Re: axfr-clarify's fraudulent claims of consensus

[gson@nominum.com (Andreas Gustafsson)]

Dean Anderson writes:
> Despite the "suggestion", all implementations prior to Bind 9
> made different (but ultimately identical) assumptions.

Rob Austein already pointed out JEEVES and CHIVES as counterexamples,
and before I worked on BIND, I wrote an authoritative-only server
which also used separate data structures for each zone.  When I
started working on BIND 9, the BIND 9 team had already independently
made the same design decision.  To both me and the BIND 9 team, it was
the obvious design, and the BIND 8 behavior was an obvious
implementation bug.

> This is nothing less than bad or lazy engineering.** As I
> mentioned previously, IXFR is in principle no different from an Oracle
> incremental replication. Clearly, one does not need to alter SQL to have
> replicated databases.  One _could_ alter SQL, but it would be a "bad
> idea".

If a version of Oracle had a bug that caused tuples to "leak" between
tables under certain conditions and the bug was only noticed when it
broke incremental replication, would you fix the bug or try to
redesign the replication protocol around the bug?  Would you do
differently if 77% of deployed Oracle servers had the bug?

> [more gratuitous insults that I'm not going to respond to]

> So perhaps no servers need changing if we standardize AXFR as I
> propose.

No servers will need changing if we standardize AXFR as I propose,
either.  There is no AXFR interoperability issue either way.  The
only difference (other than IXFR working better) is that a different
set of servers is declared non-compliant.
-- 
Andreas Gustafsson, gson@nominum.com