Re: [office-formula] Groups - OpenFormula Specification2007-06-20 (ODT) (openformula-20070620.od

["David A. Wheeler" <dwheeler@dwheeler.com>]

I said:
> > Functions and operators that receive one or more error values as an input
> > shall produce one of those error values as a result, unless the formula or
> > operator is specifically defined to do otherwise."

Andreas J Guelzow :
> Are we sure about this. I can't get to a current version of the
> openformula specs in the moment (slow dialup) but this is not true with
> respect to the latest version I have handy. For example COUNTBLANK and
> ISBLANK typically do not propagate errors but the description of those
> functions does not specifically address this issue.

I think that's ambiguity in the text above - thanks.

What I _meant_ was, COUNTBLANK(1/0) will respond with an error.
COUNTBLANK(Range) depends on what happens when errors are in the range.

> In any case I am not sure that requiring "one of those error values" as
> a result is a good idea. An implementation that has a large number of
> possible error values may want to return a less specific error value if
> many different error values come together in a formula.

Hm.  Maybe we should back off and just say "returns an Error"?

Any suggestions on how to reword this?

> PS: I guess we should stay away from the questions whether formulas such
> as A1*A2 and A2*A1 should always return the same value.

Yes please :-).  Almost everyone implements "normal" numbers by calling down to the CPU's numerical processor, and many don't fully comply with IEEE (etc.) standards.  We can't solve everything.

--- David A. Wheeler