Schism

Schism is an experimental self-hosting compiler from a subset of R6RS Scheme to WebAssembly.

This is not an officially supported Google product.

Development so far has focused on features necessary for self-hosting. The compiler itself is written in, and compiles, a very small subset of Scheme. Now that self-hosting has been achieved, development can shift towards supporting a more complete subset of Scheme.

Besides just being fun, one of the goals of this project is to explore different ways to use WebAssembly. This includes implementing garbage collection, possibly using the WebAssembly GC proposal, dynamic linking and code generation, etc.

Chez Scheme was used to bootstrap, but development no longer relies on an existing Scheme implementation and can instead run purely based on snapshots checked into the repository.

As mentioned, the goal has been to prioritize features needed for self hosting. Here are some of the current restrictions:

• No lambda, no closures.
• Functions may only be created with define, and particularly only the (define (foo args ...) body) form.
• No use of syntax-case, syntax-rules, or define-syntax.
• Only one file to start with, so we don't have to figure out how to link multiple libraries.
• Only use define to create functions, not global variables or objects.
• Use a small amount of syntax, because we won't have a proper macro expander at first. There is a pass to expand some of the simpler and more useful macros.
• Restrict data types and operations on those. For now, we can use:
  • Numbers
  • Symbols
  • Pairs
  • Strings

As more features are supported by the compiler, we will remove these restrictions.

Current Status and Next Steps

The compiler is self-hosting! Now the goal is to make a more complete language. Some of the big missing features are:

• Closures (i.e. lambda)
• Variable length argument lists
• Proper tail recursion
• Garbage collection
• Macros

Many of these features are interconnected and can be designed together to make it easier to implement them.

Testing

We currently use a very simple testing protocol. The test/ directory includes a number of Scheme libraries, each of which export a function called do-test. This function can do whatever it wants, but it must return a value other than #f to pass.

To run all the tests, do ./run-tests.sh.

The Playground

This repository includes a very simple playground.html, which gives a lightweight way to play around with the compiler. The best way to use it is to start up a web server (python -m SimpleHTTPServer should work) and point your browser at the page. Be warned, there is almost no error checking right now, so strange things can happen.

Development

The compiler code all lives in schism/compiler.ss. There is a small JavaScript runtime in rt/rt.mjs. The goal is to keep as much code as possible in Scheme, but the runtime is needed to interact with the rest of the world. In the future, the runtime should also handle dynamic module loading.

We have a staged build system. Stage0 is the compiler snapshot, stored in schism-stage0.wasm. Stage1 is generated by compiling schism/compiler.ss with the Stage0 compiler, Stage2 by compiling with the Stage1 compiler, and Stage3 by compiling with the Stage2 compiler. Stage3 should be equal to the Stage2 compiler, and currently we only generate it to verify this is the case.

To add a new feature, the usual flow is to start by adding a small test that uses it. This test will probably fail at least the Stage0 compiler. Once the feature is implemented, then the Stage1 and Stage2 compilers should pass the test. Note that you cannot use the new feature in the compiler until it works in stage2. Once this happens, you should make a new snapshot using mksnapshot.sh.