I’ll assume most of you will just roll your eyes at this title. After all, links are an essential part of the World Wide Web. Not even Emacs is safe from underlined text taking you elsewhere when clicking it. Except that sometimes an ordinary click won’t do and will just position point at the clicked position whereas a middle-click activates the link. To understand what exactly is going wrong there, I’ll go a bit more into depth than usual.
First of all, Emacs doesn’t call clickable text links. This is to be expected as it predates a hefty amount of technology we’re used to, but can throw off beginners who don’t know they’ll want to use the button package. The only problem being there that clickable text produced this way will not obey the convention of an ordinary click activating the link…
I’ve accidentally found the solution to this problem when staring at the code of info.el, wondering how it did achieve this feat of making its references behave like the links I was used to:
(define-key map [follow-link] 'mouse-face)
It isn’t unheard of for Emacs to know unusual symbols representing keys on exotic keyboards, but follow-link is new to me. Even more so is putting a face into the definition part. The manual doesn’t know of this being a valid value for keybindings either. Unbinding that definition did make normal clicks no longer work, so I did search the sources to find the following:
(defun mouse-on-link-p (pos) "..." (let ((action (and (or (not (consp pos)) mouse-1-click-in-non-selected-windows (eq (selected-window) (posn-window pos))) (or (mouse-posn-property pos 'follow-link) (let ((area (posn-area pos))) (when area (key-binding (vector area 'follow-link) nil t pos))) (key-binding [follow-link] nil t pos))))) (cond ((eq action 'mouse-face) (and (mouse-posn-property pos 'mouse-face) t)) ((functionp action) ;; FIXME: This seems questionable if the click is not in a buffer. ;; Should we instead decide that `action' takes a `posn'? (if (consp pos) (with-current-buffer (window-buffer (posn-window pos)) (funcall action (posn-point pos))) (funcall action pos))) (t action))))
Looks like this predicate checks whether there’s a follow-link property or key binding at the position, then explicitly compares its value with mouse-face. Mystery solved. As for why they did it that way, my hunch is on the mouse handling code being an incredible mess and adding extra meaning to keymaps did allow them to easily define alternative clicking behaviour for an entire mode. As usual, the manual explains the finer details.