Treesit.el Support

[AndrewSwerlick]

Hey @jcs090218 thanks for this great package, I think its a hugely important feature set in the emacs tree-sitter ecosystem. I'm curious what your plans are with regards to the new [built in treesit.el](https://github.com/emacs-mirror/emacs/blob/emacs-29/lisp/treesit.el) that is shipping with Emacs 29? Are you interested in having this package work with that?

I started experimenting with what that might mean in this PR and found the process of integrating with treesit.el to be pretty straight forward, but I wanted to understand how you were thinking about this before going any further.

If you're interested, I can go further with the PR. There are some unanswered questions, like would you want older versions to support the legacy emacs-tree-sitter package, and new versions to only support treesit.el, or one version that could swap between them etc. Interested to hear your thoughts.

[jcs090218]

Hi, Andrew! Thanks for liking this package! ❤️ Yes, I am interested to see things going with built-in tree-sitter support!

I am thinking of creating another package for built-in treesit support since ts-fold is heavily designed with emacs-tree-sitter. It depends on the complexity of switching between the two APIs! If it were easy to implement and wouldn't burden maintainers, then it should be good to build together!

If we go for the first route, it will be more straightforward since we would only have one abstraction layer regarding treesit. If we decide to build together, it's better to support both tree-sitter abstraction layers. The ultimate goal is to move from legacy emacs-tree-sitter to built-in treesit, so it's fine to drop the legacy support at some point. We should wait until treesit becomes popular, or consider stable.

Thank you for working on this. I might wait until Emacs 29 is out before I start working on this! Feel free to continue working on this! 😄

[AndrewSwerlick]

Cool, what I'll probably do is keep playing with my prototype to see if I can easily support both approaches in a single version. That will make it easier to merge changes from upstream too.

[garyo]

I'm also super interested in this! Would be willing to help.

[AndrewSwerlick]

@garyo I have a fork working here that I'm using daily with both ruby and js. It seems to be working well.

I explored trying to get something that could easily toggle between built in tree sitter and the package, and I couldn't do it without basically writing a big whole compatibility later, which didn't seem worth it.

In general the transformation was just a find and replace though, with a few minor complications

[garyo]

I'm trying your fork now (the andrew-sw/treesit-el-support branch of that fork). I had to modify the ts-fold-range-alist to add the new -ts- major modes in the latest emacs, e.g. c++-ts-mode, but then it works OK.

Perhaps an alternative to doubling the size of that alist would be to look up using symbols that remove the "-ts", something like (intern (replace-regex-in-string "-ts-mode" "-mode" (symbol-value major-mode))?

The indicator mode isn't working yet but I can look into that.

[garyo]

I also found that ts-fold-close-all needs this small patch:

@@ -307,7 +319,7 @@ If the current node is not folded or not foldable, do nothing."
   (interactive)
   (ts-fold--ensure-ts
     (let* ((node (treesit-buffer-root-node))
-           (patterns (mapconcat (lambda (fold-range) (concat "(" (car fold-range) ") " "@name"))
+           (patterns (mapconcat (lambda (fold-range) (concat "(" (symbol-name (car fold-range)) ") " "@name"))
                                  (alist-get major-mode ts-fold-range-alist) " "))
            (query (treesit-query-compile (treesit-node-language node) patterns))
            (nodes-to-fold (treesit-query-capture node query nil nil t)))

[garyo]

@AndrewSwerlick patches sent to your fork. Thanks for that!

[AndrewSwerlick]

I'm trying your fork now (the andrew-sw/treesit-el-support branch of that fork). I had to modify the ts-fold-range-alist to add the new -ts- major modes in the latest emacs, e.g. c++-ts-mode, but then it works OK.

Perhaps an alternative to doubling the size of that alist would be to look up using symbols that remove the "-ts", something like (intern (replace-regex-in-string "-ts-mode" "-mode" (symbol-value major-mode))?

It seems we could remove the non ts-mode items if we aren't going to try to support both versions at the same time.

[CrackedBeefcake]

what is the status of this? is it considered safe to use ts-fold with emacs@29?

[jcs090218]

Hi, I will work on this after April, when 29.1 is released. I've tried 29.0.60 but couldn't get it to run smoothly. I will investigate more once I better understand treesit.el.

To answer your question, yes, it's safe to use this package in 29. However, it does not use the built-in treesit module. Hope the built-in tree-sitter support will be soon in the near future!

[garyo]

I am using Andrew's fork plus some patches (see above), my patches at AndrewSwerlick#1, with emacs built from trunk a few days ago, with built-in treesit module. There are still some things to iron out with treesit-based major modes but it is basically functional.

[AndrewSwerlick]

I'll add I use my fork on my daily driver for both ruby and typescript with minimal issues. I haven't tested extensively with other languages though.

[jcs090218]

Thanks for all the information! One major issue is I use windows pretest snapshot, so it's not always on the latest master branch. Emacs-devel may have made some changes, so it causes API to be incompatible (it's normal). I will make sure I go deeper next time I try it out! ;)