jwasinger / execution-spec-tests

Tools & libraries for generating cross-client tests.

Execution Spec Tests

This repository provides tools and libraries for generating cross-client Ethereum tests. Further documentation.

Quick Start

Prerequisites

The following are required to either generate or develop tests:

1. Python >=3.10.0 < 3.11.

   • For dists. with the apt package manager ensure you have python -dev & -venv packages installed.

2. go-ethereum geth's evm utility must be accessible in the PATH, typically at the latest version. To get it:

   1. Install the Go programming language on your computer.
   2. Clone the Geth repository.
   3. Run make all.
   4. Copy build/bin/evm to a directory on the path.

   Note: To update to a different Geth branch (for example one that supports a specific EIP) all you need to do is to change the evm in the path.

3. solc >= v0.8.17; solc must be in accessible in the PATH.

Installation

To generate tests from the test "fillers", it's necessary to install the Python packages provided by execution-spec-tests (it's recommended to use a virtual environment for the installation):

git clone https://github.com/ethereum/execution-spec-tests
cd execution-spec-tests
python3.10 -m venv ./venv/
source ./venv/bin/activate
pip install -e .

After the installation, run this sanity check to ensure tests are generated. If everything is OK, you will see the beginning of the JSON format filled test.

tf --test-case yul
head fixtures/example/yul_example/yul.json

Generating the Execution Spec Tests For Use With Clients

To generate all the tests defined in the ./fillers sub-directory, run the tf command:

tf --filler-path="fillers" --output="fixtures" 

This is equivalent to running tf with no arguments. The pathsfillers/ and fixtures/ are both defaults for the respective command.

Note that the test post conditions are tested against the output of the geth evm utility during test generation.

To generate all the tests within the ./fillers/vm directory (category), for example, run:

tf --test-categories vm

This extends to sub-directories. To generate all specific tests within the ./fillers/vm/vm_arith/vm_add sub-directory, run:

tf --output="fixtures" --test-categories vm/vm_arith/vm_add

To generate all the tests in the ./fillers/*/dup.py modules, for example, run:

tf --test-module dup

To generate specific tests, such as ./fillers/*/*.py::test_dup, for example, run (remove the test_ prefix from the test case's function name):

tf --test-case dup

Testing the Execution Spec Tests Framework

The Python packages provided by the execution spec tests framework have their own test suite that can be ran via tox:

python -m venv ./venv/
source ./venv/bin/activate
pip install tox
tox -e py3

Execution Spec Tests Package Overview

ethereum_test_tools

The ethereum_test_tools package provides primitives and helpers to allow developers to easily test the consensus logic of Ethereum clients.

ethereum_test_filling_tool

The ethereum_test_filling_tool package is a CLI application that recursively searches a given directory for Python modules that export test filler functions generated using ethereum_test_tools. It then processes the fillers using the transition tool and the block builder tool, and writes the resulting fixture to file.

evm_block_builder

This is a wrapper around the block builder (b11r) tool.

evm_transition_tool

This is a wrapper around the transaction (t8n) tool.

fillers

Contains all the Ethereum consensus tests available in this repository.

Writing Tests

Purpose of test specs in this repository

The goal of the test specs included in this repository is to generate test vectors that can be consumed by any Execution client, and to verify that all of the clients agree on the same output after executing each test.

Consensus is the most important aspect of any blockchain network, therefore, anything that modifies the state of the blockchain must be tested by at least one test in this repository.

The tests focus on the EVM execution, therefore before being able to properly write a test, it is important to understand what the Ethereum Virtual Machine is and how it works.

Types of tests

At the moment there are only two types of tests that can be produced by each test spec:

• State Tests
• Blockchain Tests

The State tests span a single block and, ideally, a single transaction.

Examples of State tests:

• Test a single opcode behavior
• Verify opcode gas costs
• Test interactions between multiple smart contracts
• Test creation of smart contracts

The Blockchain tests span multiple blocks which may or may not contain transactions and mainly focus on the block to block effects to the Ethereum state.

• Verify system-level operations such as coinbase balance updates or withdrawals
• Verify fork transitions
• Verify blocks with invalid transactions/properties are rejected

Adding a New Test

All currently implemented tests can be found in the fillers directory, which is composed of many subdirectories, and each one represents a different test category.

Source files included in each category contain one or multiple test specs represented as python functions, and each can in turn produce one or many test vectors.

A new test can be added by either:

• Adding a new test_ python function to an existing file in any of the existing category subdirectories within fillers.
• Creating a new source file in an existing category, and populating it with the new test function(s).
• Creating an entirely new category by adding a subdirectory in fillers with the appropriate source files and test functions.
  • Tests within multiple sub-directories must have a __init__.py file within each directory above it (and it own), to ensure the test is found by the test filler tf.

Test Spec Generator Functions

Every test spec is a python generator function which can perform a single or multiple yield operations during its runtime to each time yield a single StateTest/BlockchainTest object.

The test vector's generator function must be decorated by only one of the following decorators:

• test_from
• test_from_until
• test_only

These decorators specify the forks on which the test vector is supposed to run.

