Skip to content

Bump reth to v2.11.0 - #95

Merged
sduchesneau merged 101 commits into
release/reth-2.xfrom
bump/reth-2.11.0
Nov 11, 2026
Merged

sduchesneau merged 101 commits into
release/reth-2.xfrom
bump/reth-2.11.0

Conversation

@dfuse-bot

@dfuse-bot dfuse-bot commented Nov 11, 2026 •

Copy link
Copy Markdown

Merges upstream reth v2.11.0 into release/reth-2.x. Merging this PR releases reth-v2.11.0-fh3.1 (the release workflow tags the merge commit from the CHANGELOG heading).

What changed

  • Upstream merge. Conflicts: Cargo.toml (version, workspace members, evm2 [patch]), crates/engine/tree/Cargo.toml, crates/evm/evm/src/execute.rs (our set_inspect_system_calls next to upstream's new with_inspector/take_inspector), Cargo.lock (upstream's, plus our entries), and docs/vocs (upstream's). Upstream dropped the inkwell patch, and so does this PR.
  • evm2 is no longer patched to the streamingfast/evm2 fork. Our only evm2 patch, opt-in inspection of system calls, went upstream in feat(evm): add opt-in inspection of system calls alloy-rs/evm2#888, so the fork had the same tree as upstream 66687c18. The evm2 [patch] block and the evm2 entries in Cargo.lock are now identical to upstream reth v2.11.0's.
  • Firehose adaptations to the evm2 API changes: call input is read from the caller's memory (MessageInput::as_bytes), initial_call_depth moved from Version to Evm, and BalWorkerError uses AnyError. The Firehose BAL tests and bench use Address::from_raw_public_key for secp256k1 0.33. The new upstream e2e test executor_replay gets the FirehoseLiveHooks impl every engine EVM config needs.
  • Fix: evm2 narrowed InstrStop::is_revert() to the REVERT opcode. A call that fails before running (insufficient balance, call depth, nonce overflow) was then traced as consuming its whole gas limit instead of 0, on every fork. The inspector uses is_frame_revert() now, and a regression test covers it. Battlefield caught it.

Effect on the Firehose block

  • Every block from Prague on: the EIP-4788 beacon-root system call now comes before the EIP-2935 block-hash system call, the order evm2's block executor runs them in (and go-ethereum's). Same calls and content; their order and ordinals swap. Transactions are unchanged.
  • Amsterdam only (Sepolia): a transaction that runs out of gas in its runtime gas phase now opens its root call before the sender's nonce change, as on go-ethereum. Only those two ordinals swap.
  • No hardfork is added, scheduled or changed: alloy-hardforks stays at 0.4.11. Glamsterdam (Amsterdam) is still unscheduled on mainnet and Hoodi, and it has been active on Sepolia since block 11856337. ACTIVATIONS.md is refreshed.
  • The block model does not change, so the tag stays fh3.1.

Firehose bug patterns

  1. Serial order: the new parallel EIP-7702 authority recovery (perf(evm): parallelize EIP-7702 authorization recovery paradigmxyz/reth#28307) runs inside one transaction's conversion, before execution, and par_iter().collect() keeps the order. It never calls the tracer. The BAL parallel path is unchanged apart from its error type. The new RPC replay through the block executor (feat(rpc): replay block transactions through the block executor paradigmxyz/reth#28266) uses the inner block executor factory; FirehoseEvmConfig only wraps batch_executor, so RPC replays never reach the tracer. JIT stays refused while tracing.
  2. Emit after acceptance: unchanged. FirehoseBlockExecutor still holds the block until the next block or into_state(). Upstream adds no post-execution check after that point.
  3. Activations: see above. Request system calls run without a code check again (fix(evm): bump evm2 to run request system calls without a code check paradigmxyz/reth#28311), which matches what v2.10.0 did.
  4. Unmapped fields: alloy is still 2.5.0 and no consensus type gained a field. evm2's InstrStop gained Reload, which no frame reports.

Open questions

  • Tag name. The newest tag on release/reth-2.x is reth-v2.9.0-fh3.1-not-a-real-version-alpha, a test of the release workflow. Applying the rule to it literally would give reth-v2.11.0-fh3.1-not-a-real-version-alpha. This PR uses reth-v2.11.0-fh3.1, following the last real release reth-v2.10.0-fh3.1.
  • Golden changes. The four Amsterdam goldens in crates/firehose-tests changed semantically (the orders described above). Please confirm.
  • Battlefield snapshot. reth-dev's deploys/contract_fail_code_copy snapshot needs updating for the Amsterdam ordinal swap once this ships.
  • Dependency table. The release tracker table lists evm2 sf/reth; that row no longer applies, since this release uses upstream evm2. evm-firehose-tracer-rs stays pinned at d35edd15, a commit on main, until firehose-tracer 5.8.0 ships.

