[tsvwg] Some personal opinions on the state of UDP ECN socket APIs

[Martin Duke <[e-mail]>]

I don't want this to go in the draft, but here are some things l like about
the various platform APIs, which would feature in a converged solution:

1) I like that on an Apple v6 socket, if it's dual stack the IPv4 stuff
"just works" without another setsockopt() call.
2) I like that the Windows API uses IP_ECN, etc., rather than TOS, to
formally separate ECN from DSCP. (I am told that Apple is the same under
the hood, although that's not clear from the name). When the TOS byte is
treated as a whole, then code that is setting ECN has to do a getsockopt()
to find out the DSCP bits and setsockopt() to send DSCP back with the
correct ECN. This is very annoying. I will say that RFC3542 (informational)
suggests IPV6_TCLASS, and I would be happy to help fix that, if it's an
obstacle for kernel developers.
3) I like that Windows has a wrapper function called WSASetIpRecvEcn()
which takes care of all the v4/v6 stuff for you, so you don't need all
these if statements. I don't like that it doesn't take care of both IP
versions on a dual stack socket, which seems like a violation of the idea
that it "just works."

I don't know if the 3-4 OSes can get together and sort this out, but I
would love for my I-D to be obsolete.

Martin