Proposals

[Jason Marshall]

>You are, however, bypassing the importance of Rice's Theorem, and I 
don't
>understand why you feel justified in that.  If it's impossible to 
decide
>which program is faster, how can you be justified in claiming that you 
can
>choose the fastest?  I myself would choose instead the smallest, not 
the
>fastest; I wouldn't always be right, but at least I would often be, 
and it's
>a decision which is easy to make.

I suspect it may be significant that all of the examples Lynn has 
offered involved code that did, provably, complete.  But optimising 
simple equations to within a hair's breadth of perfect leaves you with 
vastly inefficient control flow structures.

But the thing is, most compilers are already doing a reasonable 
approximation of this, are they not?

-Jason