KEVM: Semantics of EVM in K¶
In this repository we provide a model of the EVM in K.
These may be useful for learning KEVM and K (newest to oldest):
- Jello Paper, generated using Sphinx Documentation Generation.
- 20 minute tour of the semantics at 2017 Devcon3.
- KEVM 1.0 technical report, especially sections 3 and 5.
To get support for KEVM, please join our Riot Room.
There are three backends of K available, the OCAML backend for concrete execution, the Java backend for symbolic reasoning and proofs, and the experimental Haskell backend for developers.
This repository generates the build-products for both backends in
There is also a Haskell version of K currently under development which is meant to eventually replace the Java backend for symbolic reasoning and proofs. To read more about building and using it, please check the K Haskell Backend
The following are needed for building/running KEVM:
- Pandoc >= 1.17 is used to generate the
*.kfiles from the
- GNU Bison, Flex, and Autoconf.
- GNU libmpfr and libtool.
- Java 8 JDK (eg. OpenJDK)
- Opam, important: Ubuntu users prior to 15.04 must build from source, as the Ubuntu install for 14.10 and prior is broken.
opam repositoryalso requires
On Ubuntu >= 15.04 (for example):
sudo apt-get install make gcc maven openjdk-8-jdk flex opam pkg-config libmpfr-dev autoconf libtool pandoc zlib1g-dev
To run proofs, you will also need Z3 prover; on Ubuntu:
sudo apt-get install z3 libz3-dev
sudo pacman -S base-devel rsync opam pandoc jre8-openjdk mpfr maven z3
On OSX, using Homebrew, after installing the command line tools package:
brew tap caskroom/cask caskroom/version brew cask install java8 brew install automake libtool gmp mpfr pkg-config pandoc maven opam z3
NOTE: a previous version of these instructions required the user to run
brew link flex --force.
After fetching this revision, you should first run
brew unlink flex, as it is no longer necessary and will cause an error if you have the homebrew version of flex installed instead of the xcode command line tools version.
After installing the above dependencies, make sure the submodules are setup:
git submodule update --init --recursive
If you haven’t already setup K’s OCaml dependencies more recently than February 1, 2019, then you also need to setup the K OCaml dependencies:
Finally, you can install repository specific dependencies and build the semantics:
make deps make build
OPTIONAL: K Haskell Backend¶
The K Haskell Backend, currently under development, is meant to eventually replace the Java backend for symbolic reasoning and proofs.
In addition to the above dependencies, the Haskell Backend also depends on:
- Haskell Stack.
Note that the version of the
stacktool provided by your package manager might not be recent enough. Please follow installation instructions from the Haskell Stack website linked above.
To run proofs, you will also need Z3 version 4.7.1 or higher.
sudo apt-get install haskell-stack z3 libz3-dev
stack (if needed):
stack upgrade export PATH=$HOME/.local/bin:$PATH
After installing the above dependencies, the following command will build the Kore backend submodule dependency and then the Kore version of KEVM:
make haskell-deps make build-haskell
The following files constitute the KEVM semantics:
- krypto.md sets up some basic cryptographic primitives.
- data.md provides the (functional) data of EVM (256 bit words, wordstacks, etc...).
- evm.md is the main KEVM semantics, containing the configuration and transition rules of EVM.
These additional files extend the semantics to make the repository more useful:
- driver.md is an execution harness for KEVM, providing a simple language for describing tests/programs.
- analysis.md contains any automated analysis tools we develop.
- edsl.md defines high-level notations of eDSL, a domain-specific language for EVM specifications, for formal verification of EVM bytecode using K Reachability Logic Prover.
After building the definition, you can run the definition using
./kevm script for examples of the actual invocations of
Run the file
./kevm run tests/ethereum-tests/VMTests/vmArithmeticTest/add0.json
Run the same file as a test:
./kevm test tests/ethereum-tests/VMTests/vmArithmeticTest/add0.json
To run proofs, you can similarly use
For example, to prove the specification
./kevm prove tests/proofs/specs/vyper-erc20/totalSupply-spec.k
Finally, if you want to debug a given program (by stepping through its execution), you can use the
./kevm debug tests/ethereum-tests/VMTests/vmArithmeticTest/add0.json ... KDebug> s 1 Step(s) Taken. KDebug> p ... Big Configuration Here ... KDebug>
The tests are run using the supplied
make split-tests to generate some of the tests from the markdown files.
