Categories: parsing

Parsing

Introduction

I have been interested in various aspects of parsing since 2009. My research is particularly focused on parsing general context-free grammars, using combinator parsers. I have published several papers, formalized various proofs about parsers in the HOL4 theorem prover, and produced quite a few OCaml implementations to try out various ideas.

P0 is a very simple combinator library, a sort-of minimal basis for combinator parsing.
P1 is a combinator parsing library for all context-free grammars (CFGs). It resembles traditional combinator-parsing libraries (both in interface, and in performance) with the additional feature that it can handle arbitrary CFGs. The performance with highly ambiguous left-recursive grammars is not optimal, but can be surprisingly good. For example, on the grammar E -> E E E | "1" | eps, P1 outperforms the Haskell Happy parser generator. The implementation is based on the verified parsing work (see below).
tjr_simple_earley is my current favourite Earley parser implementation (in OCaml). A Python version is also available from github.
Example grammars are here ; see also the blog post

Verified parsing

This work produced a verified, sound and complete approach to combinator parsing. This introduced the idea of “good parse trees”. The problem with this approach is that it is too inefficient with left-recursive grammars. There is a post describing verified parsing here.

Deprecated

These are projects which I have largely abandoned:

P3 is a combinator parsing library that can parse all CFGs, whilst offering O(n³) performance (which I consider optimal for real-world usage: there are approaches that can do better than O(n³), but the real-world overhead is currently unacceptable, and the implementation complexity is high).
E3 is an Earley parser which is used by P3.
P4 is a combinator parsing library that improves on P3 in 2 main ways:
- The library user has tight control over the parsing, allowing better real-world performance than P3.
- Various extensions are supported that allow to go beyond context-free grammars. For example, indentation-sensitive parsing is possible, based on allowing infinitely-many nonterminals in the grammar.
The downside is that the interface to P4 is more complicated than that for P3.

Older links are:

P3: New versions of P3 are now available from github at https://github.com/tomjridge/p3. At the moment the documentation is on the github wiki for P3 at https://github.com/tomjridge/p3/wiki/P3.

The best documentation for P3 internals is the extended version of the following paper: Tom Ridge. Simple, efficient, sound and complete combinator parsing for all context-free grammars, using an oracle. In Proceedings of SLE 2014. ridge14sle.pdf extended version.

E3 is available from github at https://github.com/tomjridge/e3. The examples should give some guidance on usage.
P4 is available from github at https://github.com/tomjridge/p4, but there is no documentation at the moment. It was a version of P3 with support for dynamically generated grammars. For example, you could parse a grammar of the form G_n -> <the number n> | <the number n> G_{n+1}:

(* _G 1 parses "1", or "1" followed by _G 2; in general, _G n parses
   the number n, optionally followed by _G (n+1); there are an infinite
   number of nonterminals (_G n) generated lazily on demand *)

let _G = ref (fun n -> failwith "")

let _ = _G := fun n -> 
  let an = a (string_of_int n) in
  let alts = lazy(alts[
    (rhs an) >> (fun s -> c s);
    (an >-- (!_G (n+1))) >> (fun (x,y) -> (c x)^y)])
  in
  mkntparser_lazy (mk_pre_parser ()) alts

P5_scala is a version of P4 for Scala, available at https://github.com/tomjridge/p5_scala. Currently not in a very polished state. Mostly useful for me to try out Scala.