Code Splitting in GWT

A pain point for Google web applications is the time required to download the JavaScript code that supports them. Web pages do not become interactive until the supporting JavaScript code has downloaded and been parsed. Moreover, the amount of JavaScript code trends upward over the lifespan of an app, because developers add more features and those features require more code to support.

To address this, I extended the Google Web Toolkit to support code splitting. Some notable aspects of the project:

• It splits the code without respecting module boundaries, so that developers can start using it without having to reorganize their entire application according to the deployment strategy. This feature was crucial for adoption, because developers never start code splitting until their app has grown large enough to require it. Serious development tools should never force developers to rewrite from scratch.
• I developed debugging debugging support based on the GWT Compile Report. Any whole-program optimization has the downside that changes by a developer have non-local effects. Whenever the whole-program optimization cannot be avoided, it is important to provide tool support for tracing the effects of those changes.
• I developed and documented techniques for evolving a code base to minimize the non-local effects on the code splitter. In particular, I developed the Async Provider pattern to help control where lazy-loaded code is accessed synchronously.

The GWT code splitter is used by Google Adwords and Google Wallet, among many other web applications.

See the GWT code splitter documentation for more information.

Lex Spoon