NOTE: you need to clone this repo using the --recursive flag since this repo has submodules, e.g., git clone git@github.com:nevillegrech/gigahorse-toolchain.git --recursive. If you already cloned without --recursive, run git submodule update --init --recursive to fetch the souffle-addon submodule.

The Gigahorse binary lifter and toolchain

A binary lifter (and related framework) from low-level EVM code to a higher-level function-based three-address representation, similar to LLVM IR or Jimple.

Quickstart

Running/Installing Gigahorse from local clone (requires souffle)

Our dependencies installation Dockerfile can be used as a reference for installing the dependencies of Gigahorse on a debian-based system. The Dockerfile also includes the instructions to build souffle 2.4.1 from source.

In summary, you need to have the following things installed on your system:

• Boost libraries, required for souffle-addon (Can be installed on Debian with apt install libboost-all-dev)

• Z3, required for souffle-addon (Can be installed on Debian with apt install libz3-dev)

• uv, the Astral Python project manager. uv provisions the required Python interpreter (3.13, pinned in .python-version) and the project's Python dependencies for you — see the environment setup step below. (Install with curl -LsSf https://astral.sh/uv/install.sh | sh.)

• Souffle 2.3 or 2.4.1 (We only test using the release versions, later development versions may work but are untested by us. The easiest way to install this is to use the release from https://github.com/souffle-lang/souffle/releases/tag/2.4.1). Refer to the Souffle documentation for more information.

Now install the Souffle custom functors:

# builds all, sets libfunctors.so as a link to libsoufflenum.so
cd souffle-addon && make WORD_SIZE=$(souffle --version | sed -n 3p | cut -c12,13)

Finally, set up the Python environment with uv, which provisions the pinned CPython 3.13 and the project's Python dependencies into a local .venv:

uv sync

You can then run the toolchain through uv (e.g. uv run ./gigahorse.py examples/long_running.hex), or activate the environment (source .venv/bin/activate) so that ./gigahorse.py uses Python 3.13 directly. Optional feature sets are available as extras: uv sync --extra tooling (Datalog directive parsing + interactive graph visualisation) and uv sync --extra viz (Graphviz/dot rendering).

You should now be ready to run Gigahorse.

Installing Gigahorse via docker

Alternatively, you can use Gigahorse via our pre-built docker images using the following instructions:

• For amd64:

curl -s -L https://raw.githubusercontent.com/nevillegrech/gigahorse-toolchain/master/scripts/docker/install/install_amd64 | bash

For arm64/m1 (not actively tested):

curl -s -L https://raw.githubusercontent.com/nevillegrech/gigahorse-toolchain/master/scripts/docker/install/install_arm64 | bash

• Then

  ~/.bashrc

• Check if gigahorse is available using

  --help

Running Gigahorse

The gigahorse.py script can be run on a contract individually or on a collection of contract bytecode files in specified directory, and it will run the binary lifter implemented in logic/main.dl on each contract, optionally followed by any additional client analyses specified by the user using the -C flag.

The default pipeline first attempts to decompile a contract using a shrinking context-sensitivity configuration. If that times out it performs a second attempt with the scalable-fallback configuration (using a finite-precise context sensitivity algorithm, tuned for scalability). The scalable-fallback configuration can be disabled if needed using the --disablescalablefallback flag.

The Gigahorse pipeline also includes a few rounds of inlining of small functions in order to help the subsequent client libraries get more high-level inferences. The inlining functionality can be disabled with --disable_inline.

The expected file format for each contract is in .hex format.

Example (individual contract):

./gigahorse.py examples/long_running.hex

(For some Souffle versions, you will get an error message regarding the libsoufflenum.so dynamic library, during the first compilation. You can ignore this and gigahorse.py should work upon a re-run.)

Contracts that take too long to analyse will be skipped after a configurable timeout.

The decompilation results are placed in the directory .temp, whereas metadata about the execution, e.g., metrics are placed in a results.json file, as a list of triples in the form:

[filename, properties, flags]

Here, properties is a list of the detected issues with the contract in filename, where any output relations in the datalog files that are non-empty will have their relation name placed in this list. flags is a list indicating auxiliary or exceptional information. It may include "ERROR" and "TIMEOUT", which are self-explanatory.

gigahorse.py --help for invocation instructions.

