Block access list changes - BAL construction, execution and validation by jwasinger · Pull Request #32263 · ethereum/go-ethereum

jwasinger · 2025-07-23T10:22:21Z

Change summary:

when --experimental.bal is enabled:
- insertion of post-Cancun blocks which don't contain access lists will cause them to be created during block execution and embedded into the block body.
- Insertion of post-cancun blocks that contain access lists will perform transaction execution and post-state root calculation in parallel using the state diffs provided in the access list. The state diffs produced by execution are validated against the entries in the BAL.
Introduced new modified blockchain suite test runner which first imports the test chain (generating BALs), and then re-inserts the access-list-containing blocks.

Deviations from EIP-7928:

Use RLP for the encoding format instead of SSZ.
Modify BAL format to accomodate pre/post-tx execution state changes:
- entries with txindex=0 correspond to state reads/changes which occur from system contract execution before transactions.
- entries with txindx=1..len(block.transactions) correspond to state reads/changes occurring for each transaction.
- entries with txindex=len(block.transactions)+1 correspond to the post-block system contracts and EIP-4895 withdrawals.

TODO:

validation of BAL account/storage reads (it's somewhat likely that these will be removed from the spec, so I'm holding off for now).

jwasinger · 2025-07-23T10:28:16Z

A lot of the comments in the code are straight-up wrong/misleading. There's a lot of notes/reminders I made as I was implementing this, and not all of them have been removed/corrected at this point.

jwasinger · 2025-07-29T13:43:23Z

I've pushed parallel execution changes here.

Still failing 2 tests (other than modexp repricing ones which will be fixed with a rebase on master):

--- FAIL: TestExecutionSpecBlocktestsBAL/prague/eip7251_consolidations/consolidations_during_fork/consolidation_requests_during_fork.json (0.03s)
        block_test.go:233: test with config {snapshotter:false, scheme:path} failed: mismatch between BAl value and computed value
    --- FAIL: TestExecutionSpecBlocktestsBAL/prague/eip7002_el_triggerable_withdrawals/withdrawal_requests_during_fork/withdrawal_requests_during_fork.json (0.02s)
        block_test.go:233: test with config {snapshotter:false, scheme:path} failed: mismatch between BAl value and computed value

I've been trying to debug these but it's proving to be exceedingly difficult with the parallel execution enabled. The tests seem to relate to behavior of some system contracts on the fork boundaries, so I will just proceed with gathering numbers on mainnet performance for the meantime.

jwasinger · 2025-08-04T13:25:30Z

INFO [08-04|13:21:15.227] Imported new chain segment number=23,051,616 hash=228379..13abd2 blocks=0 txs=10783 mgas=1137.280 elapsed=8.072s mgasps=140.878 age=2d6h46m triediffs=271.84MiB triedirty=112.48MiB

Broke the block count on the insertion log statement here. Will try to fix tomorrow.

jwasinger · 2025-08-13T12:02:49Z

Getting some empty accounts in the BAL. example:

 {
      address: "0xff00000000000000000000000000000000008453"
  }

lightclient

Looking good so far, just a few comments.

lightclient · 2025-08-21T20:07:02Z

+		var resCh chan *ProcessResultWithMetrics
+		resCh, err = bc.processor.ProcessWithAccessList(block, statedb, bc.cfg.VmConfig)
+		if err != nil {
+			// TODO: okay to pass nil here as execution result?


Should be fine, we just use it to get out the receipts (if it is non nil).

jwasinger · 2025-08-27T14:55:45Z

We decided on the EIP breakout call to keep account/storage reads in for now. Basically, without having this information it is possible that a transaction can be crafted which does a ton of storage reads and slow down the processing of the block significantly.

So until we can evaluate worst-cases here, account/storage reads are staying in the spec.

jwasinger · 2025-08-28T08:45:38Z

-			bc.prefetcher.Prefetch(block, throwaway, vmCfg, &interrupt)
+			if block.Body().AccessList == nil {
+				// only use the state prefetcher for non-BAL blocks.
+				bc.prefetcher.Prefetch(block, throwaway, vmCfg, &interrupt)


I previously implemented a state prefetcher for BALs. I deleted it so that the import benchmarks I am running will capture the BAL block processing speed equivalent to a synced node executing blocks at chain head.

