Arc Forumnew | comments | leaders | submitlogin
1 point by binx 6105 days ago | link | parent

A proper module system which works well with non-hygienic macros seems to be difficult to make. I don't know whether it would be better if macros are made first-class.


4 points by almkglor 6105 days ago | link

I disagree. Recall that macros in Arc have precedence in expressions over non-macros; that is, if obj in (obj x) is globally a macro, even if it is used as a variable name, all (obj x) will be replaced by the macroexpansion. Note that this applies for all x.

Suppose we have nested macro expressions. The inner macro will not see the expression until the outer macro processes that value; the outer macro might even decide to completely bollix the inner macroexpression.

   (macro ....
     (sub-macro ...))
Now suppose we declare a "using" macro, which replaces all symbols inside its expressions for other symbols, i.e.

  arc> (using
      (
         foo   bar
         nitz  koo)
      '(foo bar nitz koo))
  (bar bar koo koo)
Obviously, if we postulate another macro 'foo, the replacement done by using will prevent the macro 'foo from transforming any part of the (using ...) form.

The benefit of this using macro is as a basis for module systems. Suppose we have a module system such that:

  arc> (macex
    '(module foo
        (def foo () 0)
        (mac bar () nil)))
  (do
     (= (*members* 'foo) (list 'foo 'bar))
     (def foo--foo () 0)
     (mac bar--foo () nil))
  arc> (module foo
        (def foo () 0)
        (mac bar () nil)))
  <#procedure:foo--bar 3 mac>
  arc> (macex '(w/module foo (bar)))
  (do (foo--bar))
(Damn. I've been reading too many EWD's, my language is starting to sound academic.)

-----

1 point by binx 6105 days ago | link

Rewriting all symbols is what the CL package system does. Of course, the CL package system does not really rewrite symbols after parsing, it chooses different symbols for the same string in different packages when parsing.

A true module system just rewrites all identifiers that are globally referenced, without touching local identifiers and quoted symbols.

-----

1 point by almkglor 6104 days ago | link

Why not local identifiers? It won't hurt anyway, since:

  (fn (x) x) == (fn (y) y)
Quoted symbols are more problematic. The problem is that you can't determine how macros will bash quoted symbols. For example, (tag (x) ...) implicitly creates 'x. And prior to replacement, you can't be sure if the 'tag here is pg:arc1's tag or some other module's 'tag macro.

An alternative is to make the (pr ...) versions of similarly-named symbols in different modules the same; this probably involves hacking (coerce ... 'string) to cut out the module name for symbols.

-----

1 point by binx 6104 days ago | link

If a module doesn't export any macros, the problem would go away. Just macroexpand all function definitions, then all quoted symbols can be detected.

Hygienic macros can also be dealt properly, mzscheme is an example. But I have no idea how Arc macros can work well with modules.

Maybe the CL package system is the only way.

-----

1 point by almkglor 6104 days ago | link

Hmm, probably means we need a separate macro namespace. Heck, stuff like (let obj (table) (obj 1)) doesn't do as you expect, because of the mixing of macros in the same namespace as functions and values.

-----

3 points by bramsundar 6105 days ago | link

Aren't macros already first-class?

-----