crashlink

Pure-Python HashLink bytecode Swiss Army knife.

Warning

This project is under active development. Breaking changes may be made to APIs with zero notice.

Join the Hashlink Modding Community Discord for support!

Features

• Pure Python with zero dependencies, integrates nicely in a lot of places (IDAPython compatible!)
• Deserialisation, disassembly, and (currently limited) decompilation of HashLink bytecode
• Reserialisation and first-class support for patching bytecode assembly
• A bytecode assembler for creating HashLink bytecode from scratch
• A scriptable interface for easy integration into other tools
• A little CLI similar to hlbc

Installation

pip install crashlink

Optionally, install [extras] for progress bars when parsing large files and faster bytecode save/load in-memory:

pip install crashlink[extras]

Or, for bleeding-edge features, see the Development section.

You also need to have Graphviz installed to generate control flow graphs. On most *nix systems, on Windows (with Chocolatey or Scoop), and on MacOS (with Homebrew), you can install it with your package manager under graphviz.

• Windows: choco install graphviz
• MacOS: brew install graphviz
• Debian: sudo apt install graphviz
• Arch: sudo pacman -S graphviz
• Fedora: sudo dnf install graphviz

In order to work with some of the HL/C utilities, you also need to have Hashlink's core packages installed:

haxelib git hashlink https://github.com/HaxeFoundation/hashlink.git master other/haxelib/

Usage

Either:

$ crashlink path/to/file.hl # or python -m crashlink
crashlink> funcs
f@22 static Clazz.main () -> Void (from Clazz.hx)
f@23 Clazz.method (Clazz) -> I32 (from Clazz.hx)
crashlink> fn 22
f@22 static Clazz.main () -> Void (from Clazz.hx)
Reg types:
  0. Void

Ops:
  0. Ret             {'ret': 0}                                       return

Or:

from crashlink import *
code = Bytecode.from_path("path/to/file.hl")
if code.fn(22): # 22 and 240 are typical entry points for the compiler to generate
  print(disasm.func(code.fn(22)))
elif code.fn(240):
  print(disasm.func(code.fn(240)))
# > f@22 static $Clazz.main () -> Void (from Clazz.hx)
# > Reg types:
# >   0. Void
# >
# > Ops:
# >   0. Ret             {'ret': 0}                                       return

Read the API documentation for more information.

Development

Note

This project is configured for the just command runner and uv. If you don't have them installed, you can still run the commands in the justfile manually and the pip equivalents of the uv commands, but I don't recommend it. At the very least, there's zero downside to switching to uv.

For development purposes, you can clone the repo, install development dependencies, and run the tests:

git clone https://github.com/N3rdL0rd/crashlink
cd crashlink
uv sync --extra dev
just test # or pytest

Before committing, please run just dev to format the code, run tests, and generate documentation in docs/. If you're adding new features to the core serialisation/deserialisation code (core.py), please also add a test case in tests/haxe/ for the new language feature you're adding. If you're adding a feature to the decompiler or disassembler, please add a normal test case (in Python) in tests/ that tests the new feature.

Pull requests are always welcome! For major changes, please open an issue first to discuss what you would like to change.

You can use the following pre-defined commands with just:

• just dev: Run tests, format code, and generate documentation.
• just build: Build the package.
• just install: Install development dependencies and the package in editable mode.
• just build-tests: Build test samples.
• just test: Run tests.
• just format: Format code.
• just docs: Generate documentation.
• just check: Run static analysis/typechecking.
• just clean: Clean up build artifacts.
• just profile: Run the test suite with cProfile and then open the results in a browser.
• just serve-docs: Serve the documentation locally.

Note

A lot of very WIP development is done on dev. If you want to see the latest and greatest, checkout that branch. If you want something semi-stable, checkout main instead.

crashtest CLI

crashtest is a built-in testing system that is used to score the decompiler's output against the original source code. It is used to ensure that the decompiler is working correctly, that the output is correct, that the decompiler is not regressing, and to allow those interested in the project to easily see the state of the decompiler without installing it or running the test suite themselves. You can call it with crashtest auto (or python -m crashtest auto). Make sure you call it from the root of the repository, since it uses relative paths to find the test files and the output directory.

Architecture

Note

IR and the IR optimization layers have not yet been fully implemented.

Roadmap

• Bytecode parsing
• Opcode disassembly
  • Local resolution and naming
• IR lifter (layer 0)
  • If statements
  • Loops
  • Switch opcode statements
  • Function calls
    • CallClosure
  • Closures, lambdas
• IR optimization layers
  • Resolve locals from assigns block
  • Trace optimization
  • Nested if/else/if/else -> switch
• Haxe pseudocode
• Partial recompilation (against stubs of other functions)
• GUI? (customtkinter or dearpygui)
  • Graphical disassembler
  • Embedded CFG viewer through some Graphviz bindings
  • Decompiler and patching interface
  • IR layer viewer

Portability

crashlink is written in pure typed Python with a minimum version of 3.10 (for the | operator and match statement). It should run on any modern platform, and has been tested heavily on Windows, Linux, and MacOS. As well as this, it is portable to many interpreters:

• CPython 3.10+ is the main target - deserialising Dead Cells v35 takes 103.16s and 4.6gb of RAM
• PyPy just works - the same benchmark takes 26s and 2.9gb of RAM
  • tqdm displays broken progress bars, but this is a PyPy issue and it still works.
• IronPython and Jython are not supported due to their earlier Python version targets.
• RustPython would work, but it doesn't support match statements.

Credits

• Thank you to Gui-Yom for writing hlbc and for maintaining documentation on the HashLink bytecode format, as well as for providing tests and helping me during development.
• Thank you to Haxe Foundation for creating the HashLink VM and the Haxe programming language.
• Thank you to the Dead Cells community on Discord for providing me with the motivation to start this project.
• And a big thank you to you, dear user, for being at least partially interested in this project.

❤ N3rdL0rd