Dale Stanbrough wrote:

> Here's my prediction...
>
>         the round off requirements will be changed to meet the
>         C/C++/Java implementations.
>
> Given the proliferation of business/internet software that don't even
> realise that there is a problem, legislators will be faced with a
> fait-accompli, and will accept the status quo.

Try telling that to a mutual fund manager who lives or dies on based
on a few basis points (a basis point is 1/100th of a percent)! It is
okay if you fudge the numbers and make the manager look good, but woe
to you if the numbers make the manager look bad.

This round-off problem becomes real noticable if an analyst runs through
the same set of computations with a desk calculator and gets a different
answer than the program. Business desk calculators typically support 18
decimal digits and perform all math in decimal. Analysts may accept
slightly inaccurate numbers for non-critical uses such as customer
presentations, audits and accountants want dead on accurracy. Hence the
large number of accounting applications written in COBOL or PL/1 still
running on mainframes. I do not see this situation changing any time
soon.

This issue is a strong selling point for Ada and one I raise every
chance I get. Is the accuracy of the result just another facet of
reliability? Is not reliability the big advantage of Ada over C/C++?

David Koogler
Boolean Solutions, Ltd.