Add support for Celo's stateful transfer precompile

[karlb]

Motivation

The Celo blockchain makes use of a stateful transfer precompile. Implementing this in Anvil allows running more of our tests in Anvil.

In addition to providing this feature to Celo blockchains, it also serves as an example on how to add stateful precompiles to anvil, closing #11210

Solution

Since inject_precompiles does not support stateful precompiles, a Celo-specific lookup function is added to the PrecompilesMap that provides access to the transfer precompile.

PR Checklist

• Added Tests
• Added Documentation
• Breaking changes (none)

[mattsse]

supportive, because this can serve as a reference for additional precompile support.

left some pointers, re reusing the PrecompilesMap and inject dynamically.

we can likely clean up a few more things as well,

perhaps rethink how and where we call

foundry/crates/anvil/src/eth/backend/executor.rs

Lines 337 to 345 in 79d4ab2

	let mut evm = new_evm_with_inspector(&mut *self.db, &env, &mut inspector);
	if self.odyssey {
	inject_precompiles(&mut evm, vec![(P256VERIFY, P256VERIFY_BASE_GAS_FEE)]);
	}
	if let Some(factory) = &self.precompile_factory {
	inject_precompiles(&mut evm, factory.precompiles());
	}

ideally we only have these in one location

[mattsse]

hmm, not convinced we want to add a new arg for every network that we can add custom precompiles for.

should we perhaps derive this from the chainid, although an additional celo flag doesnt really hurt

[karlb]

If we only rely on the chainid, new networks won't work with anvil until they have been explicitly supported. And getting PRs with config for short-lived and local networks sounds annoying to me. Adding a flag seems the lesser evil to me.

We can still automatically detect the flags for the most important networks based on the chainid. But if we need a flag anyway, that falls into the "nice to have" category.

[karlb]

We could also enable the transfer precompile by setting --hardfork cel2, but I'm not sure if adding chain-specific hardforks is better than adding a flag for the chain itself.

[mattsse]

I believe this is now redundant, because we can now dynamically inject precompiles into the regular evm with PrecompilesMap via

foundry/crates/anvil/src/evm.rs

Lines 28 to 29 in 79d4ab2

	evm.precompiles_mut().apply_precompile(&addr, move |_| {
	Some(DynPrecompile::from(move |input: PrecompileInput<'_>| func(input.data, gas)))

[karlb]

In its current form, this can only be used to inject stateless precompiles. The PrecompileWithAddress that is passed into this function does not take any context/state as parameter:

pub struct PrecompileWithAddress(pub Address, pub PrecompileFn);
pub type PrecompileFn = fn(&[u8], u64) -> PrecompileResult;

But now that PrecompilesMap has gained a settable lookup function, I can try to use that to include a stateful precompile. Unfortunately, the EvmInternals in the PrecompileInput don't provide access to the journal, so that I can't use the journal's transfer method like I do in this PR. I might be able to use load_account instead, since it returns a mutable Account.

[karlb]

I updated the code with this approach and it looks much better now.

[karlb]

I'm quite happy with the current implementation and updated the PR description. Before I add tests and/or docs, some feedback on the following points would be appreciated

• Is the --celo flag ok for this PR or should I change how the precompile is enabled
• What kind of tests would you recommend for this change
• Are there any major changes needed or is it a good time to start on tests and docs now?