Skip to content
Docs
/
Plugin
/
ECMAScript
/
Cheatsheet

Plugin cheatsheet

💡

This page describes the known hard points for implementing plugins for ecmascript.

Understanding types

JsWord

String allocates, and 'text'-s of source code has a special trait. Those are lots of duplicates. Obviously, if your variable is named foo, you need to use foo multiple times. So SWC interns the string to reduce the number of allocations.

JsWord is a string type that is interned. You can create a JsWord from &str, or from a String. Use .into() to convert to JsWord.

Ident, Id, Mark, SyntaxContext

SWC uses a special system for managing variables. See the rustdoc for Ident for details.

Common issues

Getting AST representation of input

SWC Playground supports getting AST from the input code.

Variable management of SWC

Error reporting

See rustdoc for `swc_common::erorrs::Handler.

Comparing JsWord with &str

If you don't know what JsWord is, see the rustdoc for swc_atoms.

You can create &str by doing &val where val is a variable of type JsWord.

Matching Box<T>

You will need to use match to match on various nodes, including Box<T>. For performance reason, all expressions are stored in a boxed form. (Box<Expr>)

SWC stores callee of call expressions as a Callee enum, and it has Box<Expr>.

use swc_plugin::*;

struct MatchExample;

impl VisitMut for MatchExample {
    fn visit_mut_callee(&mut self, callee: &mut Callee) {
        callee.visit_mut_children_with(self);

        if let Callee::Expr(expr) = callee {
            // expr is `Box<Expr>`
            if let Expr::Ident(i) = &mut **expr {
                i.sym = "foo".into();
            }
        }
    }
}

Tips

Apply resolver while testing

SWC applies plugin after applying resolver, so it's better to test your transform with it. As written in the rustdoc for the resolver, you have to use correct SyntaxContext if you need to reference global variable (e.g. __dirname, require) or top-level bindings written by the user.

fn tr() -> impl Fold {
    chain!(
        resolver(Mark::new(), Mark::new(), false),
        // Most of transform does not care about globals so it does not need `SyntaxContext`
        your_transform()
    )
}

test!(
    Syntax::default(),
    |_| tr(),
    basic,
    // input
    "(function a ([a]) { a });",
    // output
    "(function a([_a]) { _a; });"
);

Make your handlers stateless

Let's say we are going to handle all array expressions in a function expression. You can add a flag to the visitor to check if we are in a function expression. You will be tempted to do


struct Transform {
    in_fn_expr: bool
}

impl VisitMut for Transform {
    noop_visit_mut_type!();

    fn visit_mut_fn_expr(&mut self, n: &mut FnExpr) {
        self.in_fn_expr = true;
        n.visit_mut_children_with(self);
        self.in_fn_expr = false;
    }

    fn visit_mut_array_lit(&mut self, n: &mut ArrayLit) {
        if self.in_fn_expr {
            // Do something
        }
    }
}

but this cannot handle


const foo = function () {
    const arr = [1, 2, 3];

    const bar = function () {};

    const arr2 = [2, 4, 6];
}

After visiting bar, in_fn_expr is false. You have to do


struct Transform {
    in_fn_expr: bool
}

impl VisitMut for Transform {
    noop_visit_mut_type!();

    fn visit_mut_fn_expr(&mut self, n: &mut FnExpr) {
        let old_in_fn_expr = self.in_fn_expr;
        self.in_fn_expr = true;

        n.visit_mut_children_with(self);

        self.in_fn_expr = old_in_fn_expr;
    }

    fn visit_mut_array_lit(&mut self, n: &mut ArrayLit) {
        if self.in_fn_expr {
            // Do something
        }
    }
}

instead.

Ownership model (of rust)

This section is not about swc itself. But this is described at here because it's the cause of almost all trickyness of APIs.

In rust, only one variable can own a data, and there's at most one mutable reference to it. Also, you need to own the value or have a mutable reference to it if you want to modify the data.

But there's at most one owner/mutable reference, so it means if you have a mutable reference to a value, other code cannot modify the value. Every update operation should performed by the code which owns the value or has a mutable reference to it. So, some of babel APIs like node.delete is super tricky to implement. As your code has ownership or mutable refernce to some part of AST, SWC cannot modify the AST.

Tricky operations

Deleting node

You can delete a node in two step.

Let's say, we want to drop the variable named bar in the code below.

var foo = 1;
var bar = 1;

There are two ways to do this.

