t-codings and identity

jg@pa.dec.com (Jim Gettys) writes:

>>    15) (14.48) The production for <t-codings> doesn't allow "identity", but
>>        rule 3 seems to allow "identity;q=0".
>>
>Here's Henrik's clarification:
>it is not that "identity" isn't allowed - the problem
>is that chunked is special - due to backwards compatibility concerns.
>
>...

I see where I got lost now.  The production for <t-codings> allows
"chunked" or <transfer-extension>, but I missed the fact that in this
case, we use "-extension" to include choices defined in the spec.  In
all other cases we use "-extension" for extensions beyond the current
spec and define current values in the BNF (viz. <accept-params>,
<cache-response-directive> et al.)  So while the BNF is technically
correct, it deviates from normal usage in the rest of the spec.

I can live with leaving <t-codings> as it is, but I wonder if others will
also miss the subtly different use of "-extension" in transfer-extension.

Ross Patterson
Sterling Software, Inc.
VM Software Division