Code-folding using Emacs' builtin tree-sitter (
treesit
)
THIS IS A WORK IN PROGRESS, IT IS NOT STABLE
treesit-fold
is a fork of
[ts-fold](https://github.com/emacs-tree-sitter/ts-fold/actions/workflows/test.yml) that uses Emacs 29+
builtin tree-sitter (treesit
) to provide code folding based on the tree-sitter
syntax tree.
(use-package treesit-fold
:straight (treesit-fold :type git :host github :repo "abougouffa/treesit-fold"))
git clone https://github.com/abougouffa/treesit-fold /path/to/lib
then in Emacs:
(add-to-list 'load-path "/path/to/lib")
(require 'treesit-fold)
or
(use-package treesit-fold
:load-path "/path/to/lib")
The following are the functions provided by treesit-fold-mode
Commands for enabling treesit-fold
:
Commands | Description |
---|---|
treesit-fold-mode |
enable treesit-fold-mode in the current buffer. |
global-treesit-fold-mode |
enable treesit-fold-mode whenever tree-sitter is turned on and the major mode is supported by treesit-fold . |
treesit-fold-indicators-mode |
enable treesit-fold with indicators in the current buffer. See plugins section. |
global-treesit-fold-indicators-mode |
enable treesit-fold with indicators globally. See plugins section. |
treesit-fold-line-comment-mode |
enable line comment folding. |
Commands for using treesit-fold
.
Commands | Description |
---|---|
treesit-fold-close |
fold the current syntax node. |
treesit-fold-open |
open the outermost fold of the current syntax node. Keep the sub-folds close. |
treesit-fold-open-recursively |
open all folds inside the current syntax node. |
treesit-fold-close-all |
close all foldable syntax nodes in the current buffer. |
treesit-fold-open-all |
open all folded syntax nodes in the current buffer. |
treesit-fold-toggle |
toggle the syntax node at `point'. |
If evil mode is loaded, then these commands are also added to the evil folding list.
β οΈ Please sort these two lists alphabetically!
These languages are fairly complete:
- Arduino / Assembly
- Bash / Beancount
- C / C++ / C# / Clojure / CMake / CSS
- Dart
- Elisp / Elixir / Erlang
- Fish
- GDScript / GLSL / Go / Groovy (Jenkinsfile)
- Haskell / HLSL / HTML
- Jai / Java / JavaScript / JSX / JSON / Jsonnet / Julia
- Kotlin
- LaTex / Lisp / Lua
- Make / Markdown
- Nix / Noir
- OCaml / Org
- Perl / PHP / Python
- R / Ruby / Rust / reStructuredText
- Scala / Scheme / SQL / Swift
- TOML / TypeScript / TSX
- Verilog / VHDL
- XML
- YAML
- Zig
These languages are in development:
- Ada
- Agda
- Elm
- Fortran
- Pascal
- Smithy
Although treesit-fold
aims to have good folding out of the box for all supported
definitions, people will indubitably have their own preferences or desired
functionality. The following section outlines how to add your own folding
definitions and folding functions to make treesit-fold
work for you. If there
are any improvements you find for existing or new languages, please do raise a
PR so that others may benefit from better folding in the future!
treesit-fold
defines all its folding definitions in the variable
treesit-fold-range-alist
which is an alist with the key of the alist being the
mode and the value being another alist of fold definitions.
;; Example of treesit-fold-range-alist's structure
'((c-mode . c-folding-definitions) ;; <language>-folding-definitions is structured as shown below
(css-mode . css-folding-definitions)
(go-mode . go-folding-definitions)
(scala-mode . scala-folding-definitions)
...)
;; Examle of a folding definition alist
(setq css-folding-definitions
(block . treesit-fold-range-seq)
(comment . treesit-fold-range-c-like-comment))
So you can select whatever node that you want to fold on it.
To find what node you'll want to fold closed, refer to the tree-sitter
documentation about viewing nodes. tree-sitter-debug
and
tree-sitter-query-builder
are both very useful for this.
For the folding functions, treesit-fold provides some default
-
treesit-fold-range-seq
- Folds from the start of the node to the end of the node leaving a buffer of one character on each side. Usually used for code blocks that have bracketing delimiters.int main() { // <-- start of tree-sitter block node printf("Hello, World\n"); return 0; } // <-- end of tree-sitter block node // | // | '(block . treesit-fold-range-seq) // V int main() {...} // Folded node
-
treesit-fold-range-markers
- Folds the node starting from a giving delimiter character. Useful if tree-sitter's node definition doesn't align with the start of the desired folding section.NOTE: This folding function requires a lambda (or an externally defined function wrapper) so that the delimiter can be specified. You usually don't need to worry about the
node
andoffset
variables, so just pass them through.
type Dog interface {
Bark() (string, error)
Beg() (bool, error)
}
/* | Note: The tree-sitter node starts at the word interface, not at the '{'.
