Re: New defun navigation for tree-sitter (Was: Code navigation for sh-mo

Great job, Yuan.

The new `treesit--navigate-defun' does work as documented. However, the behavior does feel a little awkward to me.

For instance, if I am within a function I would expect to go to its beginning when pressing C-M-a, and not to the beginning of the first leaf function when searching backward. Although the current behavior might be desirable to some users and should be possible as well.

The navigation style that I came up with for bash-ts-mode is to navigate only to functions in the same level or higher of the tree. That seems to be the motions I usually rely on when writing/editing code. But that might just be me.

Nevertheless, I do agree with you that large changes would be required to enable proper navigation in the presence of nested functions via the existing beginning/end-of-defun functions.

On Mon, Dec 12, 2022 at 10:20 PM Yuan Fu <casouri@gmail.com> wrote:

Thanks to Alan and João, I have revamped the defun navigation of tree-sitter, now major modes can get accurate defun navigation by setting treesit-defun-type-regexp (as before), and switch between top-level-only mode and nested defun mode by treesit-defun-tactic (new).

The change I pushed doesn’t replace the old functions, my plan is to replace the existing ones if we think this change is ok for the release branch. After experimenting and implementing this, I don’t think the current beginning/end-of-defun can support nested navigation without large changes. So I’d prefer we use new beg/end-of-defun commands in tree-sitter major modes. And we later think about merging the two.

I also added comprehensive tests to the defun navigation[1], which should give us more peace in mind. Although this is a large change, I hope it can stay on the release branch because (1) it fixes current bugs (2) it has a comprehensive test.

Yuan

[1] For tests, see
treesit-defun-navigation-nested-1
treesit-defun-navigation-nested-2
treesit-defun-navigation-nested-3
treesit-defun-navigation-top-level

It tests all possible defun navigations starting from all possible starting positions (that I can think of), so it should be pretty comprehensive. I currently have test for python, js, and bash. New tests can be easily added.

The test program is like the following, we mark positions with markers like [100]

[100]
[101]class Class1():
[999] prop = 0
[102]
[103]class Class2():[0]
[104] [1]def method1():
[999] [2]return 0[3]
[105] [4]
[106] [5]def method2():
[999] [6]return 0[7]
[107] [8]
[999] prop = 0[9]
[108]
[109]class Class3():
[999] prop = 0[10]
[110]

Then we have a list of positions like this, basically saying “if point is at marker X, and we find the next/previous beginning/end of defun, point should end up at Y”, and we have X for all the possible positions in a nested defun, and Y for all four operations. So the first line says “if we start at marker 0, and find prev-beg-of-defun, we should end up at marker 103”.

;; START PREV-BEG NEXT-END PREV-END NEXT-BEG
'((0 103 105 102 106) ; Between Beg of parent & 1st sibling.
(1 103 105 102 106) ; Beg of 1st sibling.
(2 104 105 102 106) ; Inside 1st sibling.
(3 104 107 102 109) ; End of 1st sibling.
(4 104 107 102 109) ; Between 1st sibling & 2nd sibling.
(5 104 107 102 109) ; Beg of 2nd sibling.
(6 106 107 105 109) ; Inside 2nd sibling.
(7 106 108 105 109) ; End of 2nd sibling.
(8 106 108 105 109) ; Between 2nd sibling & end of parent.
(9 103 110 102 nil) ; End of parent.

(100 nil 102 nil 103) ; Before 1st parent.
(101 nil 102 nil 103) ; Beg of 1st parent.
(102 101 108 nil 109) ; Between 1st & 2nd parent.
(103 101 108 nil 109) ; Beg of 2nd parent.
(110 109 nil 108 nil) ; After 3rd parent.
)

From:	João Paulo Labegalini de Carvalho
Subject:	Re: New defun navigation for tree-sitter (Was: Code navigation for sh-mode with Tree-sitter)
Date:	Tue, 13 Dec 2022 09:11:41 -0700