Ok, I fianlly finished the defun navigation work, thanks to Alan’s suggestion, your experiment, and honorable sacrifices of my many brain cells. Now tree-sitter should give you both nested and top-level defun navigation for free, provided that the major mode sets treesit-defun-type-regexp. I’ll start a new thread about it.
I hope you can try and see if it works for bash-ts-mode, when you have time. You should only need to apply the attached patch and (goto-char (treesit--navigate-defun (point) …)) should just work. You can switch between top-level/nested with treesit-defun-tactic.
Cool. Thanks for all the work. I will test it and report my experience in the other thread.
Anyway, here is some minor points of the patch:
+;;; Tree-sitter navigation
+
+(defun sh-mode--treesit-defun-p (node)
+ "Return t if NODE is a function and nil otherwise."
+ (string-match treesit-defun-type-regexp
+ (treesit-node-type node)))
+
+(defun sh-mode-treesit-not-cs-p (node)
+ "Return t if NODE is *not* a compound-statement and nil otherwise."
+ (lambda (p)
+ (not (string-match "compound_statement"
+ (treesit-node-type p)))))
We probably want this to be an internal function, too, right?
I agree. Those are generic enough to be useful to other use cases.
+(defmacro sh-mode--treesit-parent-defun (node)
+ "Return nearest function-node that surrounds NODE, if any, or nil.
+
+This macro can be used to determine if NODE is within a function. If
+so, the macro evaluates to the nearest function-node and parent of NODE.
+Otherwise it evaluates to NIL."
+ `(treesit-parent-until ,node 'sh-mode--treesit-defun-p))
+
+(defmacro sh-mode--treesit-oldest-parent-in-defun (node)
+ "Return oldest parent of NODE in common function, if any, or NIL.
+
+This function returns the oldest parent of NODE such that the common
+parent is the nearest function-node."
+ `(treesit-parent-while ,node 'sh-mode--treesit-not-cp-p))
I'd prefer we use functions when functions will do, unless you have
particular reasons to use macros (for performance?).
Yes, I was thinking about performance but out of intuition and not evidence.
Diff artifact?
Yes, my bad.
Looks good to me, but I didn't scrutinize it line-by-line. If the new system works well, bash-ts-mode (and other major modes) wouldn't need to implemente its own navigation function. Sorry for not end up using your hard work, but your work definitely helped me to implement the more general version of defun navigation!
No problem at all. Trying things out has been a great opportunity for me to learn. Thank you for taking the time to review my patch.
Regarding the new navigation functions the do work as documented for bash-ts-mode. However, the overall behavior feels awkward to me. I will elaborate on the new thread.