Re: Treesit Regression In ec4d29c4494f32acf0ff7c5632a1d951d957f084

[Yuan Fu <casouri@gmail.com>]

> On Sep 11, 2023, at 5:41 AM, Danny Freeman <danny@dfreeman.email> wrote:
> 
> 
> Yuan Fu <casouri@gmail.com> writes:
> 
>>> I take that back. It is indeed caused by that commit. I just had a wrong Emacs build when testing. In that commit, ts_tree_cursor_goto_first_child_for_byte seem to be the problem.
> 
> Well I'm glad to know I am capable of building Emacs with the latest
> tree sitter. Not as glad that the problem still exists lol
> 
> 
>> What I found is that, for some reason, ts_tree_cursor_goto_first_child_for_byte doesn’t always work
>> as expected in clojure [1]. In your example, if I evaluate (treesit-node-parent (treesit-node-at
>> (point))) with point at the beginning of “defn”, “foo”, “[]”, and “1”, only “defn” correctly
>> evaluates to the parent node. I tried this in other modes like c-ts-mode, but things seems to work
>> fine in all the few cases I tried.
>> 
>> So now the question is, is there anything tree-sitter-clojure does differently? Not saying it does, but that seems like a starting point.
> 
> Maybe there is. The grammar has not changed in a couple months, and
> those things that did change were related to keyword literals, which are
> not involved in the example I've been using to reproduce this.
> 
> Even knowing that. I will try to narrow the problem down even further,
> and see if I can find another grammar the exhibits this behavior.
> 
>> [1] Instead of returning the first child node of NODE that extends beyond POS, ts_tree_cursor_goto_first_child_for_byte (NODE, POS) returns false (meaning not found).
> 
> I remember we found an issue with a similarly named function here
> 
> https://github.com/tree-sitter/tree-sitter/issues/2012
> https://debbugs.gnu.org/cgi/bugreport.cgi?bug=60127
> 
> You ended up writing a workaround in emacs for this called
> treesit_cursor_first_child_for_byte
> 
> I don't think there is overlap between that change and this ec4d29 one.
> However, maybe there is overlap in the tree-sitter implementation? I'm
> not sure how to tell right now, but maybe this is an upstream
> tree-sitter problem that is just brought into the light by the commit in
> question here.

Duh, I can’t believe I completely forgot about this. Yes, this is exactly it, thank you for binging it up. Before tree-sitter can fix treesit_cursor_first_child_for_byte, I’ll just replicate the workaround and we should be golden. I’ve pushed a fix to emacs-29. Once master pulls from emacs-29, the problem should go away.

BTW, I should’ve opened a bug report and move the discussion there. But since we are mostly done here, and there’s another bug report to refer to, there’s no need to do that now.

Yuan