bug#66719: mark-defun doesn't like function literals / the need for treesit-mark-defun

bug#66719: mark-defun doesn't like function literals / the need for treesit-mark-defun

[Dominik Honnef]

treesit.el currently defines treesit-beginning-of-defun and
treesit-end-of-defun, and binds C-M-a and C-M-e to them in addition to
setting beginning-of-defun-function and end-of-defun-function.

That way, C-M-a and C-M-e work well even for function literals, which do
not necessarily start and end on their own, dedicated lines. This
differs from beginning-of-defun, which always goes to the beginning
of the line.

The problem is that there is no treesit-mark-defun, and mark-defun calls
beginning-of-defun, via beginning-of-defun-comments. Thus, mark-defun
will mark more than the actual function literal. For example, in Go,
given

    // some comment
    someCall(func() {
        ...
    })

and a major mode that sets

    (setq-local treesit-defun-type-regexp (rx (or "function_declaration" "func_literal")))
    (setq-local end-of-defun-moves-to-eol nil)

mark-defun will also mark the call to someCall, and the comment above
it, while it should only mark the function literal.

Adding treesit-mark-defun would be a quick fix for the interactive use
case. However, this problem probably affects other functions that call
beginning-of-defun, or ones that build on top of mark-defun. I am not
sure how to address that, considering moving to the beginning of the
line is part of the contract of beginning-of-defun.

bug#66719: mark-defun doesn't like function literals / the need for treesit-mark-defun

[Eli Zaretskii]

> From: Dominik Honnef <dominik@honnef.co>
> Date: Mon, 23 Oct 2023 22:59:35 +0200
> 
> treesit.el currently defines treesit-beginning-of-defun and
> treesit-end-of-defun, and binds C-M-a and C-M-e to them in addition to
> setting beginning-of-defun-function and end-of-defun-function.
> 
> That way, C-M-a and C-M-e work well even for function literals, which do
> not necessarily start and end on their own, dedicated lines. This
> differs from beginning-of-defun, which always goes to the beginning
> of the line.
> 
> The problem is that there is no treesit-mark-defun, and mark-defun calls
> beginning-of-defun, via beginning-of-defun-comments.

mark-defun is defined for Lisp and Lisp-like languages, it is not a
function that's supposed to support any major mode.  If we want a
PL-independent mark-defun function, we need first to refactor
mark-defun to allow mode-specific behavior.