Sucrase

Try it out

Sucrase is an alternative to Babel that allows super-fast development builds. Instead of compiling a large range of JS features down to ES5, Sucrase assumes that you're targeting a modern JS runtime (e.g. Node.js 8 or latest Chrome) and focuses on compiling non-standard language extensions: JSX, TypeScript, and Flow. Because of this smaller scope, Sucrase can get away with an architecture that is much more performant but less extensible and maintainable. Sucrase's parser is forked from Babel's parser (so Sucrase is indebted to Babel and wouldn't be possible without it) and trims it down to focus on a small subset of what Babel solves. If it fits your use case, hopefully Sucrase can speed up your development experience!

Current state: The project is in active development. It is about 20x faster than Babel and about 8x faster than TypeScript, and it has been tested on hundreds of thousands of lines of code. Still, you may find correctness issues when running on a large codebase. Feel free to file issues!

Sucrase can build the following codebases with all tests passing:

Sucrase itself (6K lines of code excluding Babel parser fork, typescript, imports).
The Benchling frontend codebase (500K lines of code, JSX, typescript, imports).
Babel (63K lines of code, flow, imports).
React (86K lines of code, JSX, flow, imports).
TSLint (20K lines of code, typescript, imports).
Apollo client (34K lines of code, typescript, imports)
decaffeinate and its sub-projects decaffeinate-parser and coffee-lex (38K lines of code, typescript, imports).

Transforms

The main configuration option in Sucrase is an array of transform names. There are four main transforms that you may want to enable:

jsx: Transforms JSX syntax to React.createElement, e.g. <div a={b} /> becomes React.createElement('div', {a: b}). Behaves like Babel 7's babel-preset-react, including adding createReactClass display names and JSX context information.
typescript: Compiles TypeScript code to JavaScript, removing type annotations and handling features like enums. Does not check types.
flow: Removes Flow type annotations. Does not check types.
imports: Transforms ES Modules (import/export) to CommonJS (require/module.exports) using the same approach as Babel 6 and TypeScript with --esModuleInterop. Also includes dynamic import.

The following proposed JS features are built-in and always transformed:

Class fields: class C { x = 1; }. This includes static fields but not the #x private field syntax.
Export namespace syntax: export * as a from 'a';
Numeric separators: const n = 1_234;
Optional catch binding: try { doThing(); } catch { }.

JSX Options

Like Babel, Sucrase compiles JSX to React functions by default, but can be configured for any JSX use case.

jsxPragma: Element creation function, defaults to React.createElement.
jsxFragmentPragma: Fragment component, defaults to React.Fragment.

Legacy CommonJS interop

Two legacy modes can be used with the import tranform:

enableLegacyTypeScriptModuleInterop: Use the default TypeScript approach to CommonJS interop instead of assuming that TypeScript's --esModuleInterop flag is enabled. For example, if a CJS module exports a function, legacy TypeScript interop requires you to write import * as add from './add';, while Babel, Webpack, Node.js, and TypeScript with --esModuleInterop require you to write import add from './add';. As mentioned in the docs, the TypeScript team recommends you always use --esModuleInterop.
enableLegacyBabel5ModuleInterop: Use the Babel 5 approach to CommonJS interop, so that you can run require('./MyModule') instead of require('./MyModule').default. Analogous to babel-plugin-add-module-exports.

Usage

Installation:

yarn add --dev sucrase  # Or npm install --save-dev sucrase

Run on a directory:

sucrase ./srcDir -d ./outDir --transforms typescript,imports

Compile on-the-fly via a require hook with some reasonable defaults:

// Register just one extension.
import "sucrase/register/ts";
// Or register all at once.
import "sucrase/register";

Compile on-the-fly via a drop-in replacement for node:

sucrase-node index.ts

Call from JS directly:

import {transform} from "sucrase";
const compiledCode = transform(code, {transforms: ["typescript", "imports"]}).code;

There are also integrations for Webpack, Gulp, Jest and Rollup.

