Re: Discussions in IETF WGs

[[e-mail] (Martin Rex)]

Abdussalam Baryun wrote:
> 
> For example, in one of the WG discussion on list, two members of WG
> have referenced a history-discussion and informed me to read them
> regarding some subject, I did do that but was *lost in translation*. I
> now think that the memebrs' advise was to a wrong direction. We SHOULD
> NOT refer in our current discussions to any other
> history-subjected-discussions (thoes discussion had no approve by WG
> consensus nor IESG review) in any WGs.  Also referring to old
> discussions in the list result to waste time and MAY make current
> arguments long (i.e. long means more than 5 working days), or even
> makes the current argument unproductive. Old-discussions MAY be
> misleading/incorrect/invalid, even if they are helpful to gain some
> knowledge.

There is no substiantial difference between old discussions and recent
discussions.  Referencing an argument from an earlier discussion rather
than repeating it is often an improvement with a significant potential to
actually save time, and zero risk to waste time (compared to repeating
the exact same previous arguments in a new message).


> 
> We should *reference* mostly RFCs in our discussion, because RFCs are
> correct resource.

If RFCs were correct, we would neither need an Errata process, nor
maturity levels, nor -bis documents.

"RFCs are correct" is a bold and dangerous presumption.  Many RFCs
are vague or even ambigous, and they may contain numerous implications
that are non-obvious to a number of readers, and sometimes non-obvious
to implementors and document authors).

All that can be said about standard track RFCs, is that they're the result
of the IETF consensus process.  And the primary focus of the IETF consensus
process is to resolve dissent (technical or procedural objections),
_not_ correctness.  Some level of confidence about correctness can be obtained
by creating independent implementations and demonstrate that they interop
on the implemented features, but results are mixed, and most interop
tests are lacking (test only a fraction of the features and only a fraction
of the implementations in the installed base, and rarely "border cases"
of unusual encodings).


-Martin