Example (with client analysis):

./gigahorse.py  -j <number of jobs> -C clients/visualizeout.py <contracts>

(The clients following the -C flag can be a comma-separated list, with no spaces, of path-reachable or fully-qualified filenames.)

Gigahorse can also be used in "bulk analysis" mode, by replacing by a directory filled with contracts.

For additional instructions in tuning the Gigahorse framework see Advanced.md.

Textual representation of the lifted IR

Client analysis clients/visualizeout.py can be used to provide a pretty-printed textual representation of the IR produced by Gigahorse. The pretty-printed text file is named contract.tac and will be placed in the out/ folder for each analyzed contract. For example the output for ./gigahorse.py -C clients/visualizeout.py examples/longrunning.hex will be placed in .temp/longrunning/out/contract.tac.

A block visualized in contract.tac looks like:

Begin block 0x3e     prev=[0xb], succ=[0x10ee, 0x49]     =================================     0x3f: v3f(0xf42fdfb) = CONST      0x44: v44 = EQ v3f(0xf42fdfb), v32     0x10c7: v10c7(0x10ee) = CONST      0x10c8: JUMPI v10c7(0x10ee), v44

Keep in mind that the pretty-printed variable identifiers do not correspond to their identifiers in the underlying datalog facts.

Writing client analyses

Client analyses can be written in any language by reading the relational files that are written by the decompilation step (main.dl). This framework however provides preferential treatment for clients written in Datalog. The most notable example of client analysis for the Gigahorse framework is MadMax. This uses several of the "analysis client libraries" under clientlib. These libraries include customizable dataflow analysis, memory modeling, data structure reconstruction and others.

A common template for client analyses for decompiled bytecode is to create souffle datalog file that includes clientlib/decompiler_imports.dl, for instance:

#include "clientlib/decompiler_imports.dl"

.output ...

Uses of Gigahorse

The Gigahorse toolchain was originally published as:

• Grech, N., Brent, L., Scholz, B., Smaragdakis, Y. (2019), Gigahorse: Thorough, Declarative Decompilation of Smart Contracts.

In 41st ACM/IEEE International Conference on Software Engineering.

Several novel developments to Gigahorse after the original publication have been published as:

• Grech, N., Lagouvardos, S., Tsatiris, I., Smaragdakis, Y. (2022), Elipmoc: Advanced Decompilation of Ethereum Smart Contracts

Proceedings of the ACM in Programming Languages (OOPSLA).

• Lagouvardos, S., Bollanos, Y., Grech, N., Smaragdakis, Y. (2025), The Incredible Shrinking Context... in a Decompiler Near You

Proceedings of the ACM in Software Engineering (ISSTA).

The analysis of EVM "memory" operations (SHA3, xCALL, LOGx, etc.) that is bundled with Gigahorse was published as:

• Lagouvardos, S., Grech, N., Tsatiris, I., Smaragdakis, Y. (2020) Precise Static Modelling of Ethereum "Memory".

Proceedings of the ACM in Programming Languages (OOPSLA).

In addition, other research tools have been developed on top of Gigahorse, including:

• Grech, N., Kong, M., Jurisevic, A., Brent, L., Scholz, B., Smaragdakis, Y. (2018), MadMax: Surviving Out-of-Gas Conditions in Ethereum Smart Contracts. Proceedings of the ACM on Programming Languages (OOPSLA).

• Brent, L., Grech, N., Lagouvardos, S., Scholz, B., Smaragdakis, Y. (2020), Ethainter: A Smart Contract Security Analyzer for Composite Vulnerabilities.

In 41st ACM SIGPLAN Conference on Programming Language Design and Implementation.

• Grech, N., Kong, M., Jurisevic, A., Brent, L., Scholz, B., Smaragdakis, Y. (2020), Analyzing the Out-of-Gas World of Smart Contracts. Communications of the ACM.

• Smaragdakis, Y., Grech, N., Lagouvardos, S., Triantafyllou, K., Tsatiris, I. (2021), Symbolic Value-Flow Static Analysis: Deep, Precise, Complete Modeling of Ethereum Smart Contracts. Proceedings of the ACM in Programming Languages (OOPSLA).

The Gigahorse framework also underpins the realtime decompiler and analysis tool at app.dedaub.com.