Ah, this has been repeated many times, but I can't help to click "reply"...
> It's easy enough to achieve with a defmacro-style macro system - just use gensyms.
Gensym cannot prevent global identifiers from being shadowed in macro use environment.
(define-macro (my-if a b c) `(if ,a ,b ,c))
(define (boo)
(let ((if list)) ; if is shadowed
(my-if 'x 'y 'z))) ; evaluates to (x y z), not y
In CL you don't need to worry much, since (a) it is Lisp-2 so variables won't shadow operators, and (b) using packages greatly reduces accidental shadowing. In Scheme the chance is much higher; you probably don't bind 'if' locally, but you can't check every user-defined global procedures the macro inserts into the output.
It is unfortunate for abstractbill to use this incorrect explanation, because hygienic macros are still a red herring - a decent module system and a bit of restraint gets you there.
(define-macro (my-if a b c) `(std.if ,a ,b ,c))
(define (boo)
(let ((if list)) ; if is shadowed
(my-if 'x 'y 'z))) ; gives (std.if x y z)
(define (boo2)
(let ((std.if list)) ; error! cannot shadow module vars.
(my-if 'x 'y 'z)))
I've posted this around a few times asking why this isn't already used in Lisps and never seem to get a convincing argument back. The only conclusion I can draw from this is that CL-ers are ashamed of the fact that multiple namespaces aren't actually necessary and Schemers are rightly ashamed of hygienic macros. Please convincingly prove me wrong!
(ps. This ofcourse would not be limited to the languages std module but would also apply to any exported module symbols.)
That's exactly what I meant in "In CL ... (b) using packages greatly reduces accidental shadowing." So I'm not sure what you are asking to be convinced. CL certainly does not prevent you from messing with package-private symbols, but usually a convention is enough to avoid disaster. I mean, I won't write (let ((your-package::private-symbol list)) (your-macro ...) ...) unless there's absolute need to do, such as emergency workaround of a bug in a third-party library.
On the other hand, one of the reasons Scheme avoids explicit module prefix is, I think, that it breaks abstraction. Suppose my-module defines my-macro and my-function. My-macro expands into a call to my-function. But I don't really want my module users to call my-function directly, so I only export my-macro and keep my-function private.
In your proposal, my-macro will expands into prefixed-symbol like this:
But... if we allow module-private function to be called with module prefix, we can't prevent ordinary code from calling my-function as well.
This is a language design choice. Some may not mind others to mess with module-private bindings; CLers certainly would not. But if you want to hide module-private symbols, and want to allow macros expand into them, how will you do that? One idea is to mark the output of macro expansion so that the compiler can distinguish whether the reference is from legitimate macro expansion or not... well, that's what hygienic macro is doing.
Thanks for the explanation. I hadn't thought of the my-macro problem because I tend to think of macros exactly as a clever copy paste mechanism, so would expect it to always use public variables, but I guess some people might have a problem with that in some cases.
In the first point I was asking to be convinced that you can't have hygiene solely through the use of a package / module system and without needing multiple namespaces.
Although your workaround argument (for allowing variables to be shadowed) is convincing, in that case the package system wouldn't really be providing any safety at all and only the convention would be helping. I guess we can sum it up as CLers really don't see a problem at all, and Schemers require macros to be completely abstract.
(mac awhen (expr . body)
`(std.let it ,expr (std.if it (std.do ,@body))))
This assumes that awhen is being defined and exported from the std module too, so that it can also be used in expansions. Slightly more verbose, but that's optional hygiene if you want it.
Comments
Ah, this has been repeated many times, but I can't help to click "reply"...
> It's easy enough to achieve with a defmacro-style macro system - just use gensyms.
Gensym cannot prevent global identifiers from being shadowed in macro use environment.
In CL you don't need to worry much, since (a) it is Lisp-2 so variables won't shadow operators, and (b) using packages greatly reduces accidental shadowing. In Scheme the chance is much higher; you probably don't bind 'if' locally, but you can't check every user-defined global procedures the macro inserts into the output.It is unfortunate for abstractbill to use this incorrect explanation, because hygienic macros are still a red herring - a decent module system and a bit of restraint gets you there.
I've posted this around a few times asking why this isn't already used in Lisps and never seem to get a convincing argument back. The only conclusion I can draw from this is that CL-ers are ashamed of the fact that multiple namespaces aren't actually necessary and Schemers are rightly ashamed of hygienic macros. Please convincingly prove me wrong!(ps. This ofcourse would not be limited to the languages std module but would also apply to any exported module symbols.)
That's exactly what I meant in "In CL ... (b) using packages greatly reduces accidental shadowing." So I'm not sure what you are asking to be convinced. CL certainly does not prevent you from messing with package-private symbols, but usually a convention is enough to avoid disaster. I mean, I won't write (let ((your-package::private-symbol list)) (your-macro ...) ...) unless there's absolute need to do, such as emergency workaround of a bug in a third-party library.
On the other hand, one of the reasons Scheme avoids explicit module prefix is, I think, that it breaks abstraction. Suppose my-module defines my-macro and my-function. My-macro expands into a call to my-function. But I don't really want my module users to call my-function directly, so I only export my-macro and keep my-function private.
In your proposal, my-macro will expands into prefixed-symbol like this:
But... if we allow module-private function to be called with module prefix, we can't prevent ordinary code from calling my-function as well.This is a language design choice. Some may not mind others to mess with module-private bindings; CLers certainly would not. But if you want to hide module-private symbols, and want to allow macros expand into them, how will you do that? One idea is to mark the output of macro expansion so that the compiler can distinguish whether the reference is from legitimate macro expansion or not... well, that's what hygienic macro is doing.
Thanks for the explanation. I hadn't thought of the my-macro problem because I tend to think of macros exactly as a clever copy paste mechanism, so would expect it to always use public variables, but I guess some people might have a problem with that in some cases.
In the first point I was asking to be convinced that you can't have hygiene solely through the use of a package / module system and without needing multiple namespaces.
Although your workaround argument (for allowing variables to be shadowed) is convincing, in that case the package system wouldn't really be providing any safety at all and only the convention would be helping. I guess we can sum it up as CLers really don't see a problem at all, and Schemers require macros to be completely abstract.
Clojure's backquote syntax automatically resolves symbols into namespaces. `(if ~a ~b ~c) gives ('clojure.core/if a b c)
amazing language. hope it takes off.
To make it clear, the example below would become:
This assumes that awhen is being defined and exported from the std module too, so that it can also be used in expansions. Slightly more verbose, but that's optional hygiene if you want it.