[r6rs-discuss] [ANN] scheme-reports.org

[Thomas Lord]

On Mon, 2009-08-24 at 19:37 -0400, Lynn Winebarger wrote:
> On Mon, Aug 24, 2009 at 4:33 PM, Thomas Lord<lord at emf.net> wrote:
> > Scheme is interesting for more than just its math properties,
> > though.  It's also interesting because of how its math
> > properties harmonize so nicely with various implementation
> > techniques.  In any real implementation - one that minimally
> > is capable of some degree of hosting a meta-circular interpreter,
> > say - we need primitive types for symbols, numbers, characters,
> > strings, pairs, and vectors.  How else can you explain
> > tag bits to the kids?  :-)
> >
> 
> Does not a meta-circular compiler deserve as much attention
> as a meta-circular interpreter?  When you evaluate it, will its
> meaning not be as ungrounded?

I am unclear what the referent of "it" is in
your second question and unclear as to what
you mean by "ungrounded".  So, that whistling
noise I hear could well be your point sailing
cleanly over my head.  Whoosh.  Look at it go.

That said, making a guess about your meaning:

I think interpreters are primary and compilers
secondary.  I think Scheme "wants to be" an 
interpreted language with first class environments
and no chance at all of optimizing compilation 
in the general case.  I think Scheme "wants to have"
various subsets (in combination with libraries) that
an optimizing compiler can go to town on.

The default compilation output for arbitrary form
X should be something like (eval 'X appropriate-environment)
but if static analysis of X indicates that optimization
is possible, then by all means, optimize.

As for the grounding of meaning, I thought that was
what least fixed points were for.

Informally: it's a dynamic language all the way
down so it should be a dynamic language all the way
up.  The "compiler bias" of the past couple of 
Rn's never sat well with me.   I thought that when
R5 and R6 happened we ought to have been adding 
first class environments and separately writing 
about compilable subsets.

-t