Lark - a modern parsing library

Lark is a modern general-purpose parsing library for Python.

Lark focuses on simplicity, power, and speed. It lets you choose between two parsing algorithms:

• Earley : Parses all context-free grammars (even ambiguous ones)! It is the default.
• LALR(1): Only LR grammars. Outperforms PLY and most (if not all) other pure-python parsing libraries.

Both algorithms are written in Python and can be used interchangeably with the same grammar (aside for algorithmic restrictions). See “Comparison to other parsers” for more details.

Lark can automagically build an AST from your grammar, without any more code on your part.

Lark does things a little differently

1. Separates code from grammar: The result is parsers that are cleaner and easier to read & work with.

2. Automatically builds a tree (AST): Trees are always simpler to work with than state-machines. (But if you want to provide a callback for efficiency reasons, Lark lets you do that too)

3. Follows Python’s Idioms: Beautiful is better than ugly. Readability counts.

Hello World

Here is a little program to parse “Hello, World!” (Or any other similar phrase):

from lark import Lark
l = Lark('''start: WORD "," WORD "!"
            WORD: /\w+/
            %ignore " "
         ''')
print( l.parse("Hello, World!") )

And the output is:

Tree(start, [Token(WORD, 'Hello'), Token(WORD, 'World')])

Notice punctuation doesn’t appear in the resulting tree. It’s automatically filtered away by Lark.

Tiny Calculator

from lark import Lark, InlineTransformer
parser = Lark('''?sum: product
                     | sum "+" product   -> add
                     | sum "-" product   -> sub

                 ?product: item
                     | product "*" item  -> mul
                     | product "/" item  -> div

                 ?item: NUMBER           -> number
                      | "-" item         -> neg
                      | "(" sum ")"

                 %import common.NUMBER
                 %ignore /\s+/
         ''', start='sum')

class CalculateTree(InlineTransformer):
    from operator import add, sub, mul, truediv as div, neg
    number = float

def calc(expr):
    return CalculateTree().transform( parser.parse(expr) )

In the grammar, we shape the resulting tree. The ‘->’ operator renames branches, and the ‘?’ prefix tells Lark to inline single values. (see the tutorial for a more in-depth explanation)

Then, the transformer calculates the tree and returns a number:

>>> calc("(200 + 3*-3) * 7")
1337.0

Learn more about using Lark

• Read the tutorial, which shows how to write a JSON parser in Lark.
• Read the reference
• Browse the examples, which include a calculator, and a Python-code parser.
• Check out the tests for more examples.

Install Lark

$ pip install lark-parser

Lark has no dependencies.

List of Features

• Python 2 & 3 compatible
• Earley & LALR(1)
• EBNF grammar with a little extra
• Builds an AST automagically based on the grammar
• Standard library of terminals (strings, numbers, names, etc.)
• Unicode fully supported
• Extensive test suite
• Lexer (optional)
  • Automatic line & column tracking
  • Automatic token collision resolution (unless both terminals are regexps)
  • Contextual lexing for LALR
• Automatic reconstruction of input (experimental, see examples)

Coming soon

These features are planned to be implemented in the near future:

• Grammar composition
• Optimizations in both the parsers and the lexer
• Better handling of ambiguity
• Automatically convert grammars from/to Nearley, an awesome Earley library in Javascript

Planned

These features may be implemented some day:

• Parser generator - create a small parser, independent of Lark, to embed in your project.
  • Generate code in other languages than Python
• LALR(k) parser
• “Look-back” Enhancement for LALR(1)
• Full regexp-collision support using NFAs
• Automatically produce syntax-highlighters for popular IDEs

Comparison to other parsers

Lark is easier to use

• You can work with parse-trees instead of state-machines
• The grammar is simple to read and write
• There are no restrictions on grammar structure. Any grammar you write can be parsed.
  • Some structures are faster than others. If you care about speed, you can learn them gradually while the parser is already working
  • A well-written grammar is very fast
  • Note: Nondeterminstic grammars will run a little slower
  • Note: Ambiguous grammars (grammars that can be parsed in more than one way) are supported, but may cause significant slowdown if the ambiguity is too big)
• You don’t have to worry about terminals (regexps) or rules colliding
• You can repeat expressions without losing efficiency (turns out that’s a thing)

Performance comparison

Code	CPython Time	PyPy Time	CPython Mem	PyPy Mem
Lark - LALR(1)	4.2s	1.1s	0.4M	0.3M
PyParsing	32s	4.1s	0.4M	0.2M
funcparserlib	11s	1.9s	0.5M	0.3M
Parsimonious		7s		1.4M

Check out the JSON tutorial for more details on how the comparison was made.

Feature comparison

Library	Algorithm	LOC	Grammar	Builds tree?
Lark	Earley/LALR(1)	0.5K	EBNF+	Yes!
PLY	LALR(1)	4.6K	Yacc-like BNF	No
PyParsing	PEG	5.7K	Parser combinators	No
Parsley	PEG	3.3K	EBNF-like	No
funcparserlib	Recursive-Descent	0.5K	Parser combinators	No
Parsimonious	PEG	?	EBNF	Yes

(LOC measures lines of code of the parsing algorithm(s), without accompanying files)

It’s hard to compare parsers with different parsing algorithms, since each algorithm has many advantages and disadvantages. However, I will try to summarize the main points here:

• Earley: The most powerful context-free algorithm. It can parse all context-free grammars, and it’s Big-O efficient. But, its constant-time performance is slow.
• LALR(1): The fastest, most efficient algorithm. It runs at O(n) and uses the least amount of memory. But while it can parse most programming languages, there are many grammars it can’t handle.
• PEG: A powerful algorithm that can parse all deterministic context-free grammars* at O(n). But, it hides ambiguity, and takes a lot of memory to run.
• Recursive-Descent: Fast for simple grammars, and simple to implement. But poor in Big-O complexity.

Lark offers both Earley and LALR(1), which means you can choose between the most powerful and the most efficient algorithms, without having to change libraries.

(* According to Wikipedia, it remains unanswered whether PEGs can really parse all deterministic CFGs)

License

Lark uses the MIT license.

Contact

If you have any questions or want to contribute, you can email me at erezshin at gmail com.