🚀 go-pugleaf

In article <39g5bg$3tc@csnews.cs.Colorado.EDU>,
Tom Christiansen  <tchrist@mox.perl.com> wrote:

>Since no such beast exist, evidence must at best be circumstantial and
>inferred.  But similar cases (where you translate a program language X to
>language Y) nearly always should worse performance than those which start
>out in language Y.

I won't dispute that, but...

that wasn't the point you made; you said that a program written in language 
X will run faster than the same program translated from X into language Y.  
e.g. that translating a program written in python, perl, or ksh into Scheme 
would slow it down.  That depends heavily on the comparative quality of
the implementations of X and Y.

I think it's not unlikely that code written by an X->Y translator will
perform less well than code handwritten in language Y, but if Y is a much
more efficient implementation of language Y than X is of language X, there
is still quite a bit of room for speed gains.

>Counterexamples, albeit not non-existent, are at best
>scarce.

I've used several of those counterexamples.  GCL, a lisp to C compiler, 
outruns CLISP's byte-code interpreter.  ECoLisp, another lisp to C compiler,
outperforms CLISP by a factor of 6, and CLiCC outperforms CLISP by a factor
of 3-4.  Camloo compiles Caml Light into Scheme, which is then compiled
by bigloo into C, and outruns the Caml Light byte-code interpreter by
a large margin.

  -- Robert