[Cfp-interest 3298] Re: C26 Complex Related Changes

[Jim Thomas]

> Isn't the point that the canonical complex infinity is inf + i 0, while there are many values that will compute as complex infinity? It's quiet NaN all over again, where computations can produce a complex infinity that depends on the operands, but any new infinity created is canonical?

My understanding …

“Canonical” is a defined term in IEEE 754. Some values have multiple encodings, one of which is called “canonical”. Arithmetic operations return values with their canonical encoding. All encodings of a value behave the same as input. There’s no idea of different encodings having different semantics or including non-semantic auxiliary information.

Quiet NaNs are likewise all equivalent in the arithmetic, but different (from redundant encodings) in that they might contain non-semantic auxiliary information. Recommended behavior is intended to preserve such information.

Complex infinities (as defined in Annex G) are actually different values. Complex operations resulting in a complex infinity do not ordinarily return inf + i 0. Different infinities as input typically result in different real and imaginary parts of the result, as the operations use directional information in infinite real or imaginary parts. The differences have semantic significance. The attempt is to model the complex plane whose axes are the reals extended by +/- inf, which has infinitely many infinities.

A user might prefer a model of the complex plane with only one infinity. The proj function is intended as a tool for the user to create that model by applying proj wherever the different infinities might interfere, e.g. before equality comparison. The complex infinity properties in Annex G are also intended to support such an effort.

- Jim Thomas


> On Oct 2, 2024, at 8:47 AM, Jerome Coonen <jcoonen at gmail.com> wrote:
> 
> We had a task to comment on Damian's first set of proposals re. the language around complex arithmetic. Here is a file (delayed by a ay of muddled distribution attempts) with my comments overlaid on Jim's comments. 
> https://wiki.edg.com/pub/CFP/WebHome/C26-cmplx-mods-aug24-body-67-jwt_jtc.pdf
> 
> Caveat: my browser PDF viewer botches all but the first comment, but the file looks fine if downloaded. Good luck.
> 
> -Jerome Coonen
>  650.996.4738
>  jcoonen at gmail.com <mailto:jcoonen at gmail.com>
> 
> On Mon, Sep 16, 2024 at 9:11 AM Jim Thomas <jaswthomas at sbcglobal.net <mailto:jaswthomas at sbcglobal.net>> wrote:
>> Here’s a CFP wiki link to the file:
>> 
>> https://wiki.edg.com/pub/CFP/WebHome/C26-cmplx-mods-aug24-body-67-jwt.pdf
>> 
>> - Jim Thomas
>> 
>>> On Sep 15, 2024, at 9:54 PM, Jim Thomas <jaswthomas at sbcglobal.net <mailto:jaswthomas at sbcglobal.net>> wrote:
>>> 
>>> Here is a file with inline comments for the first of the four papers (C26-cmplx-mods-aug24-body-67):
>>> 
>>> <C26-cmplx-mods-aug24-body-67-jwt.pdf>
>>> 
>>> - Jim Thomas
>>> 
>>>> On Aug 25, 2024, at 5:08 AM, Damian McGuckin <damianm at esi.com.au <mailto:damianm at esi.com.au>> wrote:
>>>> 
>>>> 
>>>> Hi all,
>>>> 
>>>> I have uploaded my suggestions.  Please provide feedback before the next meeting because there is way too much to cover during a single meeting. At the start, I explain in a block before the actual changes the rationale for some of the proposals. And then for each change, I briefly note why that change has been proposed.  To make it readable, I provide the entire paragraph because most changes are more than one or two words or phrases. I do not use a 'diff' except in one case because it is so trivial.
>>>> 
>>>> I think all are editorial although in one case the word complex is missing so that the existing text says a parameter has a type of 'float' when it meant 'float complex' which seems like extremely editorial.  Most of the changes are for consistency, that is, if words are used one way in one place, and then in related section later, a different wording style is used, I try and use the style which reads the best.
>>>> 
>>>> The domains for all special cases are now inequalities.
>>>> 
>>>> Let me know if you have a problem with these links.
>>>> 
>>>> There are a small number of changes to 6/Language and 7/Library.
>>>> 
>>>> 	https://wiki.edg.com/pub/CFP/WebHome/C26-cmplx-mods-aug24-body-67.pdf
>>>> 
>>>> The major consistency changes in the effective Introductions to Annex F
>>>> and Annex G, i.e. F.10.1 and G.4.1 Introduction a Annex F (some of which
>>>> are just an exercise in reordering), and improved explanations are:
>>>> 
>>>> 	https://wiki.edg.com/pub/CFP/WebHome/C26-cmplx-mods-aug24-annex-fg-intro.pdf
>>>> 
>>>> Note that there is one clause, G.2, that could move to the end of 7.3.1..
>>>> 
>>>> The special cases in Annex F and Annex G now have a consistent syntax.
>>>> I distilled the most consistent existing syntax used for the special
>>>> cases and then applied that rigorously (although I did moved the words
>>>> about exceptions raised for each case to the end of the sentence).
>>>> 
>>>> The changes for consistency in Annex F are in:
>>>> 
>>>> 	https://wiki.edg.com/pub/CFP/WebHome/C26-cmplx-mods-aug24-annex-f.pdf
>>>> 
>>>> The changes for consistency in Annex G are in:
>>>> 
>>>> 	https://wiki.edg.com/pub/CFP/WebHome/C26-cmplx-mods-aug24-annex-g.pdf
>>>> 
>>>> Thanks - Damian
>>>> _______________________________________________
>>>> Cfp-interest mailing list
>>>> Cfp-interest at oakapple.net <mailto:Cfp-interest at oakapple.net>
>>>> http://mailman.oakapple.net/mailman/listinfo/cfp-interest
>>> 
>> 
>> _______________________________________________
>> Cfp-interest mailing list
>> Cfp-interest at oakapple.net <mailto:Cfp-interest at oakapple.net>
>> http://mailman.oakapple.net/mailman/listinfo/cfp-interest
> _______________________________________________
> Cfp-interest mailing list
> Cfp-interest at oakapple.net
> http://mailman.oakapple.net/mailman/listinfo/cfp-interest

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.oakapple.net/pipermail/cfp-interest/attachments/20241002/f82c2e39/attachment-0001.htm>