[GSoC - tcptest] - Regression Tests, Conformance Tests...

[Bruce M. Simpson]

Victor Hugo Bilouro wrote:
>> I've made a lot of changes to it; diffs are with him but I can send folk a
>> copy of my Mercurial repo.
>>     
> I would appreciate that.
>   
Sent (off-list).

As an example of the new PCS syntax and expect() stuff, I'll forward you 
the IGMPv2 test off-list. (Also sent.)

> humm, track state is needed to make TCP tests.
>   

It is something you'll have to build yourself around the expect() 
functionality.

The experimental IP reassembly code (in pcs/packets/ipv4sar.py) might be 
a good place to start. It isn't finished, but it should demonstrate the 
general principles -- i.e. you read packets in a loop and you pass them 
to an object which knows what to do, in this case, ipv4sar.

One big problem I had was that the concept of fragmentation requires 
deep copies of PCS objects. I imagine that's less of an issue for TCP 
segmentation, as the situation is made somewhat easier by the fact 
you're dealing with streams.

BTW: My snapshot of PCS fixes the IP and TCP option parsers. If you look 
at the IGMP and DHCP decoders, there is an example of a dictionary 
driven option parser. This could also be applied to TCP where it's 
likely to be useful.

I believe most of the bugs have been shaken out of expect(). The main 
problem is buffering and the fact that expect() depends on non-blocking 
I/O. pcap can return more than one packet from the kernel every time you 
call into the non-blocking dispatch function, so I did some internal 
refactoring to allow expect() to deal with that.

So your code has to be able to deal with multiple matches from the 
Connector, even if you only asked to match at *least* one packet. 
"Count" is mostly about stopping expect() from hanging the flow of 
control anyway.

The syntax and semantics are intentionally similar to PExpect for 
Python. In fact the IGMPv2 test uses PExpect to drive a QEMU virtual 
machine encapsulated as a Python object, for regression testing the IGMP 
code. So my suggestion is check out PExpect too.

> I didn't find his site, can you send me?

http://www.fsmware.com/freebsd/syntest2.py

I've added some Scapy-like syntax to PCS which can make the code look a 
bit smaller.

cheers
BMS