New flash! Bob and I agree!
On May 21, 2005, at 8:53 PM, Bob Wyman wrote:
If I have two non-identical entries, each with the same atom:id
and
the same atom:updated, I need an algorithm that can be used to
determine
which of the two entry instances should be presented to a consumer
who only
wishes to see one entry. It would be desirable if the decision
could be made
in a non-random fashion. If asked, I think 99% of consumers will
say that
they would prefer if I present them with which ever of the two
entries was
"most recently" created. The problem is that Atom does not permit
me to
satisfy this user request. Without atom:modified, I can't determine
which of
the two entry instances was "most recently" created.
Bob is right.
IF you find an two entries with the same atom:id and the same
atom:updated
AND IF you are inclined to believe that the difference will be
interesting
AND IF you believe that date-stamping is a useful way to resolve
this issue
THEN atom:modified gives you a decision procedure.
However, I believe that this kind of collision is an anomalous rare
case, and I believe that most times it happens they're going to be
actual duplicates that are exactly the same, and I believe that if
there are differences, if the publisher is to be trusted the
differences are not going to be interesting, and I believe that if
they are interesting, there's a chance you're looking at the
potential DOS attack that Bob highlighted, which case atom:modified
is not much (any?) help.
Plus, atom:modified is actively harmful because it encourages people
to believe (falsely) that there is consensus on what an "update"
means (remember the thousands of messages we invested chewing this
over because certain WG members from Norway had a violently
incompatible mental image-entirely correct for their app-of what a
"change" was?), and also to believe (falsely) that they don't have to
worry about whether to trust a feed provider.
-1 on atom:modified.
My opinion on this is not weakening. -Tim
