[lwip-users] [lwip] Back again...

[Adam Dunkels]

Hi all!

Due to other work, I have been away from the lwIP mailing list for a while and 
am currently in the process of catching up a little with the messages. There 
are over 140 unread messages in my lwip folder right now! Hopefully I'll be 
able to go through most of them today.

During my time off the list, I have been doing modifications to lwIP in 
preparation for a 0.6 version. Some of the changes are:

--- General ---
* The lwIP code has been too large for what it does. In the spririt of uIP, 
much of the code size can be reduced. I have been doing some of this already, 
especially with the TCP code.
* A new example project, proj/minimal/, will show how to reduce lwIP to its 
minimum configuration (i.e., with no operating system support, small buffers, 
etc.).
* A new API will be added to 0.6. This will be similar to how the uIP API 
works and will be in semantics as the raw API, but will not use separate 
callback functions for each event. It will use a single callback function 
that must be implemented by the application code. This is advantageous for 
those who will use lwIP to build another API on top of (similar to how Paul 
have been adding a socket layer on lwIP).

--- ARP ---
* Updated ARP implementation, with multiple queues for holding packets that 
are destined for addresses that are in the process of being resolved.
* ARP implementation now does almost all Ethernet-related stuff, which greatly 
simplifies the device driver implementation.

--- Memory management ---
* Previously, the memory managers included reclaiming functionality. When 
memory was scarce, the memory managers called registered reclaimer functions 
that deallocated some memory. This didn't really work as intended, and it 
turns out that it is much simpler to let the calling module take care of the 
problem itself - thus, reclaiming functionality is gone.
* The memp will be changed to macros instead of function calls. When 
statistics and debugging are turned off, the memp stuff will not produce any 
function calls at all.
* A configuration option will be added to support other memory managers as 
well.

--- Operating system support ---
* The sys_arch stuff is not needed when lwIP is run as a single thread (either 
in a multitasking system or without an operating system). In 0.6 there will 
be an option to turn all sys_arch code off.

--- IP ---
* Support for IP fragment reassembly will be included in 0.6.

--- TCP ---
* Quite a lot has happened with the TCP code. Parts has been greatly 
simplified and variables that previously was passed as function arguments are 
now made global to save stack space and code size (on certain platforms).
* TCP Urgent data will be supported in 0.6.
* Better separation of input and output functionality - when an incoming 
segment is being processed, no data can be sent. Instead, when input 
processing is over, the output function is called to send data.

That is the list of plans for 0.6. Hopefully, I didn't miss anything that I 
have intended to do.

/adam
-- 
Adam Dunkels <address@hidden>
http://www.sics.se/~adam

[This message was sent through the lwip discussion list.]