⚠️ This project is undergoing heavy development and is still on its early stages. There will be constant breaking changes.
This project aims to implement a Cairo Virtual Machine using Go. This is one of many other implementations that are being developed and its main goals are:
- making the Starknet ecosystem secure by reducing the risk of a single critical vulnerability,
- decentralizing development and maintenance of the different VMs,
- cross-checking and validation with other implementations and
- to foster innovation through competition.
The Cairo Virtual Machine is a crucial component of the Starknet ecosystem. It serves as the runtime environment for all smart contracts on the platform. When users write contracts in high-level Cairo, it gets compiled to Sierra, and then to CASM bytecode. The VM receives this bytecode, executes it and generates a proof of execution. This proof is then sent from a sequencer to the verifier to include the transaction in a new block.
This Virtual Machine is still in development and there is no public release available yet. Currently, it is only possible to use it by building it from source by following these instructions:
- Clone the repo to your machine:
git clone https://github.com/NethermindEth/cairo-vm-go
. - Install
Go
on your PC, instructions here. - Execute on the root folder of the repo:
make build
. - Make sure everything is running smoothly by executing:
make unit
.
After completing these steps, you can find the compiled VM in bin/cairo-vm
.
To run the VM you need to have a compiled Cairo file using the Cairo Zero compiler at cairo-lang.
First, make sure you have Python 3.9.11 installed in your machine. Since this version is quite old and can cause problem with your system's Python we suggest the use of pyenv to manage different Python versions.
Then, install the latest version of the cairo-lang package with the following command:
pip install cairo-lang==0.13.1
When the installation is completed, you can run the cairo-compile
command:
cairo-compile ./integration_tests/cairo_zero_file_tests/factorial.cairo --proof_mode --output ./factorial_compiled.json
This will compile factorial.cairo
and store the compilation result in factorial_compiled.json
. The --proof_mode
flag makes the compilation output contain special identifiers that allow the generation of a proof of execution from the VM later on.
Finally, let's use our VM to execute factorial_compiled.json
with the next command:
./bin/cairo-vm run --proofmode --tracefile factorial_trace --memoryfile factorial_memory factorial_compiled.json
When this command finishes, factorial.cairo
has run correctly starting from the main
function. The --proofmode
flag indicates that a proof of execution should be generated. The location where this proof is stored is determined by both --tracefile
and --memoryfile
flags accordingly.
To learn about all the possible options the VM can be run with, execute the run
command with the --help
flag:
./bin/cairo-vm run --help
We currently have defined three sets of tests:
- unit tests where we check the correct work of each component individually.
- integration tests where we compare that the proof of execution of our VM is the same as the proof of execution of the Python VM.
- benchmark tests to have a baseline performance indicator.
Unit tests can be automatically run with:
make unit
Integration tests are run with:
make integration
Integration tests are run with filters in the following two methods, with the first method having higher priority.
#1) set global environment variable `INTEGRATION_TESTS_FILTERS`
export INTEGRATION_TESTS_FILTERS=fib,alloc
make integration
#2) set by editing `INTEGRATION_TESTS_FILTERS=` in the `./integration_tests/.env` file
make integration
If you want to execute all tests of the project:
make testall
To benchmark the project run:
make bench
This will run benchmarks for most of the project packages. It will create a benchmark folder and a subfolder named after the current branch git information. Inside this subfolder, each package that was benchmarked will have a cpu.out, mem.out and stdout.text. The first two files will hold profiling data regarding CPU and memory usage respectively, the last one will hold allocations/operations per second per benchmark test.
To view profiling information with a web UI, run:
go tool pprof -http=:8080 benchmarks/<subfolder>/<pkg>/cpu.out
You may also be interested in benchmarking a specific package or a specific function, you can use the PKG_NAME
and TEST
flags for this.
make bench PKG_NAME="hintrunner" TEST="AllocSegment"
For convenience, we have created a makefile
that includes the most used commands such as make build
. To see all of them please run:
make help
We are planning on writing our documentation soon detailing how we adapt the theory of a non-deterministic machine to a deterministic one. Meanwhile, the next is a list of resources we are currently using to develop the VM.
- Cairo Zero Docs: How Cairo Works
- Whitepaper: Cairo a Turing-complete STARK-friendly CPU architecture
- A formalization of the whitepaper: A Verified Algebraic Representation of Cairo Program Execution
The previous list includes the most helpful documentation for the current state of the project but it does not represent all that is available. If you are interested in going beyond, there is this list made by LambdaClass which has a much broader scope.
- Cairo VM in Python by Starkware.
- VM in Rust, Go and C by LambdaClass.
- oriac a toy VM by xJonathanLEI
If you wish to contribute please visit our CONTRIBUTING.md for general guidelines.