Re: [Chicken-users] Chicken benchmarks

[Felix]

Hello, Brian!

Brian Naylor wrote:
> I am trying to make sense of the Chicken benchmark page.  The two entries for
> each program represent the two sets of compiler switches?  So we can consider
> the (a) case to be the slower version for each?

Right.

> Chicken looks like it does pretty well by these measures.  How does it stack up
> against Bigloo?

Not so good, I'm afraid. Bigloo can generate *much* better code,
since it can translate directly from Scheme to C, instead of
going through things like CPS (Chicken) or trampolines (Gambit).

But note that it doesn't really make sense to compare
Bigloo (or Scheme->C or Stalin, for that matter) with Chicken,
Gambit or other compilers that can handle full tail-call
optimization and first class continuations. Providing support for
these constructs means an implementer has to pay a certain price
for *all* code (except when you can generate pure native code,
where you have a little bit more flexibility).

Especially TCO is an issue that has different importance for
different people. So one has to take a decision. Serrano
decided to go for speed.
The same holds for conservative GC: it's fundamentally flawed
but it makes implementation so much easier that it has a
certain appeal.

(Sorry, I'm in rant-mode today. ;-)


cheers,
felix