Handy Incremental Parser

Problem

Most programming language compilers usually process source code files at once. Compilation time using this approach is proportional to the overall source code size in best case. In practice, the time may grow to seconds, and even minutes, as the code base grows.

This approach is not suitable for writing language support plugins for code editors and IDEs for two reasons:

  • The editor must always be in touch with source code's syntax and semantics.
  • Small and frequent changes that end users make to the source code should be indexed immediately, without any significant time delays.

Another problem is that PL compilers and appropriate IDE plugins are usually developed separately for each code editor.

The solution to these problems might be designing compiler's front-end to work as a compilation server in an incremental fashion:

  • It should provide external code editors with an API to obtain metadata on its internal compilation state. Such as the source code's syntax structure, semantic references and errors in the code. So the editor can utilise this information to provide end user with a set of conventional IDE tools such as code refactoring, jump-to-definition, variable renaming, syntax and error highlighting.
  • The compilation server should work continuously by caching all the metadata including AST and semantic references. This information can be reused to resume the compilation process when the source code is changed. Therefore, compilation performance will be relative not to the overall codebase size, but to the size of the changes that the end user makes to the code.

In other words, the idea is to move the code editor's conventional responsibilities to the compiler's side. So the compiler becomes universal reusable component for a set of code editors and IDEs. Whereas the editor works as a thin client that depends on compilation server. The advantage of this approach is simplifying development of PL plugins for code editors and IDEs, and unification of their support.

The heart of the compiler's front-end is a source code parser. And the parser of the incremental compiler should work in incremental fashion as well.

Here is an issue: there are a lot of tools to build ordinary parsers based on predefined programming language grammar. These include generators such as YACC and ANTLR or combinators such as Parsec or Parboiled. But too few of them provide a way to build incremental parsers. Incremental parsing feature support usually has experimental status in these projects. They are interesting from research and academic point of view, but usually are not ready for use in real products.

The goal of my project is to eliminate the lack of such instruments, by providing an easy to use library for constructing incremental parsers with a number of handy features that compiler and PL support plugin developers usually need.

Papa Carlo

I am pleased to introduce you to my project, Papa Carlo. This is a Scala library to construct recursive descent incremental parsers based on PEG grammar. The project is published on GitHub under the APL2 license.

The library provides the following features:

  • The parser's grammar is defined in the Scala code using the library's API. No explicit generation steps.
  • The API represents Parsing Expression Grammar operators. PEG is probably the most popular and easy to understand PL grammar nowadays.
  • Resulting parser builds and incrementally updates Abstract Syntax Tree out of the box. And there is no intermediate Parsing Tree build step.
  • Efficient error recovery. Resulting parser can build or update AST even from the source code that contains syntax errors.
  • Expression grammar with unary and binary precedence operators can be easily defined based on the Pratt algorithm using dedicated functions like infix() or group().

The simple example of a full-featured expression parser can be defined with Papa Carlo in one short Scala file:

object Calculator {
  private def tokenizer = {
    val tokenizer = new Tokenizer()

    import tokenizer._
    import Matcher._

    tokenCategory("whitespace", oneOrMore(anyOf(" \t\f\n"))).skip

    tokenCategory("number", choice(chunk("0"), sequence(rangeOf('1', '9'),
      zeroOrMore(rangeOf('0', '9')))))

    terminals("(", ")", "!", "+", "-", "*", "/")

    tokenizer
  }

  def lexer = new Lexer(tokenizer, new Contextualizer)

  def syntax(lexer: Lexer) = new {
    val syntax = new Syntax(lexer)

    import syntax._
    import Rule._
    import Expressions._

    mainRule("expression") {
      val rule =
        expression(branch("operand", recover(number, "operand required")))

      group(rule, "(", ")")
      postfix(rule, "!", 1)
      prefix(rule, "+", 2)
      prefix(rule, "-", 2)
      infix(rule, "*", 3)
      infix(rule, "/", 3, rightAssociativity = true)
      infix(rule, "+", 4)
      infix(rule, "-", 4)

      rule
    }

    val number = rule("number") {capture("value", token("number"))}
  }.syntax
}

The parser will be able to parse precedence expressions like this 1 - -2 / 3 * +(4 / 5) + 6 + 7! + 8, build and update AST, and recover syntax errors.

A more interesting example is a JSON parser, which incrementally parses large input files with over 600 lines of source code. In one of the included tests, these files have been changed and parsed several times:

File Difference (lines) Syntax errors Number of AST nodes Parse time (seconds)
Version 1 - 0 918 0.270
Version 2 1 1 923 0.007
Version 3 2 3 924 0.008

This is where the incremental parsing approach shows its advantage.

Also Papa Carlo includes an API to log and debug various aspects while developing a parser. This technique was used to cover the library with a number of functional tests.