What Sucrase is not

Sucrase is intended to be useful for the most common cases, but it does not aim to have nearly the scope and versatility of Babel. Some specific examples:

Sucrase does not check your code for errors. Sucrase's contract is that if you give it valid code, it will produce valid JS code. If you give it invalid code, it might produce invalid code, it might produce valid code, or it might give an error. Always use Sucrase with a linter or typechecker, which is more suited for error-checking.
Sucrase is not pluginizable. With the current architecture, transforms need to be explicitly written to cooperate with each other, so each additional transform takes significant extra work.
Sucrase is not good for prototyping language extensions and upcoming language features. Its faster architecture makes new transforms more difficult to write and more fragile.
Sucrase will never produce code for old browsers like IE. Compiling code down to ES5 is much more complicated than any transformations that Sucrase needs to do.
Sucrase is hesitant to implement upcoming JS features, although some of them make sense to implement for pragmatic reasons. Its main focus is on language extensions (JSX, TypeScript, Flow) that will never be supported by JS runtimes.
Like Babel, Sucrase is not a typechecker, and must process each file in isolation. For example, TypeScript const enums are treated as regular enums rather than inlining across files.
You should think carefully before using Sucrase in production. Sucrase is mostly beneficial in development, and in many cases, Babel or tsc will be more suitable for production builds.

Motivation

As JavaScript implementations mature, it becomes more and more reasonable to disable Babel transforms, especially in development when you know that you're targeting a modern runtime. You might hope that you could simplify and speed up the build step by eventually disabling Babel entirely, but this isn't possible if you're using a non-standard language extension like JSX, TypeScript, or Flow. Unfortunately, disabling most transforms in Babel doesn't speed it up as much as you might expect. To understand, let's take a look at how Babel works:

Tokenize the input source code into a token stream.
Parse the token stream into an AST.
Walk the AST to compute the scope information for each variable.
Apply all transform plugins in a single traversal, resulting in a new AST.
Print the resulting AST.

Only step 4 gets faster when disabling plugins, so there's always a fixed cost to running Babel regardless of how many transforms are enabled.

Sucrase bypasses most of these steps, and works like this:

Tokenize the input source code into a token stream using a trimmed-down fork of the Babel parser. This fork does not produce a full AST, but still produces meaningful token metadata specifically designed for the later transforms.
Scan through the tokens, computing preliminary information like all imported/exported names.
Run the transform by doing a pass through the tokens and performing a number of careful find-and-replace operations, like replacing <Foo with React.createElement(Foo.

Because Sucrase works on a lower level and uses a custom parser for its use case, it is much faster than Babel.

Performance

Currently, Sucrase runs about 20x faster than Babel (even when Babel only runs the relevant transforms) and 8x faster than TypeScript. Here's the output of one run of npm run benchmark:

Simulating transpilation of 100,000 lines of code:
Sucrase: 469.672ms
TypeScript: 3782.414ms
Babel: 9591.515ms

Project vision and future work

Performance improvements

Rewrite the code to run in WebAssembly, either by changing it to be valid AssemblyScript or by rewriting it in Rust.
Explore the idea of a JIT to optimize the various token patterns that need to be matched as part of code transformation.

New features

Implement more integrations, like a Browserify plugin.
Emit proper source maps. (The line numbers already match up, but this would help with debuggers and other tools.)
Rethink configuration and try to simplify it as much as possible, and allow loading Babel/TypeScript configurations.
Explore the idea of a tool that patches a Babel/TypeScript installation to use Sucrase instead, to make it even easier to try Sucrase on an existing codebase.
Explore the idea of extending this approach to other tools, e.g. module bundlers.

Correctness and stability

Add more open source projects to the suite of projects that are tested automatically.
Set up a test suite that runs the compiled code and ensures that it is correct.
Add integrity checks to compare intermediate Sucrase results (like tokens and the role of each identifier and pair of curly braces) with the equivalent information from Babel.
Fix some known correctness loose ends, like import hoisting and fully replicating the small differences between Babel and the TypeScript compiler.

License and attribution

Sucrase is MIT-licensed. A large part of Sucrase is based on a fork of the Babel parser, which is also MIT-licensed.

Why the name?

Sucrase is an enzyme that processes sugar. Get it?

sudowork / sucrase