Python-like indentation tokens for ANTLR4

A mostly-readymade solution to INDENT/DEDENT tokens in ANTLR v4. Just plug in the DenterHelper and you'll be good to go! See this blog post for some of the motivations behind this project.

antlr-helper is released under the MIT license, which basically means you can do whatever you want with it. That said, I'd really appreciate hearing from you if you find this project useful! Maybe star the project?

Usage (Java)

maven

<dependency>
  <groupId>com.yuvalshavit</groupId>
  <artifactId>antlr-denter</artifactId>
  <version>1.1</version>
</dependency>

Adding INDENT / DEDENT tokens to your lexer

1. Define INDENT and DEDENT tokens in your grammar
2. In your @lexer::members section, instantiate a DenterHelper whose pullToken method delegates to your lexer's super.nextToken()
3. Override your lexer's super.nextToken method to use DenterHelper::nextToken instead.
4. Modify your NL token to also grab any whitespace that follows (in other words, have it end in ' '*, '\t'* or similar).

DenterHelper is an abstract class, and it also takes three arguments for its constructor: the token types for newline, INDENT and DEDENT. It's probably easiest to instantiate it as an anonymous class. The whole thing should look something like this:

tokens { INDENT, DEDENT }

@lexer::header {
  import com.yuvalshavit.antlr4.DenterHelper;
}

@lexer::members {
  private final DenterHelper denter = new DenterHelper(NL,
                                                       MyCoolParser.INDENT,
                                                       MyCoolParser.DEDENT)
  {
    @Override
    public Token pullToken() {
      return MyCoolLexer.super.nextToken();
    }
  };

  @Override
  public Token nextToken() {
    return denter.nextToken();
  }
}

NL: ('\r'? '\n' ' '*);

There is also a builder available, which is especially useful for Java 8:

tokens { INDENT, DEDENT }

@lexer::header {
  import com.yuvalshavit.antlr4.DenterHelper;
}

@lexer::members {
  private final DenterHelper denter = DenterHelper.builder()
    .nl(NL)
    .indent(MyCoolParser.INDENT)
    .dedent(MyCoolParser.DEDENT)
    .pullToken(MyCoolLexer.super::nextToken);

  @Override
  public Token nextToken() {
    return denter.nextToken();
  }
}

NL: ('\r'? '\n' ' '*);

Usage (Python3)

tokens { INDENT, DEDENT }

@lexer::header{
from DenterHelper import DenterHelper
from MyCoolParser import MyCoolParser
}
@lexer::members {
class MyCoolDenter(DenterHelper):
    def __init__(self, lexer, nl_token, indent_token, dedent_token, ignore_eof):
        super().__init__(nl_token, indent_token, dedent_token, ignore_eof)
        self.lexer: MyCoolLexer = lexer

    def pull_token(self):
        return super(MyCoolLexer, self.lexer).nextToken()

denter = None

def nextToken(self):
    if not self.denter:
        self.denter = self.MyCoolDenter(self, self.NL, MyCoolParser.INDENT, MyCoolParser.DEDENT, ***Should Ignore EOF***)
    return self.denter.next_token()

}

Using the tokens in your parser

Basically, just use them. One bit worth noting is that when the denter injects DEDENT tokens, it'll prefix any string of them with a single NL. A single NL is also inserted before the EOF token if there are no DEDENTs to insert. For instance, given this input:

hello
  world
    universe
dolly

... the tokens would be (roughly):

"hello"
INDENT
"world"
INDENT
"universe"
NL
DEDENT
DEDENT
"dolly"
NL
<eof>

This is done so that simple expressions can be terminated by the NL token without worrying about surrounding context (an impending dedent or EOF). In this case, universe and dolly represent simple expressions, and you can imagine that the grammar would contain something like statement: expr NL | helloBlock;. Easy peasy!

"Half-DEDENTs"

What happens when you dedent to an indentation level that was never established?

someStatement()
if foo():
    if bar():
      fooAndBar()
  bogusLine()

Notice that bogusLine() doesn't match with any indentation level: it's more indented than if foo() but less than its first statement, if bar().

This is a buggy program in python. If you to run such a program, you'll get:

IndentationError: unindent does not match any outer indentation level

The DenterHelper processor handles this by inserting two tokens: a DEDENT followed immediately by an INDENT (the total sequence here would actually be two DEDENTs followed by an INDENT, since bogusLine() is twice-dedented from fooAndBar()). The rationale is that the line has dedened to its parent, and then indented. It's consistent with the indentation tokens for something like:

someStatement()
  bogusLine()

If your indentation scheme is anything like python's, chances are you want this to be a compilation error. The good news is that it will be, as long as your parser doesn't allow "spontaneous" indents. That is, if the example just before this paragraph fails, then so will the half-dedent example above. In both cases, the parser rules will bork on an unexpected INDENT token.

Repo layout

• Java/core: The real thing. This is what you're interested in. Maven artifact antlr-denter.
• Java/examples: Contains a real-life example of a language that uses DenterHelper, so you can see a full solution, including the pom, how to set up the parser (which is nothing extra relative to usual antlr stuff) and how to define a language that uses these INDENT/DEDENT tokens. The language itself is pretty basic, but it should get the point across. Maven artifact antlr-denter-example-examples.
• Python3: The python3 implementation

The maven run is as simple as mvn install (or your favorite goal).

Comments? Suggestions? Bugs?

Don't be shy about opening an issue!