I had a couple of conversations with Alan Bawden. I lobbied
(unsuccessfully) to get a macro writing license from him. I didn't
advance my case by mentioning that I had written a Scheme interpreter
using syntax-rules. Alan is not a fan of syntax-rules.
I agree with Alan that implementing any sort of complex macro with
syntax-rules is like forcing a program through a straw, but the
alternatives are unpleasant, too. I like the fact that syntax-rules
can be expanded without the need to invoke the interpreter, so phase
mixing is impossible.
After Olin Shivers's talk about his loop macro, it occurred to me that
Alan's observation about first-class macros having types
(Clams
got legs!) might be relevant. When a macro gets somewhat complex,
it has to code walk its arguments. It needs to be able to distinguish
between chunks of code and binding constructs. When a macro gets very
complex (like Olin's loop macro), it becomes impossible to code walk.
(Of course you can expand the inner macro away, but there are
advantages to keeping it in its original form.) This makes it
difficult to use moderately complex macros. I was thinking that
Alan's ideas of marking the types of a macro could help during the
code walking.
No comments:
Post a Comment