precision

[John P. Morgan]

I agree with all Peter's suggestions below. JP

>
>
>So how about adding the "precision" property to <statistical_properties>
>with the semantics, outlined above, that any decimal number inside its
>scope has this uniform precision.  In particular, extended functions
>inside <statistical_properties> are then understood to have this
>precision.
>
>In general, the use of precision in extended functions would be modified
>as it follows:
>
>    - precision is mandatory if the extended functions contains decimal
>      numbers
>
>    unless
>
>    - the ext-funcs are inside the scope of a subtree whose root
>      already specified a precision attributes, in which case the
>      precision is assumed uniformly within the subtree.

>
>At the moment we have decimal numbers only inside
><statistical_properties> but the above principle using subtree roots
>could be easily applied should the need arise later.
>
>I suggest a precision policy using at least precision 9 since 8 decimal
>digits cannot represent all "single precision" (IEEE standard) floating
>point numbers and nowadays even "double precision" is common.
>
>--             ,
>    Peter Dobcsanyi