Language Extensions

Language support in Zed has several components:

• Language metadata and configuration
• Grammar
• Queries
• Language servers

Language Metadata

Each language supported by Zed must be defined in a subdirectory inside the languages directory of your extension.

This subdirectory must contain a file called config.toml file with the following structure:

name = "My Language"
grammar = "my-language"
path_suffixes = ["myl"]
line_comments = ["# "]

• name (required) is the human readable name that will show up in the Select Language dropdown.
• grammar (required) is the name of a grammar. Grammars are registered separately, described below.
• path_suffixes is an array of file suffixes that should be associated with this language. Unlike file_types in settings, this does not support glob patterns.
• line_comments is an array of strings that are used to identify line comments in the language. This is used for the editor::ToggleComments keybind: <kbd class="keybinding">|</kbd> for toggling lines of code.
• tab_size defines the indentation/tab size used for this language (default is 4).
• hard_tabs whether to indent with tabs (true) or spaces (false, the default).
• first_line_pattern is a regular expression, that in addition to path_suffixes (above) or file_types in settings can be used to match files which should use this language. For example Zed uses this to identify Shell Scripts by matching the shebangs lines in the first line of a script.

Grammar

Zed uses the Tree-sitter parsing library to provide built-in language-specific features. There are grammars available for many languages, and you can also develop your own grammar. A growing list of Zed features are built using pattern matching over syntax trees with Tree-sitter queries. As mentioned above, every language that is defined in an extension must specify the name of a Tree-sitter grammar that is used for parsing. These grammars are then registered separately in extensions' extension.toml file, like this:

[grammars.gleam]
repository = "https://github.com/gleam-lang/tree-sitter-gleam"
rev = "58b7cac8fc14c92b0677c542610d8738c373fa81"

The repository field must specify a repository where the Tree-sitter grammar should be loaded from, and the rev field must contain a Git revision to use, such as the SHA of a Git commit. An extension can provide multiple grammars by referencing multiple tree-sitter repositories.

Tree-sitter Queries

Zed uses the syntax tree produced by the Tree-sitter query language to implement several features:

• Syntax highlighting
• Bracket matching
• Code outline/structure
• Auto-indentation
• Code injections
• Syntax overrides
• Text redactions
• Runnable code detection
• Selecting classes, functions, etc.

The following sections elaborate on how Tree-sitter queries enable these features in Zed, using JSON syntax as a guiding example.

Syntax highlighting

In Tree-sitter, the highlights.scm file defines syntax highlighting rules for a particular syntax.

Here's an example from a highlights.scm for JSON:

(string) @string

(pair
  key: (string) @property.json_key)

(number) @number

This query marks strings, object keys, and numbers for highlighting. The following is a comprehensive list of captures supported by themes:

Bracket matching

The brackets.scm file defines matching brackets.

Here's an example from a brackets.scm file for JSON:

("[" @open "]" @close)
("{" @open "}" @close)
("\"" @open "\"" @close)

This query identifies opening and closing brackets, braces, and quotation marks.

Code outline/structure

The outline.scm file defines the structure for the code outline.

Here's an example from an outline.scm file for JSON:

(pair
  key: (string (string_content) @name)) @item

This query captures object keys for the outline structure.

Auto-indentation

The indents.scm file defines indentation rules.

Here's an example from an indents.scm file for JSON:

(array "]" @end) @indent
(object "}" @end) @indent

This query marks the end of arrays and objects for indentation purposes.

Code injections

The injections.scm file defines rules for embedding one language within another, such as code blocks in Markdown or SQL queries in Python strings.

Here's an example from an injections.scm file for Markdown:

(fenced_code_block
  (info_string
    (language) @language)
  (code_fence_content) @content)

((inline) @content
 (#set! "language" "markdown-inline"))

This query identifies fenced code blocks, capturing the language specified in the info string and the content within the block. It also captures inline content and sets its language to "markdown-inline".

Note that we couldn't use JSON as an example here because it doesn't support language injections.

Syntax overrides

The overrides.scm file defines syntactic scopes that can be used to override certain editor settings within specific language constructs.

