outline-minor-mode doesn't behave correctly with clojure-mode

[j-cr]

TL;DR: in clojure-mode-variables, the line
(setq-local outline-regexp ";;;\\(;* [^ \t\n]\\)\\|(")
is likely incorrect; replacing it with (setq-local outline-regexp ";;;;* ") works for me.

Expected behavior

Do M-x outline-minor-mode in a clojure buffer; the command outline-insert-heading should behave correctly.

In particular, call to outline-insert-heading on the first line of each pair should result in the second line:

;;; heading:
;;; 

;;;; h2:
;;;; 

(foo)
;;; 

Actual behavior

;;; heading:
;;; h 

;;;; h2:
;;;; h 

(foo)
( 

Steps to reproduce the problem

See above, tldr; M-x outline-minor-mode, then M-x outline-insert-heading.

Environment & Version information

clojure-mode version information

clojure-mode (version 5.11.0)

Emacs version

25.3.1

[bbatsov]

Seems something's wrong with the current regexp indeed, although it doesn't seem obviously incorrect. The intention seems to be to check for an extra ; or some whitespace after ;;; which makes sense to me. \\|( looks a bit weird and I'm not sure why this was added.

[bbatsov]

Btw, it seems we copied this from lisp-mode, as there we have:

  (setq-local outline-regexp ";;;\\(;* [^ \t\n]\\|###autoload\\)\\|(")

[j-cr]

Re \\|(: if I understand correctly, the intent was to treat top-level expressions as headings, so e.g. the visibility of defuns could be cycled as if the were headings.

But there are packages (bicycle) that solve this problem by treating top-level exprs explicitly (bicycle knows how to hide them independently of what outline-regexp is), so I think it's not a problem really: if the user wants to hide top-level exprs, he can install bicycle or a similar package.

Re ;* [^ \t\n]: the intention seems to require a space and then some non-whitespace text in order to treat triple semicolon as a heading, thus:

;;;                        -- not a heading
;;;foo                     -- not a heading
;;; foo                    -- heading

but it means that the "space and then a non-whitespace character" part is included in the match, and outline-insert-heading uses that match for the heading prefix, thus producing this behavior:

;;; heading:           -- M-x outline-insert-heading =>
;;; h 

So, if we do (setq-local outline-regexp ";;;;* ") instead, the behavior will be slightly different from lisp-mode - namely, an empty line prefixed with ;;; will be considered a heading - but I don't think that's a problem.

[bbatsov]

Shouldn't we at least expect some characters (e.g. letters/numbers/punctuation) after the ;;;? Empty headings would be pretty weird IMO.

[j-cr]

For example org-mode treats empty headings (* ) as headings still, so I think it's ok (or maybe even better that lisp-mode in terms of consistency, depending on your viewpoint).