internal/ethapi: Set basefee for AccessList based on given block, not chain tip

[jwasinger]

The specification for eth_accessList takes a BlockNumberOrHash as an optional parameter (arguably, any other state than latest/pending is useless). Before this PR, we always default to sourcing the gas tip based on a value that would be appropriate for the state at the head of the chain.

However, when the base fee of a historical target block exceeds this value, the access list creation fails. This PR covers this (niche) error case.

Several functions for filling call/transaction parameters:

• setDefaults fills in the default values for transaction fields based on the current state. It is also responsible for setting blob fields and chain id, both of which are irrelevant in the the context of access list creation.
• setFeeDefaults fills tx fee fields with sensible values based the hard fork at the current block.
• CallDefaults sanitizes fields relevant to calls, ensuring they are not nil.

In this implementation, I modify access list creation to call setFeeDefaults instead of setDefaults. setFeeDefaults now sets fee fields based on a provided header (only relevant for post 1559 dynamic txs which compute the base fee based on parent header). I also compute set the nonce based on what it historically was at the given header, and call CallDefaults.

The only functionality that the new logic doesn't cover by not calling setDefaults is setting blob tx fields and gas estimation: both of which are irrelevant and not used by access list creation.

[s1na]

Yeah I think it makes sense. We shouldn't need to estimate actual gas prices or gas limit. Sure those can influence EVM execution but the user can pass in those if they care/their contract depends on it.

Only one thing I noticed: not filling the nonce causes the To address computation possibly wrong for create txes.

[jwasinger]

Only one thing I noticed: not filling the nonce causes the To address computation possibly wrong for create txes.

So this would mean that this is broken in all eth_call variants?

[jwasinger]

It's not broken. The reason is that the nonce during creation is sourced from the state of the account, not what's in the transaction args.

[s1na]

It's not broken. The reason is that the nonce during creation is sourced from the state of the account, not what's in the transaction args.

Yep for eth_call it's not broken but here it will be when using CallDefaults. It's because CreateAccessList uses the nonce from tx args and passes that to the tracer. We could either here use account state, or modify the tracer to use OnEnter && level == 0 to get the to address.

[jwasinger]

If the provided nonce was not correct, the transaction will fail in TransactionArgs.TransitionDb. So I think the current code should be okay (?)

[holiman]

the setDefaults has a lot of logic handling blobs that are passed in the args, and that will become broken if we just change this

[fjl]

We could just call setBlobTxSidecar in CallDefaults. But I'd say it's OK not to support them here.

[holiman]

But setDefaults will set the fee defaults on the latest block.

	if err := args.setFeeDefaults(ctx, b); err != nil {
		return err
	}

And regardless of what order we invoke it, it will use the latest header, and do

	if err := args.setCancunFeeDefaults(ctx, head, b); err != nil {
		return err
	}

SO it will apply cancun checks according to tip, regardless of what the given block is.

[holiman]

How about something like this

// setFeeDefaults fills in default fee values for unspecified tx fields.
func (args *TransactionArgs) setFeeDefaults(ctx context.Context, b Backend) error {
	return setFeeDefaultsUsingThisHeader(b.CurrentHeader())
}
func (args *TransactionArgs) setFeeDefaultsUsingThisHeader(ctx context.Context, b Backend, head *types.Header) error {
	// Sanity check the EIP-4844 fee parameters.

[s1na]

the setDefaults has a lot of logic handling blobs that are passed in the args, and that will become broken if we just change this

The logic there is to fill in blob proofs and commitments if they were not provided. Those matter only when submitting a tx. Since for a simulation all that can be accessed via evm is blob hashes (which still can be passed in via TransactionArgs).

All in all I think I agree with @jwasinger that CallDefaults is a better validation function to use here. All the methods that involve signing and submitting a tx use setDefaults while all the "simulation" methods use CallDefaults.

[holiman]

LGTM

[holiman]

@jwasinger please update the top-level PR description (I'm assuming the code has changed since that was written?), so that it's an accurate description. This is a typical such PR that we might revisit three years from now, when trying to figure out why something works in a particular way. So it helps if the desc is up to date with what the PR does.

Other than that, let's merge on green IMO

[jwasinger]

Done. I tried to summarize it as succinctly as possible but it got a bit long :)