[
Date Prev][
Date Next][
Thread Prev][
Thread Next][
Date Index][
Thread Index]
Re: [Stds-754] nomenclature for representable entities
On Aug 29, 2006, at 12:24 AM, Mike Cowlishaw wrote:
Jim wrote (and others commented on):
Based on a quick look at all uses of "number" in the draft, it
appears
that most are intended to exclude NaNs, which is consistent with
Warren's suggestion. ...I think following Warren's suggestion will
require fewer and less awkward changes.
I think that's correct (fewer changes). Let's not get hung up on the
spelling -- it sounds as though what we need is a term for
"floating-point
numbers and NaNs" (which might refer to formats, encodings, or just
the
set of values).
Ah, I had missed that we were looking for something ambiguous.
If we want to be loose about what we are really trying to say then
"Floating-point value" seems like it does the job since it can be taken
to mean almost anything you want.
I'm not sure that 'representable entities' is best
because we have diligently avoided using that term in a general
sense, and
in particular we use it when we are specifically not talking about
encodings.
Yes, if we want to talk about "floating-point numbers or possibly NaN"
then representable entity or floating-point entity seem to do the job.
I suggest we choose a term, make the document consistent using it,
then
worry about the spelling afterwards.
Sure, let's first make sure we know exactly what the term will and
won't refer to.
Or exactly how vague we want to be about it, if that's the case.
E.g., are the following the same or different thing as far as this
definition is concerned:
-0, +0
sNaN, qNaN
2.0, 2.00
qNaN-with-payload(42), qNaN-with-payload(42),
888, "one of the noncanonical ways of writing 888"
(Possible terms include: Michel's 'FP entity', also perhaps 'FP
item',
'FP object', 'FP article', 'FP packet', 'FP blob' ...)
I tend to think that many of these imply that you are including
information
from the representation or even encoding.
In short, I like the following terminology:
Floating point entity = {-inf, +inf, representable finite non-zero
numbers, -0, +0, NaN}
Floating point numbers = {-inf, +inf, representable finite non-zero
numbers, -0, +0}
Finite [floating-point] numbers = {representable finite non-zero
numbers, -0, +0}
Signaling vs quiet NaNs are issues of representation as is 2.0 vs 2.00.
Payloads of NaNs and non-cannonical representations are all issues of
encodings.
In usage, we can decide if floating-point number can be used
correctly, or
choose between floating-point entity or "floating-point number or NaN"
Cheers,
Jeff