Re: [Tcllib-devel] Tcllib Release 1.8 Status and Action Plan

[Gerald W. L. <Ger...@co...>]

Arjen Markus wrote:

>Hello all,
>
>I was just checking the various bug reports that are assigned to me
>and I came across one (bug 1151935) that made me think:
>
>In general what should happen if a mathematical function is to be
>evaluated for a set of argument for which it is not defined?
>
>In this case, the bug is that there is a division by zero and 
>this is reported as such, rather than as a clearer message
>that the coefficient of variation is not defined if the mean
>of the data is zero.
>
>My question:
>- Should the function return such a (clearer) message or
>- Should it return an empty string (not an error) thereby
>  leaving it to the caller to handle the _undefined_ 
>  situation
>
>I can argue either way, but we should probably try for a
>consistent practice.
>  
>
IMHO, I would emulate what sqrt() does in expr, namely:

    % expr {sqrt(-1.0)}
    domain error: argument not in valid range
    % set errorCode
    ARITH DOMAIN {domain error: argument not in valid range}
    % set errorInfo
    domain error: argument not in valid range
        while executing
    "expr {sqrt(-1.0)}"

In other words, raise an error -- but be sure to set error code to some 
appropriate list of informational elements.

Sincerly,
Gerald W. Lester