Registration (was Re: DRAFT Minutes, HTTP-WG)

We've gotten some warnings from those folks who have been involved in
the message-header registration wars to tread carefully here. There's
a quite analogous situation between HTTP headers and message headers,
and quite a bit of experience in the IETF about what can go wrong.

For example, if you merely allow anyone to register anything, then you
get poorly specified headers, conflict over interpretation of
registered headers, vanity registration, trademark conflicts, etc.

On the other hand, if you require standards-track RFCs describing the
header before the header becomes registered, you wind up with a
cumbersome process which either interferes with experimentation, or in
which the registration mechanism is ignored, the experiment proceeds,
which then interferes with registration once the experiment is

Registration mechanisms are constrained by market, financial,
political, legal issues far more then technical ones.

The default registration procedure for anything that doesn't otherwise
have one is 'create a standards-track RFC describing the item'.
I think this currently applies to

    Status codes
    Entity Header Fields
    Content codings
    Transfer codings

although it doesn't seem to apply to content-type. In lieu of any
other registration mechanism, it currently applies to URL schemes.

It's not that I don't think this issue needs to be addressed; I think
registration is very important. It's not unreasonable also to ask that
IANA would maintain an authoritative version of such a registry, even
if the only way that items could be added to the registry was via a
'standards-track document'.

One way to make better progress in HTTP-WG is to avoid 'ratholes', I
think this is one. I would like to make the issue of devising new
registration procedures 'out of scope' for HTTP-WG.

Is there is a group of individuals who are knowledgable about the
history, status, and difficulties with the registration procedures for
protocol extensions in other Internet protocols (SMTP, SNMP, media
types, Telnet options come to mind) as well as HTTP and who are
willing to work on this issue, I might feel differently.

Received on Sunday, 10 December 1995 11:37:03 UTC