[Masque] Masque needs its own encrypted ALPN

[Christian Huitema <[e-mail]>]

I hear the argument that Masque should use the H3 ALPN, in the name of
"not sticking out". But that argument would not apply if the ALPN was
encrypted, which I think it needs to be.

In the absence of encrypted ALPN, the "do not stick out" argument leads
to running every protocol on top of H3, then possibly doing strange
things with H3. For example, some application protocols care a lot about
the mapping of their protocol elements to QUIC frames and packets. But
if you run these protocols on top of H3, you should only use H3 frames,
whose mapping to QUIC frame is in theory arbitrary. If you insist on
running on top of H3, then you end up having to implement some QUIC kind
of frame mapping control inside the H3 stack. If multiple applications
have slightly different needs, that can become ugly quickly.

The ECHO proposal will encrypt the ALPN as well as the SNI. I understand
that we don't want to tie Masque development and ECHO development, but
we don't really have too. We just have to assume that when ALPN
encryption will become a concern, solutions like ECHO will enable ALPN
encryption. For now, it is sufficient to define a Masque specific ALPN,
and to specify frame mapping to QUIC  without being encumbered with H3
compatibility.

-- Christian Huitema