For example, there is a language-specific setting called word_characters that controls which non-alphabetic characters are considered part of a word, for filtering autocomplete suggestions. In JavaScript, "$" and "#" are considered word characters. But when your cursor is within a string in JavaScript, "-" is also considered a word character. To achieve this, the JavaScript overrides.scm file contains the following pattern:

[
  (string)
  (template_string)
] @string

And the JavaScript config.toml contains this setting:

word_characters = ["#", "$"]

[overrides.string]
word_characters = ["-"]

You can also disable certain auto-closing brackets in a specific scope. For example, to prevent auto-closing ' within strings, you could put the following in the JavaScript config.toml:

brackets = [
  { start = "'", end = "'", close = true, newline = false, not_in = ["string"] },
  # other pairs...
]

Range inclusivity

By default, the ranges defined in overrides.scm are exclusive. So in the case above, if you cursor was outside the quotation marks delimiting the string, the string scope would not take effect. Sometimes, you may want to make the range inclusive. You can do this by adding the .inclusive suffix to the capture name in the query.

For example, in JavaScript, we also disable auto-closing of single quotes within comments. And the comment scope must extend all the way to the newline after a line comment. To achieve this, the JavaScript overrides.scm contains the following pattern:

(comment) @comment.inclusive

Text objects

The textobjects.scm file defines rules for navigating by text objects. This was added in Zed v0.165 and is currently used only in Vim mode.

Vim provides two levels of granularity for navigating around files. Section-by-section with [] etc., and method-by-method with ]m etc. Even languages that don't support functions and classes can work well by defining similar concepts. For example CSS defines a rule-set as a method, and a media-query as a class.

For languages with closures, these typically should not count as functions in Zed. This is best-effort however, as languages like Javascript do not syntactically differentiate syntactically between closures and top-level function declarations.

For languages with declarations like C, provide queries that match @class.around or @function.around. The if and ic text objects will default to these if there is no inside.

If you are not sure what to put in textobjects.scm, both nvim-treesitter-textobjects, and the Helix editor have queries for many languages. You can refer to the Zed built-in languages to see how to adapt these.

For example:

; include only the content of the method in the function
(method_definition
    body: (_
        "{"
        (_)* @function.inside
        "}")) @function.around

; match function.around for declarations with no body
(function_signature_item) @function.around

; join all adjacent comments into one
(comment)+ @comment.around

Text redactions

The redactions.scm file defines text redaction rules. When collaborating and sharing your screen, it makes sure that certain syntax nodes are rendered in a redacted mode to avoid them from leaking.

Here's an example from a redactions.scm file for JSON:

(pair value: (number) @redact)
(pair value: (string) @redact)
(array (number) @redact)
(array (string) @redact)

This query marks number and string values in key-value pairs and arrays for redaction.

Runnable code detection

The runnables.scm file defines rules for detecting runnable code.

Here's an example from an runnables.scm file for JSON:

(
    (document
        (object
            (pair
                key: (string
                    (string_content) @_name
                    (#eq? @_name "scripts")
                )
                value: (object
                    (pair
                        key: (string (string_content) @run @script)
                    )
                )
            )
        )
    )
    (#set! tag package-script)
    (#set! tag composer-script)
)

This query detects runnable scripts in package.json and composer.json files.

The @run capture specifies where the run button should appear in the editor. Other captures, except those prefixed with an underscore, are exposed as environment variables with a prefix of ZED_CUSTOM_$(capture_name) when running the code.

Language Servers

Zed uses the Language Server Protocol to provide advanced language support.

An extension may provide any number of language servers. To provide a language server from your extension, add an entry to your extension.toml with the name of your language server and the language(s) it applies to:

[language_servers.my-language]
name = "My Language LSP"
languages = ["My Language"]

Then, in the Rust code for your extension, implement the language_server_command method on your extension:

#![allow(unused)]
fn main() {
impl zed::Extension for MyExtension {
    fn language_server_command(
        &mut self,
        language_server_id: &LanguageServerId,
        worktree: &zed::Worktree,
    ) -> Result<zed::Command> {
        Ok(zed::Command {
            command: get_path_to_language_server_executable()?,
            args: get_args_for_language_server()?,
            env: get_env_for_language_server()?,
        })
    }
}
}

You can customize the handling of the language server using several optional methods in the Extension trait. For example, you can control how completions are styled using the label_for_completion method. For a complete list of methods, see the API docs for the Zed extension API.