lark-rust 0.2.0

High-performance Rust-accelerated drop-in replacement for lark's LALR parser stack

lark-rust

High-performance Rust-accelerated drop-in replacement for lark's LALR parser stack. A faster alternative to lark-cython.

Installation

pip install lark-rust

Usage

from lark import Lark
import lark_rust

parser = Lark(grammar, parser='lalr', _plugins=lark_rust.plugins)
result = parser.parse(text)

That's it — same lark API, just faster.

Benchmarks

Measured on Apple M4 Pro, Python 3.13, median of 20 iterations:

Workload	lark	lark-cython	lark-rust	vs lark	vs cython
JSON (3 KB nested)	4.02 ms	2.39 ms	2.04 ms	2.0x	1.2x
Arithmetic (deep)	3.22 ms	1.98 ms	1.29 ms	2.5x	1.5x
Lexer (10k words)	28.66 ms	16.50 ms	11.44 ms	2.5x	1.4x

Reproduce with python benchmarks/bench.py (install lark-cython for the three-way comparison).

What's accelerated

The Rust native extension accelerates the hot-path components:

• LALR parser state machine — feed_token loop with pre-compiled parse table (HashMap lookup, no Python dict overhead)
• Token — creation, hashing, equality comparison
• Scanner — regex matching via fancy-regex
• BasicLexer — token loop with single-pass newline counting
• LineCounter — position tracking
• LexerState — lexer state management

The tree builder, tree traversal, and error recovery are implemented in Python with full lark compatibility:

• ContextualLexer, LALR_Parser (with serialize/deserialize/on_error)
• ParseTreeBuilder with ExpandSingleChild, PropagatePositions, ChildFilter
• Tree with iter_subtrees, find_data, scan_values, pretty printing

lark-cython compatibility

lark-rust is a drop-in replacement for lark-cython. All plugins supported by lark-cython are supported:

• BasicLexer, ContextualLexer, LexerThread
• LALR_Parser, _Parser, ParseTreeBuilder

Requirements

• Python >= 3.8
• lark >= 1.0

License

MIT