This document provides guidance on how we organize and write our smart contracts. For cases where
this document does not provide guidance, please refer to existing contracts for guidance,
with priority on the L2OutputOracle
and OptimismPortal
.
Optimism smart contracts follow the triple-slash solidity natspec comment style with additional rules. These are:
- Always use
@notice
since it has the same general effect as@dev
but avoids confusion about when to use one over the other. - Include a newline between
@notice
and the first@param
. - Include a newline between
@param
and the first@return
. - Use a line-length of 100 characters.
We also have the following custom tags:
@custom:proxied
: Add to a contract whenever it's meant to live behind a proxy.@custom:upgradeable
: Add to a contract whenever it's meant to be inherited by an upgradeable contract.@custom:semver
: Add toversion
variable which indicate the contracts semver.@custom:legacy
: Add to an event or function when it only exists for legacy support.@custom:network-specific
: Add to state variables which vary between OP Chains.
- Use
require
statements when making simple assertions. - Use
revert(string)
if throwing an error where an assertion is not being made (no custom errors). See here for an example of this in practice. - Error strings MUST have the format
"{ContractName}: {message}"
wheremessage
is a lower case string.
- Function parameters should be prefixed with an underscore.
- Arguments returned by functions should be suffixed with an underscore.
- Event parameters should NOT be prefixed with an underscore.
Immutable variables:
- should be in
SCREAMING_SNAKE_CASE
- should be
internal
- should have a hand written getter function
This approach clearly indicates to the developer that the value is immutable, without exposing the non-standard casing to the interface. It also ensures that we don’t need to break the ABIs if we switch between values being in storage and immutable.
We use spacer variables to account for old storage slots that are no longer being used.
The name of a spacer variable MUST be in the format spacer_<slot>_<offset>_<length>
where
<slot>
is the original storage slot number, <offset>
is the original offset position
within the storage slot, and <length>
is the original size of the variable.
Spacers MUST be private
.
All contracts should be assumed to live behind proxies (except in certain special circumstances).
This means that new contracts MUST be built under the assumption of upgradeability.
We use a minimal Proxy
contract designed to be owned by a
corresponding ProxyAdmin
which follow the interfaces
of OpenZeppelin's Proxy
and ProxyAdmin
contracts, respectively.
Unless explicitly discussed otherwise, you MUST include the following basic upgradeability pattern for each new implementation contract:
- Extend OpenZeppelin's
Initializable
base contract. - Include a
uint8 public constant VERSION = X
at the TOP of your contract. - Include a function
initialize
with the modifierreinitializer(VERSION)
. - In the
constructor
, set anyimmutable
variables and call theinitialize
function for setting mutables.
All (non-library and non-abstract) contracts MUST extend the Semver
base contract which
exposes a version()
function that returns a semver-compliant version string.
Contracts must have a Semver
of 1.0.0
or greater to be production ready. Contracts
with Semver
values less than 1.0.0
should only be used locally or on devnets.
Additionally, contracts MUST use the following versioning scheme:
patch
releases are to be used only for changes that do NOT modify contract bytecode (such as updating comments).minor
releases are to be used for changes that modify bytecode OR changes that expand the contract ABI provided that these changes do NOT break the existing interface.major
releases are to be used for changes that break the existing contract interface OR changes that modify the security model of a contract.
We have made an exception to the Semver
rule for the WETH
contract to avoid
making changes to a well-known, simple, and recognizable contract.
Where basic functionality is already supported by an existing contract in the OpenZeppelin library, we should default to using the Upgradeable version of that contract.
The following guidelines should be followed for all contracts in the src/
directory:
- All state changing functions should emit a corresponding event. This ensures that all actions are transparent, can be easily monitored, and can be reconstructed from the event logs.
Tests are written using Foundry.
All test contracts and functions should be organized and named according to the following guidelines.
These guidelines are also encoded in a script which can be run with:
tsx scripts/checks/check-test-names.ts
There is a non-intuitive behavior in foundry tests, which is documented here.
When testing for a revert on a low-level call, please use the revertsAsExpected
pattern suggested there.
Note: This is a work in progress, not all test files are compliant with these guidelines.
- Solidity
contract
s are used to organize the test suite similar to how mocha uses describe. - Every non-trivial state changing function should have a separate contract for happy and sad path tests. This helps to make it very obvious where there are not yet sad path tests.
- Simpler functions like getters and setters are grouped together into test contracts.
Test function names are split by underscores, into 3 or 4 parts. An example function name is test_onlyOwner_callerIsNotOwner_reverts()
.
The parts are: [method]_[FunctionName]_[reason]_[status]
, where:
[method]
is eithertest
,testFuzz
, ortestDiff
[FunctionName]
is the name of the function or higher level behavior being tested.[reason]
is an optional description for the behavior being tested.[status]
must be one of:succeeds
: used for most happy path casesreverts
: used for most sad path casesworks
: used for tests which include a mix of happy and sad assertions (these should be broken up if possible)fails
: used for tests which 'fail' in some way other than revertingbenchmark
: used for tests intended to establish gas costs
Test contracts should be named one of the following according to their use:
TargetContract_Init
for contracts that perform basic setup to be reused in other test contracts.TargetContract_Function_Test
for contracts containing happy path tests for a given function.TargetContract_Function_TestFail
for contracts containing sad path tests for a given function.
To minimize clutter, getter functions can be grouped together into a single test contract,
ie. TargetContract_Getters_Test
.
See the file scripts/FeeVaultWithdrawal.s.sol
to withdraw from the L2 fee vaults. It includes
instructions on how to run it. foundry
is required.