Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

Re: [P-1788]: Anything else to be said about interval overlapping?

Dan Zuras Intervals wrote:

Date: Fri, 19 Nov 2010 08:30:39 -0600
From: Ralph Baker Kearfott <rbk@xxxxxxxxxxxx>
To: stds-1788 <STDS-1788@xxxxxxxxxxxxxxxxx>
Subject: [P-1788]: Anything else to be said about interval overlapping?

P-1788 members:

I have noticed that Motion 21.2 is under discussion until
November 21, at which time we will begin a vote.  Is anything
else to be said about this motion before it is frozen and
we begin voting on it?  Also, Juergen, am I correct concerning
this motion's status?

Best regards,

Baker
--


	Folks,

	In 754, the 28 comparisons that an implementation is
	required to support were listed & organized W.R.T. how
	they affected 4 state bits {<, =, >, ||}.  As a result,
	most implementations ACTUALLY created those 4 state bits
	to represent the results of a comparison.

	This inadvertent writing style had the effect of creating
	global state, side effects, & interlock requirements that
	were, frankly, not appreciated at the time.  We live with
	that mistake to this day.

	It would have been better if we had said, "This is the
	list of comparisons you are required to support" & put
	all the rest in an informative annex.

	Motion 21.2 is much the same.

	It would be a shame if future generations of computers
	ended up creating & maintaining 13 new bits of state to
	support comparisons.  Not fatal, really.  But it might
	inhibit innovation in ways we do not yet understand.

	I would organize our description of comparisons in much
	the same way as we should have done in 754: The list in
	the normative text & the states in an informative annex.

	But with no more clear alternative, I admit this opinion
	lacks credibility.


The credible alternative is to require only the three predicates
subset, interior, and disjoint (and perhaps equal).

As argued before, the others are virtually useless, and can be
created from their definition by the very rare users who needs it.


Thus my recommendation is (in view of our peculiar voting rules)
that you state that you don't vote -- a course of action recently
legitimated by our committee leaders, which acts as a more powerful
way of voting No.


Arnold Neumaier