The comment on SQL NaN was based on comments in web discussions on NaN.
Several discussions stated that some SQL systems used NaN as the way to mark
unknown or not applicable data. The comment is not based on experience or
direct knowledge. If no databases use NaN this way, and they can't be a
source for introducing them into J, then great.
----- Original Message -----
From: "bill lam" <[EMAIL PROTECTED]>
To: "Beta forum" <[email protected]>
Sent: Monday, February 25, 2008 11:41 PM
Subject: Re: [Jbeta] Use of the name 'NaN' deprecated
The man page said _. will be created in
DLL call (SQL database, LAPACK, etc.)
Why do you think SQL database will return NaN ?
IIRC NULL is stored and returned in a separated bit/byte array in SQL
database.
Roger Hui wrote:
The change is much more than substituting NaN for _. or vice versa. You
should read the page for _. in the J6.02 qbeta.
Absolutely no plans to change 0%0 giving 0.
----- Original Message -----
From: [EMAIL PROTECTED]
Date: Monday, February 25, 2008 16:25
Subject: Re: [Jbeta] Use of the name 'NaN' deprecated
To: [email protected]
I agree with Henry.
As the Dictionary reads now (and has always read since the 1991
version), the only reference to _. is its "Indeterminate" entry
itself, namely:
Indeterminate _.
The indeterminate _. results from expressions such as _-_ (infinity
minus infinity) and from expressions (such as 3+_.) in which an
indeterminate argument occurs.
A clarification how _. relates to = (and, by induction i. e. ~: >:
<: etc.) is dearly missing in my opinion. (3=_. ? _.=_. ?)
If the clarification should be that the results are "entirely
fortuitous" [Roger's words] or "platform-dependent", that is fine
with me, too. I just think the rules need to be spelled out.
(I also think it should also be made explicit that _. propagates only
in "arithmetic expressions" as opposed to, say, <_. or 2#_. .)
I've spent enough days on making _/__/_. work on various platforms
and am certainly aware of the close ties to IEEE arithmetic. However,
just shelling another moniker such as "NaN" for _. in the Dictionary
doesn't help a bit to define how _. is supposed to work. If anything,
any spurious or explicit reference to NaN or IEEE just raises further
questions just how IEEEish J's arithmetic is supposed to be, without
answering this either. Just stick to the word "indeterminate", as
Henry suggests.
Beyond the references to "NaN", I have another quarrel with the
two new foreigns: I am missing a rank specification for these
verbs.
Martin
PS:
(And the only way to deal with _. in J data is to get rid of it!)
We'd love too, but up to J Rel. 6.01 the Dictionary has been silent
on how to do this.
PPS: Any plans to kill off "0%0 ==> 0"?
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