Test Execution Summary

Build Status

  • cargo check --workspace --all-targets (reth, bump/reth-2.11.0 at f557f816d): PASSED
  • cargo build --release --bin reth: PASSED

LLVM 23 (needed by evm2-jit through llvm-sys 231) was installed with the repository's own .github/scripts/install_llvm_ubuntu.sh 23, because the build image only ships LLVM 14 and 15.

Unit Tests

Package Status Notes
cargo test -p reth-firehose PASSED 81 tests, including the new call_failed_on_insufficient_balance_consumes_no_gas
cargo test -p reth-firehose-tests PASSED 9 tests. 4 Amsterdam goldens regenerated, see "Golden file changes"
cargo test -p reth-engine-tree --lib --tests PASSED 240 lib tests (27 BAL ones, including the Firehose BAL path), 16 e2e-testsuite, firehose_live_tracing, firehose_live_tracing_amsterdam, firehose_live_tracing_prewarm_reuse
cargo test -p reth-evm -p reth-evm-ethereum --lib --tests PASSED 21 + 26 + 15 tests
cargo test -p reth-firehose -p reth-firehose-tests at 961bf9b5b (upstream evm2) PASSED 90 tests. The other runs above used the streamingfast/evm2 fork, whose tree is byte-identical to upstream 66687c18

Golden file changes (semantic, need a reviewer's OK)

Four crates/firehose-tests Amsterdam goldens changed. Each change was checked to be only a reordering:

  1. amsterdam_slot_number, amsterdam_block_access_list, amsterdam_builder_execution_requests, amsterdam_create_runtime_out_of_gas: the two pre-block system calls swap places. evm2's block executor, which reth v2.11.0 now runs Ethereum blocks on, calls EIP-4788 (beacon roots) before EIP-2935 (block hash history). reth v2.10.0 called them the other way round. go-ethereum uses the new order. Same calls, same content; only their order and their ordinals change. Transactions are unchanged.
  2. amsterdam_create_runtime_out_of_gas only: the root CREATE call now begins (ordinal 7) before the sender's nonce change (ordinal 8); it was the reverse. evm2 now fires the depth-0 frame hooks for a transaction that runs out of gas in its runtime gas phase, as go-ethereum's haltTopFrame does, so the tracer no longer synthesizes that root call. Status, gas and failure reason are unchanged.

Integration Tests

  • Status: NOT RUN (no integration suite beyond the tests above for this fork)

Battlefield Tests

  • Status: PASSED on reth-devnet; on reth-dev, 1 expected diff (snapshot needs updating, see below)
  • reth-dev (Amsterdam from genesis, as on Sepolia): 93 passing, 1 failing, 43 pending. Compare blocks passed.
    1. Deploys / Contract fail not enough gas after code_copy (deploys/contract_fail_code_copy): the root CREATE beginOrdinal 3 → 2 and the sender nonce change ordinal 2 → 3. This is the same evm2 change as golden change 2 above. It is an expected diff: every other chain's snapshot of this test, geth included, puts the call's begin before the nonce change. The reth-dev snapshot needs SNAPSHOTS_UPDATE="^deploys/contract_fail_code_copy$" once this release is merged.
  • reth-devnet (Prague, as on mainnet and Hoodi): 84 passing, 0 failing, 54 pending. Compare blocks passed.
  • The first reth-dev run also failed calls/nested_fail_with_native_transfer and suicide/create_contract_kill_it_and_try_to_call_within_same_call. A call that fails on insufficient balance reported its whole gas limit as consumed instead of 0. Cause: evm2 narrowed InstrStop::is_revert() to the REVERT opcode and moved the old meaning to is_frame_revert(). This affected every fork, Prague included. It is fixed in f557f816d with a regression test, and both tests pass in the run above.
  • The Battlefield system-call snapshots do not cover system call order, so the swap in golden change 1 does not show there.