We should probably add back the BAL prefetcher and measure the import performance.

jwasinger · 2025-08-28T09:36:13Z

Also TODO: need to implement validation of storage/account reads against the BAL.

jwasinger · 2025-08-28T10:34:34Z

 	// The tx context and all occurred logs in the scope of transaction.
 	thash   common.Hash
 	txIndex int
+	sender  common.Address


remove trhis.

gballet · 2025-08-28T10:34:37Z

 	// The tx context and all occurred logs in the scope of transaction.
 	thash   common.Hash
 	txIndex int
+	sender  common.Address


leftover, remove

gballet · 2025-08-28T10:38:28Z

this should not be committed

Yeah, it's leftover test data for a unit test that no longer exists (but should be updated and added back in).

fynnss · 2025-11-21T08:48:35Z

-		return nil, err
+		vstart := time.Now()
+		var err error
+		err = bc.validator.ValidateState(block, statedb, resWithMetrics.ProcessResult, false, false)


I would like to ask why validate state is needed here. The root has been calculated and compared during parallel processing.

It's because ValidateState performs some other checks beyond root comparison that we still need to do.

Got it, thanks.
Since the root is already computed and checked in parallel processing, should ValidateState still call statedb.IntermediateRoot()? Or should it only run the extra checks?

I thought about optionally-enabling the call to IntermediateRoot via a flag parameter to ValidateState but decided against it because it felt dirtier than just always calling IntermediateRoot and incurring the overhead of an extra hash check. Also, a call to IntermediateRoot caches the result on the statedb instance so calling it a second time is a effectively a no-op.

I have essentially rewritten the BAL code path for state root calculation in order to optimize it. I haven't yet pushed those changes to this branch because I am trying to sort out a bug in them.

…ontract' in the same transaction

…ding state, move readOnly call context check into gas func to avoid unecessary state reads in the gas handler in case where these are called in a static context.

…ate access if the component of the call price which is independent from the state is not sufficient to cover the provided gas.

This PR is an alternative to ethereum#32556. Instead of trying to be smart and reuse `geth init`, we can introduce a new flag `--genesis` that loads the `genesis.json` from file into the `Genesis` object in the same path that the other network flags currently work in. Question: is something like `--genesis` enough to start deprecating `geth init`? -- ```console $ geth --datadir data --hoodi .. INFO [10-06|22:37:11.202] - BPO2: @1762955544 .. $ geth --datadir data --genesis genesis.json .. INFO [10-06|22:37:27.988] - BPO2: @1862955544 .. ``` Pull the genesis [from the specs](https://raw.githubusercontent.com/eth-clients/hoodi/refs/heads/main/metadata/genesis.json) and modify one of the BPO timestamps to simulate a shadow fork. --------- Co-authored-by: rjl493456442 <garyrong0905@gmail.com>

…the Amsterdam tests aren't filled with bpo1/2 activated

…e number of CPUs

… cover the state access costs

…correct)

…oesn't contain one. remove the new pre/post-tx tracer hooks and use the system contract start/end hooks instead in the bal tracer

jwasinger · 2026-01-19T05:20:26Z

This got stale and I plan to propose a slightly different set of changes as an initial target for merging EIP-7928 support: we will merge a version that doesn't include the performance improvements initially.

