[Stds-754] The names have been changed to protect the innocent...

[Michel Hack (1-914-784-7648) <hack@xxxxxxxxxxxxxx>]

Comments on Dan Z's changes.

  page 13, 3.2.4 canonical encoding:

This applies to NaNs too, so should use "floating-point datum".
(2 instances)

  page 30, 7.2:  (I see it on page 28)

This section (decimal exponent calculation) is much clearer now!

  page 32, 7.3.2:  (I see it on page 29)

It would still help to say "For finite decimal operands...", since
Infinity and NaNs are mentioned separately for quantize().

  page 39, isCanonical:  (I see it on page 35)

Section 7.7.2 keeps mentioning "x" which is however not defined here.
Perhaps "x" should become "the operand".  (Unrelated to Dan's changes.)

  page 44, 7.10   (I see it on page 40)

Now I'm getting uncomfortable.  The revised text reads fine, but it
makes me wonder exactly what a "floating-point datum" is (and we have
no glossary entry yet -- that could be exciting).  What bothers me
here (and perhaps I should have reacted already for Table 1) is that
FPD appears to stand for the "entity", which, in the case of numbers,
refers to the value and not to the representation.  Yet I thought that
what we had voted on was a replacement for "Nancy", which I understood
to be what can be found in a floating-point variable, encompassing all
that is encoded therein, so that DFP 1e2 is not the same FPD as 100e0
even though they have the same numeric value.  (Perhaps I misunderstood.)

I guess this is an almost-unavoidable level-confusion issue, which would
affect any other choice for "Nancy".

I note that the totalOrder() definition introduces yet another technical
term, "canonical member of a format".  In other contexts this has been
called "representation of a floating-point datum" (as distinct from
"encoding"), except for the restriction to "canonical" encodings.
(I'm not complaining, btw.)

In passing, I note that the effects of Jim's motion to clean up 7.12.3
have not been reflected yet.

All in all, this is a very positive development.  Thanks, Dan!

Michel.
Sent: 2006-09-11 20:34:42 UTC