Commands (from a streamingfast/battlefield-ethereum checkout at ccbad7c, fireeth v2.24.1):

pnpm install
# reth-dev
RETH_BINARY=<reth>/target/release/reth ./scripts/run_firehose_reth_dev.sh   # terminal 1
pnpm test:fh3.0:reth-dev                                                    # terminal 2
# reth-devnet
./scripts/ethereum_devnet/run_playground_devnet.sh                          # terminal 1
RETH_BINARY=<reth>/target/release/reth ./scripts/run_firehose_reth_devnet.sh # terminal 2
pnpm test:fh3.0:reth-devnet                                                 # terminal 3

CHANGELOG

## reth-v2.11.0-fh3.1

### Changed

- Update to upstream reth [v2.11.0](https://github.com/paradigmxyz/reth/releases/tag/v2.11.0).
  evm2 now comes from the upstream revision reth pins instead of the `streamingfast/evm2` fork,
  because its only patch is upstream. No hardfork is scheduled or changed.
- From Prague on, the EIP-4788 beacon root system call is traced before the EIP-2935 block hash
  system call, the order evm2's block executor runs them in. Their content does not change.
- On Amsterdam, a transaction that runs out of gas in its runtime gas phase opens its root call
  before the sender's nonce change, as on go-ethereum. The block content does not change, only
  the order of those two ordinals.

pepyakin and others added 30 commits November 5, 2026 03:11
Co-authored-by: Matthias Seitz <matthias.seitz@outlook.de>
Co-authored-by: Matthias Seitz <19890894+mattsse@users.noreply.github.com>
Co-authored-by: Matthias Seitz <19890894+mattsse@users.noreply.github.com>
Co-authored-by: Matthias Seitz <19890894+mattsse@users.noreply.github.com>
Co-authored-by: Matthias Seitz <matthias.seitz@outlook.de>
Make DNS entry parsing more consistent and extend coverage for edge cases.
…z#28312)

Checks for cancellation between calls in `eth_simulateV1`, as `eth_callMany` and the trace methods already do since paradigmxyz#27462.
Co-authored-by: Matthias Seitz <19890894+mattsse@users.noreply.github.com>
…8307)

Co-authored-by: Matthias Seitz <19890894+mattsse@users.noreply.github.com>
lean-apple and others added 21 commits November 10, 2026 14:58
…gmxyz#28254)

Co-authored-by: Matthias Seitz <19890894+mattsse@users.noreply.github.com>
Co-authored-by: Matthias Seitz <19890894+mattsse@users.noreply.github.com>
…d space (paradigmxyz#28414)

Co-authored-by: Matthias Seitz <19890894+mattsse@users.noreply.github.com>
…xyz#28424)

Co-authored-by: Matthias Seitz <19890894+mattsse@users.noreply.github.com>
…8423)

Co-authored-by: Matthias Seitz <19890894+mattsse@users.noreply.github.com>
Co-authored-by: Soubhik Singha Mahapatra <soubhiksmp2004@gmail.com>
Co-authored-by: Ravan Mammadov <95362051+Ravan002@users.noreply.github.com>
Co-authored-by: Matthias Seitz <matthias.seitz@outlook.de>
Co-authored-by: Matthias Seitz <19890894+mattsse@users.noreply.github.com>
# Conflicts:
#	Cargo.lock
#	Cargo.toml
#	crates/engine/tree/Cargo.toml
#	crates/evm/evm/src/execute.rs
#	docs/vocs/bun.lock
#	docs/vocs/package.json
#	docs/vocs/vocs.config.ts
Call inputs are read from the caller's memory and the initial call depth moved from Version to Evm.
evm2 runs the EIP-4788 system call before EIP-2935 and opens the root frame
of a transaction that runs out of gas in its runtime gas phase.
evm2 narrowed InstrStop::is_revert to the REVERT opcode. Insufficient balance,
call depth and nonce overflow failures return their gas, now is_frame_revert.
@github-actions

github-actions Bot commented Nov 11, 2026 •

Copy link
Copy Markdown

Merging this PR will release reth-v2.11.0-fh3.1

  • Creates the tag reth-v2.11.0-fh3.1 on the merge commit
Release notes

Changed

  • Update to upstream reth v2.11.0.
    evm2 now comes from the upstream revision reth pins instead of the streamingfast/evm2 fork,
    because its only patch is upstream. No hardfork is scheduled or changed.
  • From Prague on, the EIP-4788 beacon root system call is traced before the EIP-2935 block hash
    system call, the order evm2's block executor runs them in. Their content does not change.
  • On Amsterdam, a transaction that runs out of gas in its runtime gas phase opens its root call
    before the sender's nonce change, as on go-ethereum. The block content does not change, only
    the order of those two ordinals.

Our only evm2 patch is upstream now, so the fork had the same tree as alloy-rs/evm2@66687c18.
@sduchesneau

Copy link
Copy Markdown

@maoueh FYI:
this comes with 2 more "ordinal flips" like we saw when we went from geth to reth.

@sduchesneau
sduchesneau merged commit 3e576a6 into release/reth-2.x Nov 11, 2026
5 checks passed
@sduchesneau
sduchesneau deleted the bump/reth-2.11.0 branch November 11, 2026 14:37
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.