Notes for meeting at Intel in Sunnyvale at 1:00 on Thursday
	9/9/04.  Jeff Kidder hosted us in room 231.  Prof Kahan, & Dan
	Zuras attended.

	We began again with Annex Z.

	My split infinitives & language in general was corrected.  Many
	statements were made more clear.  We discussed examples of
	applications which make the need for arbitrary precision clear.

	Jeff argued that perhaps we should provide references to a list
	of secondary sources to much of the mathematical problems we
	are trying to use as examples of numerical or roundoff problems.

	I will put the example that Prof Kahan provided into the
	rationale section.

	We took our afternoon break around 4:00.

	After the break we went to alternate exception handling.

	Jeff proposed that a caller specifying break-on-invalid should
	PERMIT the callee to break out early as soon as it knows that
	an invalid has happened.  A caller specifying a break-on-0*inf
	would permit the callee to use its default.  Same with caller
	specified 0*inf<-7.0 as a presubstitution.

	I think that specification of a transfer of control on error
	should be different (in its inheritence properties) from a
	presubstitution on error.  I don't know if that means I agree
	with Jeff or not.

	Jeff initially wanted to do this with specified NaNs (eg: the
	0*inf NaN) but it looks like we're trying some other method.

	We spent some time discussing the difficulties of the case
	analysis of x^y which illustrates many of our problems.