bug(`anvil`): historical state includes anvil_set* changes made after the block

[zexoverz]

Component

Anvil

What version of Foundry are you on?

anvil 1.8.1, and master at 4cd4320 (1.8.4-dev) behaves the same

What command(s) is the bug in?

eth_getStorageAt, eth_getProof (any historical state read) after anvil_set*

Operating System

macOS (Apple Silicon)

Describe the bug

After anvil_setStorageAt (or anvil_setBalance / anvil_setCode) followed by a mine, reads at the previous block return the new value. The proof root for that block matches no block header.

cast rpc anvil_setCode 0x…aa 0x00
cast rpc anvil_setStorageAt 0x…aa 0x…01 0x…11 && cast rpc evm_mine   # block 1
cast rpc anvil_setStorageAt 0x…aa 0x…01 0x…22 && cast rpc evm_mine   # block 2
cast storage 0x…aa 1 --block 1                                        # 0x…22, expected 0x…11

The same change made through real transactions reads back correctly. It looks like the state for block N is snapshotted when N+1 starts mining, so it's really the pre-state of N+1. Moving the snapshot to right after N would break tracing of N+1 transactions that rely on a cheat, so it probably needs a separate post-state when state changes between blocks. Happy to do that if you're ok with the approach.

[mablr]

Yes, this seems valid. We snapshot the parent state when mining the next block, so anvil_set* changes made between blocks can appear in historical reads for the parent. Keeping a separate post-block state for historical reads and pre-execution state for tracing sounds like the right direction. Please also cover reads and proofs for the current head after anvil_set*, before another block is mined.