Re: [certi-dev] Thoughts on IPv6 support

[Eric Noulard]

2010/5/4 Gotthard, Petr <address@hidden>:
> Hi,
> May be someone else has a better insight, but from my perspective it's
> not very difficult and manageable with the current CERTI.

I do agree with this.

> The IP dependent code is quite separated from the "core functionality",

CERTI has a "Socket" class in libCERTI with derived classes:

Socket
   SocketMC (multicast)
   SocketTCP
        SecureTCPSocket
        SocketHTTPProxy
   SocketUDP

There is a design anomaly with the "SocketUN" class which unfortunalety
does not derive from Socket.

Currently certi uses:
     SocketTCP object for RTIA <--> RTIG "Reliable" communication on
all platforms
     SocketUDP object for RTIA<--> RTIG "Best Effort" communication on
all platforms
     SocketUN object for Federate <--> RTIA
          SocketUN is "Unix Socket" on Unix and "TCP Socket  on Windows"

SocketMC is there efficient collective "BestEffort" communication
but it has been untested for a while so I do not really know its
current functional status...

SecureTCPSocket was a GSSAPI usage for secured WAN federation we used
in the past.
SocketHTTPProxy was added by Petr for handling connections over a proxy.

I would say that adding IPV6 support should be doable with some extra
kind of polymorphic
Socket object in most of the code, after the specific creation CERTI
do "only" send/receive
using the provided objects.

> so it can be changed quite easily. There might be problem with Windows
> as winsock does (or at least did few years ago) not implement all POSIX
> functions.

Don't know for that.

I'll have to have a look at "source-code level" IPV4 vs IPV6,
may be a starter would be:
http://people.redhat.com/drepper/userapi-ipv6.html

on windows there seems to be starting points on MSDN too:
http://msdn.microsoft.com/en-us/library/ms738649%28VS.85%29.aspx


-- 
Erk
Membre de l'April - « promouvoir et défendre le logiciel libre » -
http://www.april.org