Since it's only 30 lines of code, here's the relevant section from SBCL (as found in src/code/early-setf.lisp):
(defmacro-mundanely setf (&rest args &environment env)
#!+sb-doc
"Takes pairs of arguments like SETQ. The first is a place and the second
is the value that is supposed to go into that place. Returns the last
value. The place argument may be any of the access forms for which SETF
knows a corresponding setting form."
(let ((nargs (length args)))
(cond
((= nargs 2)
(let ((place (first args))
(value-form (second args)))
(if (atom place)
`(setq ,place ,value-form)
(multiple-value-bind (dummies vals newval setter getter)
(sb!xc:get-setf-expansion place env)
(declare (ignore getter))
(let ((inverse (info :setf :inverse (car place))))
(if (and inverse (eq inverse (car setter)))
`(,inverse ,@(cdr place) ,value-form)
`(let* (,@(mapcar #'list dummies vals))
(multiple-value-bind ,newval ,value-form
,setter))))))))
((oddp nargs)
(error "odd number of args to SETF"))
(t
(do ((a args (cddr a))
(reversed-setfs nil))
((null a)
`(progn ,@(nreverse reversed-setfs)))
(push (list 'setf (car a) (cadr a)) reversed-setfs))))))
Okay, so it's not the
simplest thing around, and it hinges upon a bunch of other things also defined in the source. Nonetheless it's not very "oatmeal and fingernail clippings" after all. It's pretty straightforward. "Here is my docstring, let `nargs` be the number of arguments, if this is 2, then we've either got the primitive case (setf a 1) which is actually a `setq`, or else we macro-expand the place which should be a list like `(car a)`, pull off that first symbol `(car place)`, get its function -- call that function `inverse`, then return an expression which will call `inverse` on the `(cdr place)` with the `value-form` we're trying to set to. If this inverse doesn't exist then there seems to be some transformation of the list (a comment would have been nice here, because I'm not sure what this describes). If the number of args is odd then that's an error; if it's an even number greater than two then we assume that the syntax was (setf a 1 b 2) and recurse on the child pairs."
As readability goes, it's not bad.