Hyperlark

Hyperlark is a Rust reimplementation of the Lark parsing toolkit, built with a focus on ergonomics, performance and modularity.

Hyperlark can parse all context-free languages. To put it simply, it means that it is capable of parsing almost any programming language out there, and to some degree most natural languages too.

Hyperlark implements the vast majority of Lark's features, and can often be used as a drop-in replacement, with none or very few code changes.

Status: 0.1.0-beta.1 — pre-1.0, APIs may still change between betas.

Install

pip install hyperlark

Wheels are CPython 3.10+ (stable-ABI, abi3), for Linux (glibc and musl, x86-64 and aarch64), macOS 13+ (x86-64 and arm64) and Windows x64.

Not supported: PyPy, and free-threaded CPython (3.13t / 3.14t)

Hello World

import hyperlark

parser = hyperlark.Lark(r"""
    start: WORD+
    %import common.WORD
    %ignore " "
""")

tree = parser.parse("hello world")
print(tree.pretty())

For complete, runnable programs — a calculator, JSON, an indentation-sensitive language, ambiguity, interactive parsing, error reporting — see the examples page.

Features

• An industry-standard grammar. Lark is widely used, its grammar syntax was adopted by OpenAI for their official API, and .lark files get syntax highlighting on GitHub.
• Choose between LALR(1) and Earley (SPPF) parsers. LALR(1) is fast, linear-time and low on memory. Earley can parse every context-free grammar, and can efficiently handle and store every ambiguity, for later queries.
• Choose between several lexers, or provide your own. Hyperlark provides sophisticated lexers that can help LALR(1) disambiguate tokens based on the parser state, and that help Earley handle lexical ambiguities much faster than scannerless methods. A postlex pass, such as the bundled Indenter, covers indentation-sensitive languages.
• Automatic tree construction. Use the Transformer / Visitor / Interpreter classes on the parse tree, or hand the transformer to transformer= to skip tree construction entirely.
• Interactive parsing (for LALR). Drive the parser and query it at any point of the parse. Useful for error handling, checkpoints & backtracking, and even debugging.
• Line positions. Hyperlark keeps track of the line and column of every token, and will even propagate the ranges to the tree nodes when propagate_positions=True.
• Grammar composition. Import rules, terminals, or entire grammars into your grammar, using an inheritance-like interface (namespaces, overrides), for better re-use and modularity.

Performance

10× lark engine for engine on LALR, and 16–22× on Earley.

Actual speed depends on grammar, input size and how Hyperlark is used.

For more information, see the benchmarks.

Differences from Lark

fast_tokens

To avoid creating a Python object for each token, Hyperlark uses its own Token object. As a consequence, hyperlark.Token does not inherit from str. It does provide some of the str methods (.upper(), .startswith(), …) for convenience, but reach for .value wherever a real str is required.

This can be disabled with the fast_tokens=False option, which restores Lark-identical str-subclass tokens everywhere, at some cost in performance.

Not implemented

The reconstructor, tree templates, the standalone tool, and TextSlice inputs.

These options are recognized but not supported. Each raises a clear ConfigurationError at construction, rather than silently doing nothing:

Beyond these, a handful of behaviors differ in detail: lexer callbacks, the parser object's surface, class identity under mixed lark + hyperlark imports, and recursion inside your own callbacks. See the full list of divergences.

Links

• Documentation — the grammar language, guides, and the API reference
• Lark — the API and grammar language Hyperlark reimplements
• Questions and bug reports: erezshin@gmail.com
• Commercial licensing: hyperlark@eshsoft.com

Hyperlark's source hasn't been published at this time.

License

Free for noncommercial use, including evaluation and testing inside commercial organizations. Commercial use requires a separate license — contact hyperlark@eshsoft.com.

Copyright © 2026 Esh Software LLC. Full text in the LICENSE file in the distribution.