jwasinger mentioned this pull request Jul 23, 2025

core/state: implement optional BAL construction in statedb #31959

Closed

jwasinger commented Aug 3, 2025

View reviewed changes

Comment thread core/types/bal/bal.go Outdated

jwasinger force-pushed the bal-execution branch 2 times, most recently from ab1c562 to 8f200db Compare August 4, 2025 08:51

jwasinger marked this pull request as ready for review August 4, 2025 10:13

jwasinger requested review from MariusVanDerWijden, fjl, gballet, lightclient, rjl493456442, s1na and zsfelfoldi as code owners August 4, 2025 10:13

jwasinger force-pushed the bal-execution branch from 869b300 to e14a463 Compare August 19, 2025 07:40

lightclient reviewed Aug 21, 2025

View reviewed changes

jwasinger commented Aug 28, 2025

View reviewed changes

gballet reviewed Aug 28, 2025

View reviewed changes

jwasinger force-pushed the bal-execution branch from 7d3b230 to 93b4210 Compare September 11, 2025 11:14

jwasinger mentioned this pull request Sep 11, 2025

all: generate and embed block-access-lists in block bodies when --experimentalbal is enabled #32586

Closed

jihoonsong mentioned this pull request Sep 19, 2025

feat: support amsterdam time ethpandaops/ethereum-package#1187

Closed

jwasinger force-pushed the bal-execution branch 3 times, most recently from 856a882 to fe61a7a Compare September 29, 2025 12:53

jwasinger added 3 commits November 12, 2025 15:16

add state prefetch timer

abd8612

cleanup bal state transition

74ca164

more cleanup

b3d8591

jwasinger force-pushed the bal-execution branch from 74ca164 to b24306e Compare November 12, 2025 17:45

correctly set the error in IntermediateRoot of BALStateTransition

2614e20

fynnss reviewed Nov 21, 2025

View reviewed changes

jwasinger and others added 20 commits November 22, 2025 16:33

minor refactor. instrument the bal tracer with some debug logging

159bbcd

fix triggered by a 'contract creation -> revert -> creation of same c…

2b003dc

…ontract' in the same transaction

remove logging statement

e364107

small refactors

6b6d3f2

core/vm: for selfdestruct/sstore whose gas funcs are dependent on rea…

0a8f3b0

…ding state, move readOnly call context check into gas func to avoid unecessary state reads in the gas handler in case where these are called in a static context.

core/vm: refactor call gas pricing such that Geth will not perform st…

f61df31

…ate access if the component of the call price which is independent from the state is not sufficient to cover the provided gas.

add method to log a state diff from two sets of AccountMutations

10a4b09

validate that accounts in the BAL are unique

1ef2b5d

update test release to bal experimental v1.8.0

fe72cf6

comment out BPO1/2 in amsterdam fork test config to make tests pass: …

85fc414

…the Amsterdam tests aren't filled with bpo1/2 activated

revamp BAL validation error logging

64558ff

perf experiment: limit the number of concurrently-executing txs to th…

898ff11

…e number of CPUs

perf experiment: fully concurrent origin storage loading

6f3e3de

wip: special storage key/val encoding

283c334

fix the last commit

c112198

only perform state reads in selfdestruct if we have sufficient gas to…

00136d1

… cover the state access costs

attempt to align behavior with other clients (spec tests currently in…

abaef48

…correct)

fix crash when attempting to call debug_getAccessList on block that d…

630a79e

…oesn't contain one. remove the new pre/post-tx tracer hooks and use the system contract start/end hooks instead in the bal tracer

attempt miner fix (untested)

158e44a

jwasinger force-pushed the bal-execution branch from 1c1ab63 to 158e44a Compare December 18, 2025 19:09

remove TODO that is no longer relevant

54d9e4f

jwasinger closed this Jan 19, 2026

Conversation

jwasinger commented Jul 23, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

jwasinger commented Jul 23, 2025

Uh oh!

jwasinger commented Jul 29, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

jwasinger commented Aug 4, 2025

Uh oh!

jwasinger commented Aug 13, 2025

Uh oh!

lightclient left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

jwasinger commented Aug 27, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

jwasinger commented Aug 28, 2025

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

jwasinger commented Jan 19, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

jwasinger commented Jul 23, 2025 •

edited

Loading

jwasinger commented Jul 29, 2025 •

edited

Loading

jwasinger commented Aug 27, 2025 •

edited

Loading