Re: (DNS) consensus wording

jg@w3.org:
>
>This sparked no fewer than 16 messages to the mailing list since I posted
the
>draft on Tuesday.
>
>However, no one argued with the requirement itself; all comments appeared
>to be related to the implementation of the requirement rather than the
>requirement itself.  There wasn't even any comment on the wording.

I think you are mis-interpreting the responses to your message.

I do not interpret the messages as not arguing with the requirement
itself.  The first comment was "It is not as simple as it might
seem.".  I think "it" was your "HTTP clients MUST observe the TTL
(Time To Live) information reported by the name server.".  The second
comment was that writing code for observing this information now would
be "a terrible idea given how quickly and continuously BIND has been
evolving".

The subsequent discussion convinced me that, if we put this MUST in the
spec, it is very well possible that we require more effort from
browser authors than most can afford to spend, especially given the
fact that their code may require updating to interface with a more
secure DNS system one year from now.

The result will be creative mis-interpretations of this part of the
spec in order to achieve compliance with your "must", or, even worse,
browser vendors taking a lot of time hacking together low level DNS
interface code and then strongly resisting any change at this
interface level, so that DNS can't evolve into a more secure system.

[...]

>Please let me know if you believe consensus for this requirement does not
>exist, 

I believe consensus for this requirement does not exist.

I propose that the "must" is changed into a "should" or "are strongly
encouraged to", and that text is added requiring some reasonable
things of clients which can't get TTL.

>and please let me know if you believe there is better wording
>possible.

Included below.

>                                - Jim Gettys

[....]
>If clients cache the results of name lookups for performance
>reasons, HTTP clients MUST observe the TTL (Time To Live) information 
>reported by the name server.

Change to:

 If clients cache the results of name lookups for performance reasons,
 they are strongly encouraged to get and observe the TTL (Time To
 Live) information reported by the name server.  If this is not
 feasible, clients must not cache the results of name lookups for
 longer than 10 minutes, and must immediately discard a name lookup
 result if a network error occurs when using the result to initiate a
 connection.


Koen.

Received on Saturday, 30 March 1996 05:24:15 UTC