I'll take a little space at the top of this message to pop out a level
and discuss what seems to be an underlying issue here. It is true that
many of the things in Scheme can be expressed in a language that just
has lambda, variables, and applications, but I do not think that means
such things should be dropped from the semantics.
While did think carefully about what would be redundant when writing
the semantics (and did avoid many redundant things), the decision of
whether or not something went into the semantics was based on how
useful it would be for people to use the semantics to understand how
their programs (or some kernel of their programs) behave, presumably
exploiting PLT Redex's stepper to do so.
As an anecdote, I recently did that with some code from the 80s (from
the Constraining Control paper by Friendman/Haynes). That paper
contains code that nearly runs in the semantics. There were a few
minor things that I had to write out, namely "and" and "case" and a
few things I had to define in the semantics "not" and "for-each" and
this wasn't a big deal.
But, I had to rewrite out "let" expressions. This turned out to be a
big deal. It is much harder to follow along with the reductions and
understand what's happening when your let expressions are "inside out"
as lambda applications.
Indeed, I've been thinking about adding let to the semantics, just to
make such exploration easier.
So, that is the spirit in which I think that the library top-level
should remain. It was easy for me to stick a few definitions into my
program (in the above anecdote) without having to inline the
definitions into the uses (and possibly have to rewrite more,
depending on how they would be used).
Of course, there are other arguments to include the top-level, if the
semantics become what Kent asked for in his formal comment (since it
will have its own unique subtleties that will probably be of
interest).
On 3/15/07, AndrevanTonder <andre_at_het.brown.edu> wrote:
> On Thu, 15 Mar 2007, Robby Findler wrote:
>
> > lambda cannot express letrec.
>
> Does the current semantics of DEFINE express LETREC*?
Not as written, no (since the defines have to be at the top-level).
> Regarding expressing LETREC* in terms of LAMBDA, giving some interpretation to
> <undefined> would seem to be a low hurdle to clear to make this possible.
You also need assignment (unless you allow yourself to do non-local
rewrites, of course).
Robby
Received on Thu Mar 15 2007 - 12:26:13 UTC