Comments on: Clojure: The 'apply' function

By: Rafał Dowgird

Rafał Dowgird — Thu, 04 Mar 2010 12:07:25 +0000

@Thomas: "I always thought apply was the same as (eval (cons f args))"

An important implication of the above article is that apply is lazy with respect to args. (apply (constantly 7) (repeat 1)) evaluates to 7 even though the args list is infinite. The eval-cons cannot do that.

By: Brian McKeough

Brian McKeough — Thu, 26 Nov 2009 02:35:47 +0000

Thanks, you inspired me to look at the innards of Clojure in a way I hadn't before. It appears to me that str is itself compiled into an IFun instance (actually two, one extending AFunction and one extending RestFn). Look for (and decompile) core$str* in the clojure.jar. Each of these implement "invoke" according to the logic of str. At this point I frankly don't understand why there are the two different classes. As you and Ola already indicated "invoke" is called by applyTo. I believe AFn (AFunction's parent class) implements applyTo as a call to this.invoke with the appropriate parameter count.

By: Thomas Winant

Thomas Winant — Wed, 25 Nov 2009 14:12:19 +0000

I always thought apply was the same as (eval (cons f args))

By: Mark Needham

Mark Needham — Wed, 25 Nov 2009 05:50:04 +0000

Yeah that's right as far as I can see but then inside the 'applyTo' method there seems to be a call to 'fn.invoke()' which I'm assuming calls the Clojure function?

By: Ola Bini

Ola Bini — Wed, 25 Nov 2009 05:27:25 +0000

Actually, the magic in the apply method is [#^clojure.lang.IFn f & args] which (I'm pretty sure) is a type assertion. The . is a Java method call, and applyTo is a method defined on IFn, which is the common interface for all functions in Clojure – so that means Java integration is used to do str.applyTo(seq([1 2 3]))