Cairo VM stands for Cairo Virtual Machine (VM). Cairo stands for CPU AIR and is a Turing complete, STARK friendly, CPU like instruction set and programming language. At a high level, it lets the developer write program as with any other programming language, but provable programs.
What is a provable program? Well, a program for which a given execution can be proven without the need to re-execute it. In other words, with a provable program P, you can run P on a given input I, returning output O, and convince anyone that P(I) = O in logarithmic time. An application can be for example that you delegate some heavy computation of cost N to a given cloud provider, and can verify in cost (log N)^2 that they actually did the job and not just returned a random number.
The Cairo VM is currently powering the Starknet blockchain, though its usage is not limited to blockchain applications. There are currently five open sources implementations of the spec, as defined in the Cairo paper:
- The original Python implementation, deprecated since July 2023
- The Starknet current VM, in rust, from LambdaClass
- A primarily educational Cairo VM in go, also from LambdaClass
- Another go implementation, by Nethermind
- Yet another educational implementation in typescript, by Kakarot
- And here we are, this spreadsheet with its associated app script!
At the heart of the Cairo VM lies the concept of felt and relocatable. Because we are here to keep it simple, say that:
- a felt (or Field Element) is a constant number between 0 and p (a given prime number), much like (here the cryptographers die) in usual computation a variable is a uint of some kind (uint32 or uint64 in most of the languages and architectures, uint256 in the Ethereum Virtual Machine for example);
- a relocatable is a reference to a felt or to another relocatable (?). While this sounds complicated, it appears that spreadsheets are by design exactly this, a mix of cells with constant values and cells referencing other cells.
In the end, the task of the Cairo VM is just to fill some cells such that, at each line, the constraints defined by the program holds. For example, one may have at a given line an instruction like:
| Opcode | dst | src |
|---|---|---|
| AssertEq | 10 | =D33 |
In this context, the run of the Cairo VM would consist in writing in D33 the value 10 such that the equality with B33 holds. That's it, you've become a cryptographer specialized in ZK.
The aim of this implementation is definitely not to reach prod perf (though everything is possible in gsheet) but to provide an easy educational, plug-and-play tool for running step by step compiled Cairo codes.
It provides few new functions that can be used directly from the sheet:
- DECODE_INSTRUCTION(hex): disassemble a Cairo compiled program
It also adds a menu "Cairo VM" which allows to:
- load a compiled cairo program
- run it step by step
- run it until the end
This spreadsheet is under active development. Currently, only the DECODE_INSTRUCTION is working. Next priorities are to fix the Step and Run menu over a small fibonacci program. Then, we will add the option to load any compiled program. Eventually, we may stop there or keep improving the VM and support Cairo 1.
To start contributing:
- fork this repo and
git clone(as usual) - run
npm installto get the clasp CLI tool - create a copy of the main Google sheet in the UX
- clone your app script project locally with the CLI
npx clasp login
npx clasp clone <project id> --rootDir .- the project id can be found by:
- opening the menu Extensions > Apps Script
- go to project setting on the left navigation bar
- copy the ID therein
The clasp CLI is not so convenient and you should take a bit of time to play
around and read the doc. Especially, check that everything is set up correctly
by using
npx clasp pathsand make sure that is shows:
project <project root>/.clasp.json
ignore <project root>/.claspignore
auth $HOME/.clasprc.jsonIt is recommended to have jq
installed to be able to check the status of the non-ignored files only
npx clasp status --json | jq ".filesToPush"The npx clasp push command has a watch mode that allows you to make sure that
the remote code running on the sheet is always up to date:
npx clasp push -wYou can run functions locally using npx clasp run. Follow
this documentation to
set your project:
- create a GCP project associated with the script in project setting > GCP project
- add the Apps Script API to your project
- create an OAuth Client ID for a Desktop app followingthis doc
- download the credentials as
creds.jsonand put them at the root of the project npx clasp login --creds creds.jsonclasp run