"Darren New" <dne### [at] sanrrcom> wrote

> Humans are notoriously bad at guessing where the time in their programs
> go, and notoriously bad at keeping track of things like whether the
> value in R147 is going to be needed before the value in R93 is.
>
> And if a compiler already generates optimum code, you obviously can't
> improve on it.

Since *humans* write compilers, your argument is self defeating.

I know what you are *trying* to say. However, no matter how efficient, high
level to low level is a translation layer, and if the application writer is
as proficient as the compiler writer, he can write at the very least the
same quality, often better, direct low level code than that can be achieved
by writing high level code and having it translated by someone else's
compiler, linked to someone else's APIs and libraries. However, in practice,
it's an economical decision, and that's why there are many, many layers of
abstraction between the machine and the application programmer. The
compromise is that modern applications may be a couple of orders of
magnitude more bloated, in both footprint and performance, than that could
be achived with careful low level coding. "Hello World" windows application
in Delphi, for instance, takes several hundered kilobytes. Don't tell me
that an assembler version can not be more optimized.

Post a reply to this message