The following subsume all other tests:
make test: All of the quick tests.
make test-all: All of the quick and slow tests.
These are the individual test-suites (all of these can be suffixed with
-all to also run slow tests):
If you also want to build the Jello Paper, you’ll additionally need:
sudo apt-get install python-pygments python-sphinx python-recommonmark git clone 'https://github.com/kframework/k-editor-support' cd k-editor-support/pygments easy_install --user .
For the 2017 Devcon3 presentation, you’ll need
pdflatex, commonly provided with
sudo apt-get install texlive-full
Any pull requests into this repository will not be reviewed until at least some conditions are met. Here we’ll accumulate the standards that this repository is held to.
Code style guidelines, while somewhat subjective, will still be inspected before going to review. In general, read the rest of the definition for examples about how to style new K code; we collect a few common flubs here.
Writing tests and more contract proofs is always appreciated. Tests can come in the form of proofs done over contracts too :).
Hard - Every Commit¶
These are hard requirements (must be met before review), and they must be true for every commit in the PR.
- If a new feature is introduced in the PR, and later a bug is fixed in the new feature, the bug fix must be squashed back into the feature introduction.
The only exceptions to this are if you want to document the bug because it was quite tricky or is something you believe should be fixed about K.
In these exceptional cases, place the bug-fix commit directly after the feature introduction commit and leave useful commit messages.
In addition, mark the feature introduction commit with
[skip-ci]if tests will fail on that commit so that we know not to waste time testing it.
- No tab characters, 4 spaces instead.
Linux-style line endings; if you’re on a Windows machine make sure to run
dos2unixon the files. No whitespace at the end of any lines.
Hard - PR Tip¶
These are hard requirements (must be met before review), but they only have to be true for the tip of the PR before review.
- Every test in the repository must pass.
We will test this with
make split-tests ; make test -j12.
Soft - Every Commit¶
These are soft requirements (review may start without these being met), and they will be considered for every commit in the PR.
Comments do not live in the K code blocks, but rather in the surrounding Markdown (unless there is a really good reason to localize the comment).
You should consider prefixing “internal” symbols (symbols that a user would not write in a program) with a hash (
Place a line of
-after each block of syntax declarations.
syntax Foo ::= "newSymbol" // -------------------------- rule <k> newSymbol => . ... </k>
Notice that if there are rules immediately following the syntax declaration, a commented-out line of
-is inserted afterward. Notice that the width of the line of
-matches that of the preceding line.
Place spaces around parentheses and commas in K’s pretty functional-style syntax declarations.
syntax Foo ::= newFunctionalSyntax ( Int , String ) // ---------------------------------------------------
When multiple structurally-similar rules are present, line up as much as possible (and makes sense).
rule <k> #do1 => . ... </k> <cell1> not-done => done </cell1> rule <k> #do1Longer => . ... </k> <cell1> not-done => done-longer </cell1> rule <k> #do2 => . ... </k> <cell2> not-done => done2 </cell2> rule <k> #doShort => . ... </k> <cell2> nd => done2 </cell2>
This makes it simpler to make changes to entire groups of rules at a time using sufficiently modern editors. Notice that if we break alignment (eg. from the
#do1group above to the
#do2group), we put an extra line between the groups of rules.
Line up the
rule(if it’s not all on one line). Similarly, line up the end of
andBoolfor extra side-conditions with the end of
rule <k> A => B ... </k> SOME_LARGE_CONFIGURATION requires A > 3 andBool isPrime(A)
For more information about The K Framework, refer to these sources:
- The K Tutorial
- Semantics-Based Program Verifiers for All Languages
- Reachability Logic Resources
- Matching Logic Resources
- Logical Frameworks: Discussion of logical frameworks.