RE: Rounding errors in financial app transforms

> MulDiv, as, for example the Windows API function,
> 
> "The MulDiv function multiplies two 32-bit values and then 
> divides the 64-bit result by a third 32-bit value.
> The return value is rounded up or down to the nearest integer."
> 

The specs do say that implementations must support 18 digits of precision
for xs:integer, so it's very likely they will use 64 bits or more. There's
no absolute rule that intermediate results of xs:integer calculations must
retain at least 64 bits, but it's a reasonable expectation. xs:decimal
leaves more scope for implementations to vary, especially with division and
in their overflow behaviour, but I think you'd be unlikely to find an
implementation that doesn't allow any two 32-bit values to be multiplied
without loss. 

Michael Kay
http://www.saxonica.com/