* | '(interface_type . (lambda (node offset)
* | (treesit-fold-range-markers node offset "{" "}")))
* V
*/
type Dog interface {...}
-
treesit-fold-range-block-comment
- Folds multi-line comments that are of the form/*...*/
. Should show a summary if the commentary plugin is turned on./* * The main function that gets run after program is compiled * Doesn't take any parameters */ int main() { printf("Hello, World\n"); return 0; } // | // | '(comment . treesit-fold-range-block-comment) // V /* <S> The main function that gets run after program is compiled */ int main() { printf("Hello, World\n"); return 0;
-
treesit-fold-range-line-comment
- For languages that have one line comment blocks with the comment delimiter starting each line. Condenses all the comment nodes into a single fold.Note: This folding function requires a lambda (or an externally defined function wrapper) so that the delimiter can be specified. You usually don't need to worry about the
node
andoffset
variables, so just pass them through.# show the long form of ls # and also display hidden files alias ll='ls -lah' # | # | (comment (lambda (node offset) (treesit-fold-range-line-comment node offset "#")))) # V # show the long form of ls... alias ll='ls -lah'
-
treesit-fold-range-c-like-comment
- A shortcut for the large number of languages that have the c style comment structures/*...*/
and// ...
. Smartly picks the correct folding style for the comment on the line./* * The main function that gets run after program is compiled * Doesn't take any parameters */ int main() { // print hello world // and a new line printf("Hello, World\n"); return 0; } // | // | '(comment . treesit-fold-range-c-like-comment) // V /* <S> The main function that gets run after program is compiled */ int main() { // <S> print hello world printf("Hello, World\n"); return 0;
Now that you know what kinds of folds are easily available in treesit-fold, you can
go ahead and add new fold definitions to treesit-fold-range-alist
and be good to go!
Let's look at a quick example of adding a new folding definition. Let's say you
want to add folding to go-mode
's field_declaration_list
. The folding
definition that is needed will be
'(field_declaration_list . treesit-fold-range-seq)
. To add this to the
treesit-fold-range-alist
, you can do something like the following.
(push '(field_declaration_list . treesit-fold-range-seq) (alist-get 'go-mode treesit-fold-range-alist))
Now the new fold definition should be usable by treesit-fold!
With the functions listed above you'll be able to define most folding behavior that you'll want for most languages. However, sometimes you'll have a language where the delimiter is a word instead of a single character bracket and you want to offset your fold by a certain amount to accommodate it. That's where offsets come in. When adding a fold definition to a a language's fold alist, you can either provide the folding function directly as you've seen so far:
'(block . treesit-fold-range-seq)
Or you can provide the folding function with an offset:
'(block . (treesit-fold-range-seq 1 -3))
When a range is provided, it provides extra room on the ends of a fold. The way
this works is most easily shown using an example. Lets say we want to write a
fold for bash's for...do...done
construct to look something like this:
for i in 1 2 3 4 5
do
echo "Welcome $i times"
done
# |
# | '(do_group . <some folding function>)
# V
for i in 1 2 3 4 5
do...done
The do...done
block is represented in tree-sitter as the node named
do_group
. However, if we just use '(do_group . treesit-fold-range-seq)
, then
we'll get results like the following:
for i in 1 2 3 4 5
d...e
which is hard to read. Instead, we can use the definition
'(do_group . (treesit-fold-range-seq 1 -3))
to offset the fold a bit to get our
desired result!
If the built in functions don't fit your needs, you can write your own fold parser! Folding functions take two parameters:
node
- the targeted tree-sitter node; in this example,block
will be the targeting node.offset
- (optional) a cons of two integers. This is handy when you have a similar rule with little of positioning adjustment.
Then the function needs to return a position range for the fold overlay in the
form '(start-of-fold . end-of-fold)
. If nil
is returned instead of a range,
then no fold is created. This can be useful if you want to add extra conditional
logic onto your fold.
As an example of a folding function, take a look at the definition of the
basic treesit-fold-range-seq
.
(defun treesit-fold-range-seq (node offset)
"..."
(let ((beg (1+ (tsc-node-start-position node))) ; node beginning position
(end (1- (tsc-node-end-position node)))) ; node end position
(treesit-fold--cons-add (cons beg end) offset))) ; return fold range
treesit-fold comes with a couple of useful little additions that can be used or turned off as desired.
This plugin adds interactive visual markers in the gutter that show where folds can be made. They can be clicked on to fold or unfold given nodes.
treesit-fold-indicator-mode
is loaded when treesit-fold-mode
is and the functionality
should be auto-loaded in, however if that's not working then you may want to
explicitly declare the package in in your config.
