCVTIMEO
  * SO_SNDTIMEO
  * TCP_NODELAY

One of them (SO_RCVTIMEO) is also useful for UDP sockets.

>
>
> The stream timeout can be accessed using 'ext:stream-timeout, see:
> http://www.scieneer.com/scl/doc/function/ext/stream-timeout.html

What exact socket option does this function actually set? SO_RCVTIMEO  
or SO_SNDTIMEO, or both?

>
>
> The stream deadline can be accessed using the 'ext:stream- 
> expiration, see:
> http://www.scieneer.com/scl/doc/function/ext/stream-expiration.html

I see. Thanks!

>
>
> Adding a 'socket-set-option would also work well.  A 'nodelay  
> argument can be
> added to 'ext:connect-to-inet-socket and/or 'ext:accept-tcp- 
> connection if this
> is the preferred interface.

I think no need to add more keywords into EXT:CONNECT-TO-INET-SOCKET,  
for options we can set after socket creation, just use a separate  
function to set is just OK.

(The LOCAL-HOST and LOCAL-PORT keyword cannot be set after socket  
creation, so I requested to add them into EXT:CONNECT-TO-INET-SOCKET  
weeks ago.)

>
>
> There are internal functions: 'ext::set-socket-option, and 'ext::set- 
> socket-tcp-nodelay
> and one or both could be exported if it helps support usocket.

Great! And we (usocket) will try to use them (as internal functions).

>
>
> Regards
> Douglas Crosher
>
>
> Chun Tian (binghe) wrote:
>> Hi, Douglas
>> By reading SCL 1.3.8 documents, I saw the TIMEOUT keyword in  
>> EXT:CONNECT-TO-INET-SOCKET, so I guess just use this keyword will  
>> be OK to add support for TIMEOUT in USOCKET:SOCKET-CONNECT. more  
>> questions:
>> 1. From which version of SCL, the TIMEOUT keyword be added? All SCL  
>> version or there's some way we can detect whether TIMEOUT can be  
>> used? (i.e *features*, fboundp, ...)
>> 2. How to support DEADLINE in SCL? (I can find details from SCL doc)
>> 3. What if we (usocket) don't split out DEADLINE from SOCKET- 
>> CONNECT but add a new SET-SOCKET-OPTION to set the behavior of a  
>> already created socket?
>> Regards,
>> Chun Tian (binghe)
>>> Hello Erik,
>>>
>>> Thank you for the update.
>>>
>>> The nodelay socket option can be supported in the Scieneer CL.  It  
>>> may also
>>> be useful to add this option to 'socket-accept, or perhaps split  
>>> this out
>>> into a separate function that can be used to enable or disable the  
>>> Nagle
>>> algorithm.
>>>
>>> The Scieneer CL streams support both the timeout and the deadline,  
>>> but also
>>> allow these to be reset and for the remaining time to be read  
>>> which is important
>>> for many applications.  For example a web server can reset the  
>>> deadline at
>>> the start of a new request.  So perhaps this functionality could  
>>> be split out
>>> from 'socket-connect.
>>>
>>> Regards
>>> Douglas Crosher
>>>
>>> ...
>>>> A number of smaller features (options to SOCKET-CONNECT) have also
>>>> been added, however, these may not be available on all
>>>> implementations:
>>>>
>>>> * :no-delay, which disables Naggle's algorithm
>>>> * :timeout, which sets a per-operation timeout on the socket
>>>> * :deadline, which sets a maximum life-time to the socket
>>> ...
>>>> As always, any feedback - bugs or hugs - is greatly appreciated. If
>>>> you use usocket, please let us know what for!
>>>
>>>
>>> _______________________________________________
>>> usocket-devel mailing list
>>> usocket-devel at common-lisp.net
>>> http://common-lisp.net/cgi-bin/mailman/listinfo/usocket-devel
>