Re: Or probably just fix the org-ctags hook functions? (was: Should we accept breaking changes ...)

[Jens Schmidt]

On 2023-08-08  10:48, Ihor Radchenko wrote:

> The situation is a bit more complex.

To make it even more complex, here a different point of view.  Sorry,
I don't have any earlier mails ready I could reply to ...

* TL;DR

Probably the problem is not the side-effects done by loading
=org-ctags=, but rather that these hook function from =org-ctags= try
to do their job when the environment is not really ready for them,
i.e. when there is no known tag file around.

* Full Story

I tried to get this thing running.  Some issues and observations I
came across:

1. When I execute function `org-ctags-create-tags' in file
   =org-manual.org= of Org =main=, an empty =TAGS= file gets created.
   Presumably because the asterisk in the generated command line is
   quoted and the warning generated by that (=cannot open source file
   "/home/jschmidt/work/org-mode/doc/*"=) is not shown in Emacs:

   #+begin_example
     ctags-exuberant --langdef=orgmode --langmap=orgmode:.org \
       --regex-orgmode="/<<([^>]+)>>/\1/d,definition/" \
       -f "/home/jschmidt/work/org-mode/doc/TAGS" -e -R \
       "/home/jschmidt/work/org-mode/doc/*"
      ^^^                                ^^^
   #+end_example

   If I execute the statement on the command line w/o the quotes on
   the final parameter, I get a non-empty =TAGS= file.

   Besides that, I somehow doubt that =-R .../*= (=-R= meaning
   recursive operation) makes actually sense, this probably should be
   just =-R ...= instead.

   Besides *that* the following sexps from the function look fishy:

   #+begin_src emacs-lisp
     (expand-file-name (concat dir-name "/TAGS"))
     (expand-file-name (concat dir-name "/*"))
   #+end_src

   Why not =(expand-file-name "TAGS" dir-name)=?

2. When one looks at the =TAGS= file generated thus, one immediately
   notices that the regular expression from parameter
   =--regex-orgmode= used to match these <<targets>> matches also
   non-targets, like in the following example section:

     #+begin_example
       1. one item
       2. <<target>>another item
       Here we refer to item [[target]].
     #+end_example

   But that is probably intentional or not avoidable.  Probably these
   are even valid Org targets?

Ok, that was more or less cheap criticism.  Or a bug report?  Anyway,
what is more interesting is:

3. When one later visits =org-manual.org=,\\
   and there is a =TAGS= file in that directory,\\
   and =org-ctags= previously has been loaded, then,\\
   by virtue of the following snippet from =org-ctags.el=:

   #+begin_src emacs-lisp
     (add-hook 'org-mode-hook
          (lambda ()
            (when (and org-ctags-enabled-p
                       (buffer-file-name))
              ;; Make sure this file's directory is added to default
              ;; directories in which to search for tags.
              (let ((tags-filename
                     (expand-file-name
                      (concat (file-name-directory (buffer-file-name))
                              "/TAGS"))))
                (when (file-exists-p tags-filename)
                  (visit-tags-table tags-filename))))))
   #+end_src

   that =TAGS= file gets automatically visited, meaning that future
   tag look-ups with =C-c C-o= do not ask about any tag files to load.

   (Yes, there is again that funny =(expand-file-name (concat ...))=
   pattern used above.)

4. From that and also from the header documentation:

   #+begin_example
     ;; When you click on a link "[[foo]]" and org cannot find a
     ;; matching "<<foo>>" in the current buffer, the tags
     ;; facility will take over.  The file TAGS in the active
                                      ^^^^^^^^^^^^^^^^^^^^^^^
     ;; directory is examined to see if the
        ^^^^^^^^^
     ;; tags facility knows about "<<foo>>" in any other files.
     ;; If it does, the matching file will be opened and the
     ;; cursor will jump to the position of "<<foo>>" in that
     ;; file.
   #+end_example

   I conclude that the whole =org-ctags= functionality is based on the
   assumption that a =TAGS= file lives in the working directory of the
   currently visited Org mode file.  Why not test for that condition
   in the hook functions:

   #+begin_src emacs-lisp
     org-ctags-find-tag
     org-ctags-ask-rebuild-tags-file-then-find-tag
     org-ctags-rebuild-tags-file-then-find-tag
     org-ctags-ask-append-topic
     org-ctags-append-topic
     org-ctags-ask-visit-buffer-or-file
     org-ctags-visit-buffer-or-file
     org-ctags-fail-silently
   #+end_src

   in some way or other, probably also testing variable
   =tags-file-name=, and just skip their execution returning =nil=, if
   the environment does not seem to be ready for a tag search.  Then
   regular link operation would kick in.

Of course, that is all based on educated guesses and ad-hoc poking in
the code, not on long-term usage.  Is there a known user of
=org-ctags= who one could ask?

I think the number of people using =org-ctags= is much smaller than
the number of people not using it, in particular because of the issue
described in the first item.  If above assumption is true and hook
functions get wrapped as indicated, we could keep the former happy
without affecting the latter with unexpected and inexplicable tag file
prompts.