File: cconv.el.html

This takes a piece of Elisp code, and eliminates all free variables from lambda expressions. The user entry point is cconv-closure-convert. All macros should be expanded beforehand.

Here is a brief explanation how this code works. Firstly, we analyze the tree by calling cconv-analyze-form. This function finds all mutated variables, all functions that are suitable for lambda lifting and all variables captured by closure. It passes the tree once, returning a list of three lists.

Then we calculate the intersection of the first and third lists returned by cconv-analyze-form to find all mutated variables that are captured by closure.

Armed with this data, we call cconv-convert, that rewrites the tree recursively, lifting lambdas where possible, building closures where it is needed and eliminating mutable variables used in closure.

We do following replacements :
(lambda (v1 ...) ... fv1 fv2 ...) => (lambda (v1 ... fv1 fv2 ) ... fv1 fv2 .)
if the function is suitable for lambda lifting (if all calls are known)

(lambda (v0 ...) ... fv0 .. fv1 ...) =>
(internal-make-closure (v0 ...) (fv0 ...) <doc>
  ... (internal-get-closed-var 0) ... (internal-get-closed-var 1) ...)

If the function has no free variables, we don't do anything.

If a variable is mutated (updated by setq), and it is used in a closure we wrap its definition with list: (list val) and we also replace var => (car-safe var) wherever this variable is used, and also
(setq var value) => (setcar var value) where it is updated.

If defun argument is closure mutable, we letbind it and wrap it's definition with list.
(defun foo (... mutable-arg ...) ...) =>
(defun foo (... m-arg ...) (let ((m-arg (list m-arg))) ...))

Defined variables (2)

Defined functions (21)

Defined faces (0)