Re: [lwip-users] Problem with lwip_close in TCP

[Christiaan Simons]

Hi Frédéric,

> After using this code to send my document, I do a lwip_close on my
> socket. The problem is that all my data is not sent if I don't do a
> "sleep" between my "Send" call, and the lwip_close. With a 10ms
> "sleep" before the lwip_close, all my datas is received. It seems
> that all "enqueued" tcp datas are dropped when I do the lwip_close.

I think this is related to
http://savannah.nongnu.org/bugs/?func=detailitem&item_id=15926
since the socket layer builds upon the netconn layer.

Since the netconn_close() is broken (by design),
the socket close() is probably broken as well.

I'm only doing very minor bugfixing on this code,
since I'm not the author. I've placed this in
the CHANGELOG todo list as well, since this will take
a lot of work to get it right.

Currently I'm working on a "raw API" SNMP agent
so I don't have time to work on the sockets.

For now I strongly discourage the use of both
the socket and netconn API. If possible try
to use the raw API, this is a little harder to
program but it gives the best performance and stability.

A crude example of this is in contrib/ports/unix/proj/minimal,
and there is a short intro in doc/rawapi.txt

Some users created their own private socket
implementations but none of these seem to have
made it into our contrib or lwip tree. :-(

Christiaan Simons

Hardware Designer
Axon Digital Design

+31 (0)13 511 66 66
+31 (0)13 511 41 51

http://www.axon.tv