wycats / looper

Static analysis tools for ES6

Looper is a project that provides tools for analyzing ES6 code in an effort to make transpiled output smaller and more efficient.

Scope Analyzer

The first part of Looper is the Scope Analyzer. The goal of the scope anlayzer is to preprocess an AST and tag nodes with additional information about the scopes they represent.

For example, consider this code:

{
  let { x, y } = a.b;

  if (x) {
    _ref = x._ref;
  }
}

In this case, there are three scopes:

• The top-level scope, which contains one Block, no bindings, one downstream binding mutation (_ref), and no downstream usages of direct eval.
• The Block, which contains two bindings (x and y), one upstream binding reference (a), one downstream binding reference (x), one downstream binding mutation (ref) and no downstream usages of direct eval.
• The Block inside the IfStatement, which contains one upstream binding mutation (_ref), one upstream binding reference (x) and no downstream usages of direct eval.

This information can be used when transpiling the destructuring assignment. In particular, this destructuring assignment needs to assign a.b to a temporary variable.

Because we know that x and _ref are used downstream, we should not use those names for our temporary variable.

We could transpile the destructuring assignment to something like this reliably (the let would be transpiled separately, using similar information to keep it reliable and small):

{
  // since we know that t is never used downstream, we can be sure that
  let t = a.b, x = t.x, y = t.y;

  if (x) {
    _ref = x._ref;
  }
}

Let's look at a similar example with just let:

function log(string) {
  if (string) {
    let string = `${string}!`;
    console.log(string);
  } else {
    let string = "No string provided to log";
    console.warn(string);
  }
}

In this case, the let variables inside the blocks can be reliably converted into vars since they are not used in another child scope of the containing function scope.

In this case, when transpiling the let, we would go up to the function scope, and check to see whether there are direct child scopes (other than the conditional's scopes) that reference or mutate the string binding. If not, we are safe to just transpile the let directly into a var.

By tracking all downstream variable references and mutations in a single pass, we can avoid having to constantly do arbitrary-depth tree-walks during the transpilation itself.

Scope Annotations

(NOTE: This section is slightly ahead of the implementation; that will be corrected soon)

Every node that represents a new scope is annotated with a looper:scope property that is a Scope object.

The scope has a number of useful properties on it:

• node: the node the scope represents
• parent: the parent scope of this scope
• children: a list of child scopes of this scope
• es6scope: if this is true, the scope represents a let scope. If false, it represents a function scope.
• upstream: An object with two properties, get and set.
  • get is a Set of bindings that this scope references that are not declared on this scope.
  • set is a Set of bindings that this scope mutates that are not declarared on this scope.
• downstream: An object with two properties, get and set that represents the union of all upstream properties on downstream scopes.
• unbound: An object with two properties, get and set.
  • get is a list of all binding references that escape the current scope.
  • set is a list of all binding mutations that escape the current scope.
• bindings: a Set of bindings declared on this scope. Note that var bindings are always considered declared on their function scope. If initialized, they are considered unbound mutations.

Note that it's possible to get a list of all "global" references and mutations from a program by unioning the get and set properties of the Program Scope's unbound property.

It also has some useful methods.

• parentFunctionScope: the closest parent scope that is a function scope, rather than an ES6 Block scope.