RE: Representation of floating point values.

>>Oh, well.  I guess those who care about large numbers will just find a
>>way to bypass this schema limitation.

>Yes.  The decimal datatype allows large numbers, and the proposed
>pDecimal will allow both large numbers and scientific notation.


The issue that I was referring to is that the current mapping loses
information, not that a large number can not be represented.

A value outside of the valid range is almost always an error or a
special case and an application may need to know about it.  Explicit
Infinity value may be a valid value, while a too large literal may be a
typo.

For example literals in XQuery are typed in by a programmer and silent
mapping of them into an Infinity will conceal an error.

Or some schema may use double for efficiency, and roll back to pDecimal
when too large numbers are encountered (but still support Infinity as a
valid value). All those options for handling overflow/underflow in IEEE
754 are probably there for a reason.


Daniel;

Received on Wednesday, 9 March 2005 17:44:29 UTC