-
use-package
(use-package treesit-fold-indicators :straight (treesit-fold-indicators :type git :host github :repo "emacs-tree-sitter/treesit-fold"))
-
(add-to-list 'load-path "/path/to/lib") (require treesit-fold)
or
(use-package treesit-fold-indicators :load-path "/path/to/lib")
You can then enable this manually by doing either of the following:
M-x treesit-fold-indicators-mode
M-x global-treesit-fold-indicators-mode
Please note that turning on treesit-fold-indicators-mode
automatically turns on
treesit-fold-mode
as well. Though, turning off treesit-fold-indicators-mode
does not
turn off treesit-fold-mode
-
To enable this automatically whenever
tree-sitter-mode
is enabled, use the global indicator mode:(global-treesit-fold-indicators-mode 1)
Else, a hook can be added to tree-sitter directly.
(add-hook 'tree-sitter-after-on-hook #'treesit-fold-indicators-mode)
-
To switch to left/right fringe: (Default is
left-fringe
)(setq treesit-fold-indicators-fringe 'right-fringe)
-
To lower/higher the fringe overlay's priority: (Default is
30
)(setq treesit-fold-indicators-priority 30)
-
To apply different faces depending on some conditions: (Default is
nil
)For example, to coordinate line-reminder with this plugin.
(setq treesit-fold-indicators-face-function (lambda (pos &rest _) ;; Return the face of it's function. (line-reminder--get-face (line-number-at-pos pos t)))) (setq line-reminder-add-line-function (lambda (&rest _) (null (treesit-fold--overlays-in 'treesit-fold-indicators-window (selected-window) (line-beginning-position) (line-end-position))))) (advice-add 'line-reminder-transfer-to-saved-lines :after ;; Refresh indicators for package `treesit-fold'. #'treesit-fold-indicators-refresh)
This plugin automatically extracts summary from the comment/document string, so you can have a nice way to peek at what's inside the fold range.
-
If you don't want this to happen, do: (Default is
t
)(setq treesit-fold-summary-show nil)
-
Summary are truncated by length: (Default is
60
)(setq treesit-fold-summary-max-length 60)
-
The exceeding string are replace by: (Default is
"..."
)(setq treesit-fold-summary-exceeded-string "...")
-
To change summary format: (Default is
" <S> %s "
)(setq treesit-fold-summary-format " <S> %s ")
Just like with fold definitions, you can create your own summary definitions.
Summary definitions are defined in treesit-fold-summary-parsers-alist
and has one
summary function per major mode '(java-mode . fold-summary-function)
. The
summary function takes in the doc string which is all the text from a doc node
and then returns a string to be displayed in its stead. Unlike with the folding
functions, there aren't a set of general summary functions to fall back on.
However, there are lots of examples and helper functions present in
treesit-fold-summary.el
. Let's look at one example here.
(defun treesit-fold-summary-javadoc (doc-str)
"Extract summary from DOC-STR in Javadoc."
(treesit-fold-summary--generic doc-str "*")) ;; strip the '*' and returns the first line
As can be seen treesit-fold-summary--generic
is a very helpful function since it
removes the provided delimiter and returns the first line. often this will be
enough.
This plugin makes line comment into foldable range.
M-x treesit-fold-line-comment-mode
Enable tree-sitter-mode
first, then tree-sitter-query-builder
is useful to test
out queries that determine what syntax nodes should be foldable and how to fold
them. emacs-tree-sitter
has an excellent documentation on how to write tree-sitter
queries.
When adding a new folding parser, add the folding definition function to
treesit-fold.el
itself near where the other range functions live and then add the
parser to treesit-fold-parsers.el
file. Finally, if you are adding support for a
new language, remember to add it to the treesit-fold-range-alist
variable.
When creating a new parser, name it treesit-fold-parsers-<language>
.
When creating a new folding function, name it
treesit-fold-range-<language>-<feature>
or something similar.
Here are some techniques for finding your desired nodes in tree-sitter.
To look for the correct node you have three options:
- look at the
tree-sitter-[lang]/grammar.js
implementation. In the above example,block
node is defined in the tree-sitter-c-sharp'sgrammar.js
file - open a file of your language choice in emacs and
M-x tree-sitter-debug-mode
. This will display the whole s-expr representing your file (message "%S" (tsc-node-to-sexp))
in your function to display what your function is seeing
β οΈ WarningMake sure you look into the correct repository. Repositories are managed under tree-sitter-langs's using git submodule. Some tree-sitter module aren't using the latest version!
treesit-fold-summary.el
module is used to extract and display a short description
from the comment/docstring.
To create a summary parser, you just have to create a function that could
extract comment syntax correctly then register this function to
treesit-fold-summary-parsers-alist
defined in treesit-fold-summary.el
.
The display and shortening will be handled by the module itself.
Functions should be named with the prefix treesit-fold-summary-
followed by
style name
. For example, to create a summary parser for Javadoc style, then it
should be named treesit-fold-summary-javadoc
.