Keith Moore wrote:
I would strongly object to a change to our process that removed
the requirement to demonstrate interoperability.
If we need additional incentives to advancement, perhaps we should
require that proposed standards revert to informational or historic
if no action is taken within three years.
(action being: recycle at proposed, advance to draft)
As I understand it, the original draft requirement (demonstrating interop)
was to improve the specification in several aspects:
- ensure that there are implementations of the specification,
because implementing a spec is a good method to uncover inconsistencies,
ambiguities and contradictions
- ensure that there are multiple independent implementations in order
to find out whether independent implementers understand the spec
in the same fashion.
- find out which of the (interoperable) features of a spec are necessary
and which one are more in the direction of bloat
Recycling a spec on proposed will primarily add new features, and include
errata and potential clarifications, but rarely drop features.
Therefore I'm no sure that "recycling" at proposed should be considered
a valid substitute for a demonstration of interoperability between
independent implementations. At least, there should be a limit as to
how often recycling a spec at proposed should "exempt" a working group
from demonstrating interoperability.
Otherwise, there might grow a disconnect between the most recent spec
and what is most commonly used on the internet. The TLS protocol is
a victim of the "recycle at proposed" rather than performing serious
interop testing. With the result that the most recent spec is
TLS v1.2 (08/2008), whereas the protocol version that is most widely
used, and the _only_ TLS protocol version that can be safely used by
a client that does not implement an application-level reconnect
fallback is TLS v1.0 (01/1999). If the forward extensibility options
in the TLS protocol had been more appropriately interop tested in
the years after TLSv1.0, this problem would likely be *much* smaller.
The PKIX Interet Certificate and CRL Profile has similar being
recycled at proposed serveral times (2459,3280,5280), and has
grown feature bloat rather than seen cleanup & feature reduction.
Ietf mailing list