They also automatically append necessary information for the ethereum_test_filling_tool to process when the generator is being executed to fill the tests.

The test vector function must take only one str parameter: the fork name.

StateTest Object

The StateTest object represents a single test vector, and contains the following attributes:

• env: Environment object which describes the global state of the blockchain before the test starts.
• pre: Pre-State containing the information of all Ethereum accounts that exist before any transaction is executed.
• post: Post-State containing the information of all Ethereum accounts that are created or modified after all transactions are executed.
• txs: All transactions to be executed during the test vector runtime.

BlockchainTest Object

The BlockchainTest object represents a single test vector that evaluates the Ethereum VM by attempting to append multiple blocks to the chain:

• pre: Pre-State containing the information of all Ethereum accounts that exist before any block is executed.
• post: Post-State containing the information of all Ethereum accounts that are created or modified after all blocks are executed.
• blocks: All blocks to be appended to the blockchain during the test.

Pre/Post State of the Test

The pre and post states are elemental to setup and then verify the outcome of the state test.

Both pre and post are mappings of account addresses to account structures (see more info).

A single test vector can contain as many accounts in the pre and post states as required, and they can be also filled dynamically.

storage of an account is a key/value dictionary, and its values are integers within range of [0, 2**256 - 1].

txs are the steps which transform the pre-state into the post-state and must perform specific actions within the accounts (smart contracts) that result in verifiable changes to the balance, nonce, and/or storage in each of them.

post is compared against the outcome of the client after the execution of each transaction, and any differences are considered a failure

When designing a test, all the changes must be ideally saved into the contract's storage to be able to verify them in the post-state.

Test Transactions

Transactions can be crafted by sending them with specific data or to a specific account, which contains the code to be executed

Transactions can also create more accounts, by setting the to field to an empty string.

Transactions can be designed to fail, and a verification must be made that the transaction fails with the specific error that matches what is expected by the test.

Writing code for the accounts in the test

Account bytecode can be embedded in the test accounts by adding it to the code field of the account object, or the data field of the tx object if the bytecode is meant to be treated as init code or call data.

The code can be in either of the following formats:

• bytes object, representing the raw opcodes in binary format
• str, representing an hexadecimal format of the opcodes
• Code compilable object

Currently supported built-in compilable objects are:

• Yul object containing Yul source code

Code objects can be concatenated together by using the + operator.

Verifying the Accounts' Post State

The state of the accounts after all blocks/transactions have been executed is the way of verifying that the execution client actually behaves like the test expects.

During their filling process, all tests automatically verify that the accounts specified in their post property actually match what was returned by the transition tool.

Within the post dictionary object, an account address can be:

• None: The account will not be checked for absence or existence in the result returned by the transition tool.
• Account object: The test expects that this account exist and also has properties equal to the properties specified by the Account object.
• Account.NONEXISTENT: The test expects that this account does not exist in the result returned by the transition tool, and if the account exists, it results in error. E.g. when the transaction creating a contract is expected to fail and the test wants to verify that the address where the contract was supposed to be created is indeed empty.

The Account object

The Account object is used to specify the properties of an account to be verified in the post state.

The python representation can be found in src/ethereum_test_tools/common/types.py.

It can verify the following properties of an account:

• nonce: the scalar value equal to a) the number of transactions sent by an Externally Owned Account, b) the amount of contracts created by a contract.

• balance: the amount of Wei (10^-18 Eth) the account has.

• code: Bytecode contained by the account. To verify that an account contains no code, this property needs to be set to "0x" or "".

  It is not recommended to verify Yul compiled code in the output account, because the bytecode can change from version to version.

• storage: Storage within the account represented as a dict object. All storage keys that are expected to be set must be specified, and if a key is skipped, it is implied that its expected value is zero. Setting this property to {} (empty dict), means that all the keys in the account must be unset (equal to zero).

All account's properties are optional, and they can be skipped or set to None, which means that no check will be performed on that specific account property.

Verifying correctness of the new test

A well written test performs a single verification output at a time.

A verification output can be a single storage slot, the balance of an account, or a newly created contract.

It is not recommended to use balance changes to verify test correctness, as it can be easily affected by gas cost changes in future EIPs.

The best way to verify a transaction/block execution outcome is to check its storage.

A test can be written as a negative verification. E.g. a contract is not created, or a transaction fails to execute or runs out of gas.

These verifications must be carefully crafted because it is possible to end up having a false positive result, which means that the test passed but the intended verification was never made.

To avoid these scenarios, it is important to have a separate verification to check that test is effective. E.g. when a transaction is supposed to fail, it is necessary to check that the failure error is actually the one expected by the test.

Failing or invalid transactions

Transactions included in a StateTest are expected to be intrinsically valid, i.e. the account sending the transaction must have enough funds to cover the gas costs, the max fee of the transaction must be equal or higher than the base fee of the block, etc.

An intrinsically valid transaction can still revert during its execution.

Blocks in a BlockchainTest can contain intrinsically invalid transactions but in this case the block is expected to be completely rejected, along with all transactions in it, including other valid transactions.