lef!, One of My Best Macros
While I was looking for any unneeded or outdated packages I could remove from my Emacs configuration I came across noflet, an external package that implements "dynamic, local, advice for Emacs-Lisp code". It is a small package consisting of just one macro: noflet. It is similar in nature to cl-flet with two main differences: it binds functions globally instead of locally and it allows you to refer to the original function with the symbol this-fn.
Adapted from noflet's README, the following example showcases the effect of using noflet. Normally, expand-file-name would interpret "#/thing" as a path in the current directory as is seen from the return value of the raw call to expand-file-name below. However, during the evaluation of its body noflet overrides expand-file-name, replacing it with a function that acts differently when a filename begins with a #. Note, the binding for find-file-no-select actually has no effect because expand-file-name never calls find-file-no-select in its body1--I think it was included for didactic purposes. But if anything did call find-file-no-select it would use it would use the version specified by noflet.
(expand-file-name "#/thing")
;; => "/home/luis/Documents/blog/org/posts/#/thing"
(noflet ((find-file-noselect (file-name)
(if (string-match-p "^#.*" file-name)
(this-fn "/tmp/mytest")
(this-fn file-name)))
(expand-file-name (file-name &optional thing)
(if (string-match-p "^#.*" file-name)
(concat "/tmp" (file-name-as-directory
(substring file-name 1)))
(funcall this-fn file-name thing))))
(expand-file-name "#/thing"))
;; => "/tmp/thing/"
Although undeniably useful, noflet had been causing me some inconveniences. It had been triggering a cl depreciated warning because it required cl instead of cl-lib. Additionally, it uses dash but does not explicitly declare it as a dependency. Not knowing this, I had faced an error when I tried to require noflet before I had required dash. It was apparent From noflet's issues page that I had not been the only one to notice these things (see #26 and #19). From amount of time numerous easy to fix issues have been open it is equally apparent that noflet is unmaintained. To fix these things I had a few options: I could use my own fork of noflet with these fixes applied or I could just create my own macro. Considering how relatively easy it would be to write a replacement for this package--it is just one short macro after-all--I decided to just roll it out myself.
The closest built-in alternative for noflet I could find is cl-letf2. Like noflet, cl-letf can dynamically bind variables.
However, the (symbol-function #'FUNCTION) syntax it uses to dynamically bind
functions is too verbose for my taste. Moreover, it lacks noflet's ability to
refer to the original function using the symbol this-fn. If not vital, the
latter ability to refer to the original function is at least.
In the following example I use lef! to replace the built-in function +, which as
its name implies returns the sum of two or more numbers, with a function that
returns double the sum of exactly two numbers. In the expression (* 2 (funcall
this-fn a b)), this-fn refers the original + function. The built-in + returns
3 when adding 1 and 2 whereas the new + returns 6.
(+ 1 2)
;; => 3
(lef! ((+ (a b) (* 2 (funcall this-fn a b))))
(+ 1 2))
;; => 6
Performing only one step of macroexpansion demonstrates the conversion. The original function bound to + is stored in the generated symbol orig-fn899 and let-bound to this-fn and this-function for the body of the function that replaces +.
(cl-letf* ((orig-fn899 (when (fboundp '+) (symbol-function '+)))
((symbol-function '+) (lambda (a b) (let ((this-fn orig-fn899) (this-function orig-fn899)) (* 2 (funcall this-fn a b))))))
(+ 1 2))
This macro converts the concise and convenient place binding declarations we specify into what we want it to actually do via cl-letf*. For each binding in lef! we create two cl-letf* bindings, one to store the original function (if there is one) and one to globally bind the specified function.
(defmacro lef! (bindings &rest body)
"Bind each symbol in BINDINGS to its corresponding function during BODY.
BINDINGS is a list of either (SYMBOL FUNCTION), where symbol is the symbol to be
bound and FUNCTION is the function to bind it to; or (SYMBOL ARGS BODY). In
each of BINDINGS if the symbol is an existing function symbol let-bind the
original function to `this-fn', otherwise bind `this-fn' to nil."
(declare (indent 1))
(let (binds orig-fn)
(pcase-dolist (`(,sym . ,rest) bindings)
(setq orig-fn (gensym "orig-fn"))
(push `(,orig-fn (when (fboundp ',sym) (symbol-function ',sym))) binds)
(push (list `(symbol-function ',sym)
(pcase rest
(`(,fn . nil)
`(lambda (&rest args)
(let ((this-fn ,orig-fn)
(this-function ,orig-fn))
(apply ,fn args))))
(`(,args . ,function-body)
`(lambda ,args
(let ((this-fn ,orig-fn)
(this-function ,orig-fn))
,@function-body)))))
binds))
`(cl-letf* ,(nreverse binds) ,@body)))
Overall, I am pleased with lef. I am happy I was able to reuse the built-in cl-letf instead of implementing dynamic binding myself. Additionally, I am proud I was able to implement the this-fn functionality. Yet I still have some lingering thoughts. Perhaps one critique of lef! is that the symbols this-fn and this-function are bound regardless of whether you invoke the original function anaphrocally. If you do not reference the original function, after all, you would be able to expect these symbols are unbound. Although, I am hard-pressed to think of a realistic example of how this could cause problems I cannot help but think that to be "perfect" this excessive binding should be avoided--good topic for another blog post.