Chiming in with my minor voice. I prefer my mode configurations generally self contained and not centralized as a single global configuration variable would encourage. When I share my configurations with others, it also suggests that if they adopt a global configuration variable on top of what I share, it will introduce subtle changes in behavior that I may not at first understand the source of when helping others. I definitely prefer proper lisp over an opaque configuration variable.

-Stephane

On Fri, Nov 29, 2024 at 12:04 PM Stefan Monnier <monnier@iro.umontreal.ca> wrote:
> Since Emacs 29, I see many people ask about how to configure
> tree-sitter modes by setting some variable.

This is not specific to tree-sitter.

> It seems that people much prefer setting a variable than adding
> a major mode hook that calls some functions.

But that boxes them into the subset of possibilities that have been
pre-imagined by those who designed the set of variables.

> What do you guys think about something like this:
>
> (setq treesit-global-configuration
>       '((c-ts-mode
>          ;; Set treesit-font-lock-level to 4
>          (font-lock-level . 4)
>          ;; Disable tree-sitter’s outline support
>          (outline . disable)
>          ;; Enable these features on top of the default ones.
>          (font-lock-enable . (function property variable))
>          ;; Disable these features.
>          (font-lock-disable . (definition))
>          ;; Add extra font-lock rules
>          (font-lock-extra-rules
>           ( :feature 'my-rules
>             :language 'c
>             ((some_query) @some-face)))
>          (simple-indent-extra-rules
>           (c
>            (matcher anchor offset))
>           (d
>            (matcher anchor offset)))
>          )))

Sounds to me like this is inventing a new programming language, just one
that's a lot more restrictive than ELisp.

Is it really better than ELisp which could look like:

    (add-hook 'c-ts-mode-hook #'my-c-ts-mode-preferences)
    (defun my-c-ts-mode-preferences ()
      (setq-local treesit-font-lock-level 4)
      (treesit-outline-mode -1)
      (treesit-font-lock-enable '(function property variable))
      (treesit-font-lock-disable '(definition))
      (treesit-font-lock-add-rules
       '( :feature 'my-rules
          :language 'c
          ((some_query) @some-face)))
      (treesit-indent-add-rules
       '((c
          (matcher anchor offset))
         (d
          (matcher anchor offset)))
      ))

Admittedly, the add-hook/defun dance could be improved, and that would
benefit more than just tree-sitter.  E.g. a new macro like

    (defmacro custom-set-hook (hook &rest body)
      (declare (indent 1) (debug (sexp def-body)))
      (let ((funname (intern (format "custom-set-hook--%s" hook))))
        `(progn (add-hook ',hook #',funnmame)
                (defun ,funname () ,@body))))

> One thing I don’t like is how it handles languages. In this POC
> language-specific settings are nested under the mode. I’m ok with
> mode-language hierarchy, but the nesting adds a lot of nesting levels
> to the variable.  And the language nesting isn’t consistent, some
> settings have language nesting, some don’t.

I don't really understand what you're referring to, but for
`treesit-font-lock-add-rules` would could try and auto-add the `:language`?


        Stefan