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

Re: Motion P1788/M007.01_NaI: NO, several NaIs please



Dan & all
<< George, this counts as a vote, please. >>

Sorry not to reply before start of voting period.

On 24 Aug 2009, at 20:04, Dan Zuras Intervals wrote:
	P.S. - On the merits:  I would argue in favor of no NaI or
	just one.  The argument about the historical uselessness
	of the NaN payload is a compelling one in my opinion.

Seems to me its "historical uselessness" (nice phrase) has much to do with no _official_ payload values and meanings having been defined.

If, say, (following IBM's cunning scheme Michel Hack described on Aug 6th) 754-1985 had stated that payloads either shall be, or are recommended to be, something like
  Payload after bit reversal            Meaning
     integer 0                          result of invalid operation
     integer 1                          missing data
     ...                                ...
then you can betcha Matlab, for instance, would by now be using (NaN with payload 1) for its missing data feature in the Matlab plotting interface. And programmers would know that, and exploit it.

The many-Nans or many-NaIs idea may be compared with the ASCII control-characters -- character codes 0 to 31.
  - Some of them, such as ESC and CTRL-C and BEL, have kept
    close to their originally intended meaning across many
    years and in many application programs.
  - CR and LF, which originally described two different
    mechanical motions on a printer or typewriter, now
    mostly mean the same thing but on different systems
    (causing us no end of trouble converting between Unix,
    Apple & Windows).
  - Some, such as Vertical Tab VT and Form Feed FF, have
    little use in modern technology.

So, some lasted and others didn't. But it's a good thing the ASCII group _did_ define these meanings, instead of saying "you can use characters 0 to 31 for any non-printing meaning you feel like".

Let's do the same and define some NaIs with meanings that seem useful to us now. Like CTRL characters, some meanings will stand the test of time and others won't. When we reconvene in years to come as P1788-R, we shall revisit them.

I take a different message from Dan, about the "historical uselessness" of NaN payloads, and it's a message Dan feels strongly about:
                   DEFINE, DEFINE, DEFINE.

So I cast my vote for "NO, I want several NaIs", and I want us to specify them.

Regards

John