Re: electric-pair-mode as a minor mode?

[Stefan Monnier]

>     (defun message--syntax-propertize (beg end)
>       ;; first, remove all `syntax-table' properties
>       (remove-text-properties beg end '(syntax-table nil))

This should not be necessary (it's done by syntax-propertize before it
calls syntax-propertize-function).

>       ;; propertize smileys as "generic comments"
>       ;;
>       (goto-char beg)
>       (while (search-forward-regexp ":-?[()]" end 'noerror)
>         (add-text-properties (match-beginning 0) (match-end 0)
>                              '(syntax-table (14 . nil))))

14 is definitely not right.  A "comment fence" means "this char starts
or ends a comment, depending on whether we're inside a comment".
So putting this on 3 consecutive chars tells Emacs that the first
2 chars make up a comment and the 3rd is the beginning of a comment
(which might extend til the end of the buffer).

>       ;; idem for citations
>       ;;
>       (goto-char beg)
>       (while (search-forward-regexp
>               (concat "^" message-cite-prefix-regexp ".*")
>               end 'noerror)

You might want to move the "concat" out of the loop ;-)

>         (add-text-properties (match-beginning 0) (match-end 0)
>                              '(syntax-table (14 . nil)))))

> * I have no idea how "expensive" this naive approach is.

Shouldn't be much more expensive than doing the same thing in font-lock.

Not sure who's considered maintainer of message.el, but you should check
with Lars, see what he thinks of it.

> * Although the `syntax-table' property seems to be set correctly, and
>   both show-paren mode and electric-pair-mode seem to dtrt, I saw many
>   cases where M-: (nth 4 (syntax-ppss)) didn't return non-nil when it
>   should have, and vice versa. Is this part of the
>   `parse-sexp-lookup-properties' sematics, that `syntax-ppss' becomes
>   meaningless?

That's because you mis-used the comment markers.  They should only be
put at the beginning and at the end of the comment.

> * I'm using `message-cite-prefix-regexp' to detect citations. It's what
>   font-lock in lisp/gnus/message.el uses so it seems ideal. But there
>   are also a lot of vars like `message-yank-prefix', whose docstring
>   contains the following line which baffles me:
>> Fix `message-cite-prefix-regexp' if it is set to an abnormal value.

Hopefully Lars can say something useful about that.


        Stefan