Re: Late vs. early binding Was: DISFAVORED Was: name for 3 PICK

Jonah Thomas <jethomas5@[EMAIL PROTECTED]
> wrote:
>Josh Grams <josh@[EMAIL PROTECTED]
> wrote:
>
>>How would I put an execute-only word in a macro without using
>>POSTPONE (or EVALUATE)?
>
>What I'm imagining is that the macro would use EVALUATE . So you
>never get an xt for the execute-only word. You can't use COMPILE,
>on it, or POSTPONE .
>
>The result is a simpler system because all the things you can't
>do don't need methods to do them. And it's easier to understand
>because a collection of methods that are kind of complex to
>follow are gone.
>
>But the missing functionality might actually get in the way. I'm
>trying to see if there are some Gotchas there and not just Yes
>Buts.

I think we'll have to agree to disagree here.  But just for kicks,
here's my point of view.

The main gotcha that I know of is that someone could redefine one
of the words you're using in your macros.  I've had that happen to
me, and it's ugly.  One of the things I like most about Forth is
the early binding, and I do reuse short common words as internal
names.  And it just feels wrong to me to have the macro system be
late binding when the rest of the system is explicitly early
binding.

You can sort of get around the late-binding problems with
EVALUATEd macros by aliasing the words you use into a custom
wordlist, and then setting the search-order before you EVALUATE,
but...then you need a standard way to create an alias to an
IMMEDIATE or execute-only word.

I'm not saying that I'm entirely happy with the current state of
affairs.  But I do think that lisp/forth-style macros (allowing
user code to execute at parse-time) is an intrinsically better
approach.  I think we would benefit from a standard set of words
for manipulating code as a list of words, but manipulating
code-as-data rather than manipulating code-as-text just seems like
a good idea to me.

I also suspect that it would be possible to make the current
execute/compile/postpone conceptually cleaner, especially if you
drop sup****t for combined words.  It does bug me a little that
parsing words don't work the same in compilation state.  But it
may not be possible to do something about that in a low-level
language like Forth.  One of these years I'd like to re-read
Matthew Flatt's paper on PLT Scheme's macros and see if I can come
up with a way to apply some of that reasoning to Forth.

--Josh