Re: AUCTeX and "builtin latex mode" integration

[Stefan Monnier]

> (a) The mode names change as follows:
>     plain-tex-mode -> plain-TeX-mode
>     latex-mode     -> LaTeX-mode
>     doctex-mode    -> docTeX-mode
>     context-mode   -> ConTeXt-mode
>     texinfo-mode   -> Texinfo-mode
>     ams-tex-mode   -> AmSTeX-mode
>     japanese-plain-tex-mode -> japanese-plain-TeX-mode
>     japanese-latex-mode     -> japanese-LaTeX-mode
>     context-en-mode  DELETED
>     context-nl-mode  DELETED
>     TeX-tex-mode     UNCHANGED
>     TeX-mode         NEW

LGTM.

> (a-2) Now `TeX-add-local-master' adds entry of new mode names such as
> %%% Local Variables:
> %%% mode: LaTeX      <-- not `latex'
> %%% End:

I don't like this very much: IMO the `mode:` cookie should really not specify
the "mode" to use but rather describe the file-type (Emacs could/should
aim to support "mime types" kind of names), and then which mode to use
for each file-type is decided based on the user's configuration.

> (b) All new modes (except TeX-tex-mode) are defined by
>     `define-derived-mode'. The inheritance relations are:
[...]
> (c) The compatibility with the former mode names with respect to
>     function call are retained by redirections or aliases.

No objection here.

> [Incompatibility]
> 1. directory local variables
>   Currently, directory local variables are prepared in a entry like
>  (latex-mode (reftex-label-alist . (("equation" ?e "eq:%f-" "\\eqref{%s}")
>                                     ("align" ?e "eq:%f-" "\\eqref{%s}")
>                                     ("gather" ?e "eq:%f-" "\\eqref{%s}")))))
>   in .dir-locals.el. In this feature branch, mode name is `LaTeX-mode',
>   thus this entry is ignored.
>
>   The facility of major-mode-remap-list still does not handle this
>   situation.
>
>   We can announce this fact and ask users to rewrite their
>   .dir-locals.el, but that can't be done if the user doesn't have write
>   access for that particular .dir-locals.el. (Maybe we can ignore such
>   corner cases?)

We should fix this for Emacs-30.
[ And then add some backward compatibility hack (presumably using
  an advice for when AUCTeX is used on Emacs<30).  ]

Maybe using `major-mode-remap-list` is not the right answer, tho (some
settings may be specific to a particular major mode and may cause
problems for other major modes used for the same file-type; I'm
thinking here of remapping between things like doc-view-mode and
pdf-view-mode, or image-mode and postscript-mode, or
image-mode and C-mode (for XPM files), ...).

We should make the directory-local code pay attention to:

- `derived-mode-p`.
- some additional "pseudo-derived-p" thingy for cases like `LaTeX-mode`
  where there isn't a true `derived-mode-p` relation, but the other mode
  can be treated as a sort of parent nevertheless.  Maybe we should change
  `derived-mode-p` so it handles that directly (i.e. offer some way for
  a major mode to declare itself as being a descendant of other
  modeS in addition to its "real" parent.  We could (re)user
  `define-child-mode` for that).

Maybe in addition to that, we need to be able to say in `dir-locals.el`
that a setting applies only for exactly this major mode, and not
its derivatives, maybe with a syntax like:

 ((= latex-mode) (reftex-label-alist
                  . (("equation" ?e "eq:%f-" "\\eqref{%s}")
                     ("align" ?e "eq:%f-" "\\eqref{%s}")
                     ("gather" ?e "eq:%f-" "\\eqref{%s}"))))

> 2. simple-minded reference to auto-mode-alist in emacs core
>   As discussed in [2], when elisp codes look at auto-mode-alist
>   directly and compare its content with `major-mode', new mode names
>   such as `LaTeX-mode' get no hits and lead to failure.
>   I added workaround for cedet/semantic/symref/grep.el, but similar
>   problematic code still exists at least in cedet/semantic/idle.el:
> ----------------------------------------------------------------------
> (defun semantic-idle-scheduler-work-parse-neighboring-files ()
> ...(snip)...
>     ;; Collect all patterns matching files of the same mode we edit.
>     (mapc (lambda (pat) (and (eq (cdr pat) major-mode)
>                            (push (car pat) matching-auto-mode-patterns)))
>         auto-mode-alist)
> ----------------------------------------------------------------------

IME, references to the `major-mode` variable in a package is a code
smell :-)

CEDET's code should obey `derived-mode-p` (and its own
`define-child-mode`), indeed.  It shouldn't be hard to do.


        Stefan