Repository navigation
feat(spec-specs,test-forks,tests): charge EIP-8250 keyed nonce access in intrinsic gas and the calldata floor - #3756
Conversation
… in intrinsic gas and the calldata floor Implements ethereum/EIPs#12457: every non-zero nonce key pays KEYED_NONCE_ACCESS_COST (EIP-8038's COLD_STORAGE_ACCESS + STORAGE_WRITE, 12,100 gas) for the cold read and the write of its nonce manager slot, in both frame_tx_intrinsic_gas and calldata_floor_gas. The legacy key set pays nothing. The framework's frame intrinsic and floor calculators price it through a nonce access hook that EIP-8250 overrides. test_keyed_nonce_access_cost builds its expected gas from the EIP's constant, independently of the calculators, for the legacy key set and for fresh, used, two and sixteen keys, with the standard cost and with the calldata floor binding. test_block_access_list now also requires the written slots to be absent from storage_reads, with one and two keys, as the EIP's block access list section states. The reference spec pin moves to the merged EIP text.
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## eips/bogota/eip-8250 #3756 +/- ##
=======================================================
Coverage ? 94.95%
=======================================================
Files ? 677
Lines ? 41945
Branches ? 3961
=======================================================
Hits ? 39830
Misses ? 1464
Partials ? 651
Flags with carried forward coverage won't be shown. Click here to find out more. ☔ View full report in Codecov by Harness. 🚀 New features to boost your workflow:
|
| if floor_bound: | ||
| worker = pre.deploy_contract(code=Op.STOP) | ||
| worker_gas = fork.frame_entry_gas_calculator()(target_warm=False) | ||
| worker_data = FLOOR_PADDING |
There was a problem hiding this comment.
Having a constant data padding is suboptimal because: (a) it might break the test in the future unknowingly, and (b) we would not be able to catch a potential off-by-one bug in a client. If possible, consider doing a binary search to find the minimum amount of bytes that bind the floor.
If you consider it out of scope for this PR (since this constant already existed and it's used by other tests) please open a tracking issue and assign to the original author of this file. cc @spencer-tb
There was a problem hiding this comment.
Done in 208aa47, for both tests in this file that used the constant. The constant also wasn't what bound the floor: with a no-op worker, the signature and nonce bytes priced at the floor rate already put it above the standard cost, so the 30,000 bytes only widened the margin. Both tests now run a worker whose cold reads keep the standard cost above the floor and binary-search the fewest bytes of frame data that bind it, and test_keyed_nonce_access_cost gains a below_floor case one byte short of that. Making the spec count one byte too many in the floor fails floor and below_floor; one byte too few fails floor.
test_gas_settlement.py on eips/bogota/eip-8141 has the same 30,000-byte padding (FLOOR_PADDING_DATA); happy to give it the same treatment in a separate PR against that branch, or open an issue if you prefer.
There was a problem hiding this comment.
Will update these in some follow ups!
…m padding FLOOR_PADDING lifted the floor with a fixed 30,000 bytes, a margin no one-byte miscount of the charged data could cross, and it was not what made the floor bind: the frames' signature and nonce bytes, priced at the floor rate, already outweighed the stand-in worker's gas. Both floor tests now run a worker whose cold reads keep the standard cost above the floor, and a binary search over the worker's frame data finds the fewest bytes that bind the floor. test_keyed_nonce_access_cost adds a below_floor case one byte short of that, where the standard cost binds again. A spec that counts the floor's data one byte too many fails the below_floor and floor cases; one that counts one byte too few fails floor.
… in intrinsic gas and the calldata floor (#3756)
… in intrinsic gas and the calldata floor (#3756)
… in intrinsic gas and the calldata floor (#3756)
… in intrinsic gas and the calldata floor (#3756)
3 participants
Description
Implements ethereum/EIPs#12457 for EIP-8250: every non-zero nonce key pays
KEYED_NONCE_ACCESS_COST(EIP-8038'sCOLD_STORAGE_ACCESS + STORAGE_WRITE, 12,100 gas) for the cold read and the write of its nonce manager slot, in bothframe_tx_intrinsic_gasandcalldata_floor_gas. The legacy key set pays nothing. In the spec it is a newGasCosts.KEYED_NONCE_ACCESSterm in the frame transaction's intrinsic base, which the intrinsic cost and the calldata floor share. The framework prices it through a_frame_transaction_nonce_access_costhook thatEIP8250overrides, so the existing EIP-8250 tests, which price gas through the fork calculators, pick it up.The new
test_keyed_nonce_access_costbuilds its expected gas from the EIP's constant rather than from the calculators, for the legacy key set and for fresh, used, two and sixteen keys, with the standard cost binding and with the calldata floor binding.test_block_access_listnow also requires the written slots to be absent fromstorage_reads, with one and two keys, as the EIP's new block access list section states. The spec already recorded them that way.Filling the EIP-8250 tests without the spec change fails 77 fixtures, all 24 keyed
test_keyed_nonce_access_costcases among them. The filled fixtures pass against a client implementing #12457 (ethrex), and 31 fixture files fail against the same client without it.This changes frames-devnet-1 consensus:
tests-frames-devnet@v1.0.0pins EIP-8250 at 17669e9, before #12457, so it is meant for the next v1 minor release.Related Issues or PRs
frames-devnet@v1.0.0#3727.Checklist
just static<type>(<area>): <title>, where<type>and<area>come from an appropriateC-<type>, respectivelyA-<area>, label. The title should match the target squash commit message.Cute Animal Picture