Chisel

Extending Chisel

Writing your own language on the chisel foundation

This page sketches how to build a small language using chisel primitives. The Notch parser itself is the canonical worked example; this is an outline.

1. Configure a tokenizer

Compose token types into a Tokenizer:

Tokenizer T = new Tokenizer()
    .withTokenType(WHITESPACE)
    .withTokenType(BOOL)
    .withTokenType(MY_KEYWORD)
    .withTokenType(IDENT)
    .withTokenType(INT)
    .withTokenType(STR)
    .withTokenTypes(TokenTypePunct.common());

TokenTypePunct.common() returns the common punctuation tokens. You can add your own TokenType implementations by subclassing or composing.

2. Write a parser

Extend BasicParser:

public class MyParser extends BasicParser {
    public MyParser(String fileId, String src) {
        this(T.tokenize(fileId, src));
    }

    public MyExpression parseExpression() {
        // recursive descent here
    }
}

The base class provides:

location(), take(), peek() - position-tracking primitives.
ignoreTypes(TokenType...) - skip tokens of the listed types when scanning (typical use: whitespace and comments).
temporaryEndTokens(TokenType[], () -> T) - run a sub-parser that treats the listed tokens as end-of-stream. Useful when parsing a list element that should stop at , or ] without consuming them.

3. Add coercions

If your language uses JVM types, register coercions so type mismatches at method-call sites get bridged automatically. See src/main/java/edu/montana/notch/types/coercions/Coercion.java for the abstract base.

4. Render errors with spans

Throw ParseException from your parser. Catch at the entry point and use the Spanned interface to render a file:line:column caret message.

1. Configure a tokenizer

2. Write a parser

3. Add coercions

4. Render errors with spans

See also