Compiling large switch statements

[Brian Goetz]

> I suggest using brute classfile size as your metric for choosing your
> switch.  It will never be far off from the model you propose, and the
> differences won't matter once the JIT gets hold of the code.  As a
> bonus, you'll slightly decrease the size of jars, which surely counts
> for something (you don't even have to run the code to collect your
> wins).

Our friends in the ME world would be happy if we did this.