Mark & Delete

The first way is to mark it as invalid and delete it later. This is typically more convinient.


use swc_plugin::*;

impl VisitMut for Remover {
    fn visit_mut_var_declarator(&mut self, v: &mut VarDeclarator) {
        // This is not required in this example, but you typically need this.
        v.visit_mut_children_with(self);


        // v.name is `Pat`.
        // See https://rustdoc.swc.rs/swc_ecma_ast/enum.Pat.html
        match v.name {
            // If we want to delete the node, we should return false.
            //
            // Note the `&*` before i.sym.
            // The type of symbol is `JsWord`, which is an interned string.
            Pat::Ident(i) => {
                if &*i.sym == "bar" {
                    // Take::take() is a helper function, which stores invalid value in the node.
                    // For Pat, it's `Pat::Invalid`.
                    v.name.take();
                }
            }
            _ => {
                // Noop if we don't want to delete the node.
            }
        }
    }

    fn visit_mut_var_declarators(&mut self, vars: &mut Vec<VarDeclarator>) {
        vars.visit_mut_children_with(self);

        vars.retain(|node| {
            // We want to remove the node, so we should return false.
            if node.name.is_invalid() {
                return false
            }

            // Return true if we want to keep the node.
            true
        });
    }

    fn visit_mut_stmt(&mut self, s: &mut Stmt) {
        s.visit_mut_children_with(self);

        match s {
            Stmt::Decl(Decl::Var(var)) => {
                if var.decls.is_empty() {
                    // Variable declaration without declarator is invalid.
                    //
                    // After this, `s` becomes `Stmt::Empty`.
                    s.take();
                }
            }
            _ => {}
        }
    }

    fn visit_mut_stmts(&mut self, stmts: &mut Vec<Stmt>) {
        stmts.visit_mut_children_with(self);

        // We remove `Stmt::Empty` from the statement list.
        // This is optional, but it's required if you don't want extra `;` in output.
        stmts.retain(|s| {
            // We use `matches` macro as this match is trivial.
            !matches!(s, Stmt::Empty(..))
        });
    }

    fn visit_mut_module_items(&mut self, stmts: &mut Vec<ModuleItem>) {
        stmts.visit_mut_children_with(self);

        // This is also required, because top-level statements are stored in `Vec<ModuleItem>`.
        stmts.retain(|s| {
            // We use `matches` macro as this match is trivial.
            !matches!(s, ModuleItem::Stmt(Stmt::Empty(..)))
        });
    }
}

Delete from the parent handler

Another way to delete the node is deleting it from the parent handler. This can be useful if you want to delete the node only if the parent node is specific type.

e.g. You don't want to touch the variables in for loops while deleting free variable statements.

use swc_plugin::*;

struct Remover;

impl VisitMut for Remover {
    fn visit_mut_stmt(&mut self, s: &mut Stmt) {
        // This is not required in this example, but just to show that you typically need this.
        s.visit_mut_children_with(self);

        match s {
            Stmt::Decl(Decl::Var(var)) => {
                if var.decls.len() == 1 {
                    match var.decls[0].name {
                        Pat::Ident(i) => {
                            if &*i.sym == "bar" {
                                s.take();
                            }
                        }
                    }
                }
            }
            _ => {}
        }
    }


    fn visit_mut_stmts(&mut self, stmts: &mut Vec<Stmt>) {
        stmts.visit_mut_children_with(self);

        // We do same thing here.
        stmts.retain(|s| {
            !matches!(s, Stmt::Empty(..))
        });
    }

    fn visit_mut_module_items(&mut self, stmts: &mut Vec<ModuleItem>) {
        stmts.visit_mut_children_with(self);

        // We do same thing here.
        stmts.retain(|s| {
            !matches!(s, ModuleItem::Stmt(Stmt::Empty(..)))
        });
    }
}

Referencing parent node from handler of child node

This includes usage of paths and scope.

Caching some information about an AST node

You have two way to use informantion from a parent node. For first, you can precompute information from the parent node handler. Alternatively, you can clone the parent node and use it in the child node handler.

Alternatives for babel APIs

generateUidIdentifier

This returns a unique identifier with a monotonically increasing integer suffix. swc does not provide API to do this, because there's a very easy way to do this. You can store an integer field in transformer type and use it while calling quote_ident! or private_ident!.


struct Example {
    // You don't need to share counter.
    cnt: usize
}

impl Example {
    /// For properties, it's okay to use `quote_ident`.
    pub fn next_property_id(&mut self) -> Ident {
        self.cnt += 1;
        quote_ident!(format!("$_css_{}", self.cnt))
    }

    /// If you want to create a safe variable, you should use `private_ident`
    pub fn next_variable_id(&mut self) -> Ident {
        self.cnt += 1;
        private_ident!(format!("$_css_{}", self.cnt))
    }
}

path.find

Upward traversal is not supported by swc. It's because upward traversal requires storing information about parent at children nodes, which requires using types like Arc or Mutex in rust.

Instead of traversing upward, you should make it top-down. For example, if you want to infer name of a jsx component from variable assignments or assignments, you can store name of component while visiting VarDecl and/or AssignExpr and use it from the component handler.

state.file.get/state.file.set

You can simply store the value in the transform struct as an instance of transform struct only process one file.

Last updated on April 30, 2022