Re: AUCTeX and "builtin latex mode" integration

[Stefan Monnier]

>    This is much unsatisfactory. I'm currently thinking to apply the
>    attached patch to sort out the disorders of mode names. It works as
>    follows:
>    (a) It uses the names without `TeX-' prefix for LaTeX-mode,
>        plain-TeX-mode and TeX-mode.
>    (b) Those mode definitions are overwritten, prior to emacs 29, when
>        tex-mode.el is loaded. So the patch saves the function
>        definitions of AUCTeX major modes in the symbol property
>        `auctex-function-definition', and restores them after tex-mode.el
>        is loaded via `with-eval-after-load'.

Looks good to me (I mean: it doesn't literally look good, but given what
we have, this seems about as clean as it gets).

Only detail, I'd move the

    (if (eq (symbol-function 'plain-TeX-mode) #'plain-tex-mode)
        (defalias 'plain-TeX-mode nil))
    (if (eq (symbol-function 'LaTeX-mode) #'latex-mode)
        (defalias 'LaTeX-mode nil))
    (if (eq (symbol-function 'TeX-mode) #'tex-mode)
        (defalias 'TeX-mode nil))

into a function and merge it with the similar code which uses
`auctex-function-definition`.

> 2. ConTeXt modes
>    The design of ConTeXt modes doesn't match well with
>    `define-derived-mode' as explained below.
>    ConTeXt modes consist of "primary" mode `ConTeXt-mode' and "sub"
>    modes such as `ConTeXt-en-mode' and `ConTeXt-nl-mode'. Sub modes are
>    expected to work when they are called explicitly as well.
>    At first sight, it seems natural to implement the sub modes as
>    derived modes of the primary mode. However, simple implementation
>    doesn't work because the primary `ConTeXt-mode' is supposed to
>    detect the document's language (english, dutch...) and call the sub
>    mode according to the detected language. Thus simple implementation
>    leads to loop of the primary mode and duplication of hook calls.
>    I chose to implement the sub modes as derived modes of the primary
>    mode and move the "detecting the language and calling the sub mode"
>    phase of the primary mode into `change-major-mode-after-body-hook'.
>    In addition, the phase now bypasses call to the sub mode and calls
>    the "body" part of the sub mode independently. Hence the sub mode's
>    keymap and hook are ignored in that case. This might be considered as
>    a bad behavior.
>    In addition, I'm not sure whether such usage is suitable for
>    `change-major-mode-after-body-hook' or not.

That's not very pretty, indeed.  The fundamental problem you're trying
to solve is the same as the one we have with `tex-mode` since that
function is used both as "parent mode" and as "dispatch to the
appropriate child-mode".

In `tex-mode` we've tried various approaches over the years, none of
which are elegant :-(

What have now is:

    (define-derived-mode tex-mode text-mode "generic-TeX"
      [...]
      (tex-common-initialization))
    
    (advice-add 'tex-mode :around #'tex--redirect-to-submode
                ;; Give it lower precedence than normal advice, so
                ;; AUCTeX's advice takes precedence over it.
                '((depth . 50)))
    (defvar tex-mode--recursing nil)
    (defun tex--redirect-to-submode (orig-fun)
      [...]
      (if (or delay-mode-hooks tex-mode--recursing)
        [...]))

In the case of `ConTeXt-mode`, maybe the better option is to make the
submodes be minor modes?

If at all possible, I recommend to use different names for the parent
mode and the dispatch function :-)

> 3. Additional menu item
>    I chose `text-mode' as the parent of the base mode
>    `TeX--VirTeX-mode' and `Texinfo-mode'. This is reasonable because
>    those modes called `text-mode-hook' so far. On the other hand, there
>    is a side effect that "Text" menu appears in the menu bar in all
>    AUCTeX major modes, which may be distracting for some users.

Maybe `nil` (a.k.a. `fundamental-mode`) is a better parent, which more
closely matches the historical behavior of AUCTeX?

TeX is one of those formats that sits halfway between `text-mode` and
`prog-mode` (like (SG|X)ML, as well).

> 4. Directory local variables
>    (a) I simplified the way to have `LaTeX-mode' as the value of
>        `major-mode' of `japanese-LaTeX-mode' while it can read directory
>        local variables prepared for japanese-LaTeX-mode. Now there is a
>        line
> :after-hook (setq major-mode 'LaTeX-mode)
>        in the definition of `(define-derived-mode japanese-LaTeX-mode...'
>        and cumbersome interaction with `hack-local-variables' is
>        discarded.
>        This solution works for most cases, but would break if the mode
>        hooks contain user codes which assume mode name without
>        `japanese-' prefix.
>        If it turns out that such breakage does exist and no workaround
>        is found, we'd have to go back to the former approach.

FWIW, that's a cleaner solution than what I had come up when I looked
at it.


        Stefan