core, core/vm, params: adopt the EIP-2780 and EIP-8037 spec changes (#35318)

[pratikspatil024]

Important

Status, 2026-10-01: #2345, #2346 and #2354 have merged, so this PR now targets upstream-merge-v1.17.4 directly. It is no longer part of a GitHub stack: merging #2354 made GitHub rebase this branch onto the base automatically (754d732aa → ada1fc786), so the stack was removed. The content is unchanged except that the switch-dispatch generator went back to #2346's chargeRegular/chargeState form, which charges gas identically. Merge with a merge commit. Don't use "Update branch" with rebase, or any rebase. The notes below describe the earlier stacked layout.

Stacked on #2354 (base ppatil-upstream-v1.17.5), not develop. Part of the go-ethereum upstream sync (POS-2549), which merges into upstream-merge-v1.17.4 and reaches develop once, at the end.

Merge with a merge commit. Never squash. Squashing rewrites this branch's SHAs and breaks every PR stacked above it.

Review bottom-up: #2308, #2319, #2325, #2328, #2337, #2340, #2341, #2342 and #2343 have merged into the base. The open stack is #2345 → #2346 → #2354 → #2455 → #2461. Start at #2345.

Review fixes and cascade, 2026-09-29: three findings from Marcello's review were fixed where the code lives and cascaded up. #2346 f2fde2692: the switch-dispatch interpreter (runSwitch, on by default on mainnet BPs) now charges EIP-8037 state gas like Run, with a new Amsterdam differential test. #2354 0166dcc9d: the pebble v1 path, which every existing bor database opens through, keeps bor's tuning and metrics (new databases use pebble v2; upgrade command tracked in POS-3746). #2354 aac7dd01e: V2's FinaliseFast settles EIP-8246 self-destructs like serial Finalise. Cascade: #2354 256e10c39 → #2455 d96fcc81b → #2461 440ed1ac4; at #2455 and #2461 the generator mirrors that level's renamed Run helpers. All dormant until Amsterdam except the pebble fix, which restores today's behaviour for existing nodes. Every hop builds and passes the dispatch/V2/parity tests, and the witness regeneration tests pass on the top of the chain.

Review follow-up and cascade, 2026-09-30: per Jerry's review on #2346 (49aa12b54), both BlockSTM processors run their pre-execution system calls through the serial PreExecution instead of inline copies (no behaviour change there). Merged up into #2354 (ab0921f62), that closes a dormant gap: #35285 had moved the EIP-7997 factory insert (first Amsterdam block) into PreExecution, which V1/V2 didn't call, so they would have disagreed with serial on the root at activation; TestV2PreExecEIP7997Activation covers it. Cascade: #2354 ab0921f62 → #2455 c27e0f29a → #2461 caeadc89b, where upstream's #35458 drops the insert altogether and the test goes with it. Every hop builds and passes the parity/V2 tests, and the witness regeneration tests pass on the top of the chain.

Cascade, 2026-09-28: #2342 merged into the base (cd9097f32) and was cascaded up the stack: #2343 663809636 → #2345 4aa9ff8f0 → #2346 8971e231a → #2354 778958987 → #2455 f1c5c8571. It brings only #2342's final changes (develop's otel 1.45 go.mod/go.sum bump, the nightly-govulncheck workflow and one eth/70 receipt test). Every hop test-compiles, and the witness regeneration tests pass on the top of the chain.

2026-09-29: #2343 merged into the base (cc07116a3, a two-parent merge whose tree matches 663809636), so nothing new cascades from the base. Review fixes since: #2354 64709a22f (Finalise errTerminated log filter, override-flag help text), merged up into #2455 as c4922a57b; #2455 855630f34 (EIP-8037 test configs run without Bor forks, block-based Amsterdam entry in tests.Forks). The v1.17.6 sync, #2461, now sits on top of #2455, so the release covers v1.17.4 + v1.17.5 + v1.17.6.

Summary

Adopts upstream #35318 (EIP-2780 and EIP-8037 spec changes, upstream 1ef0ffb98). It was deferred out of the v1.17.5 sync with the recorded decision "take as its own stacked PR". It has to land before the v1.17.6 sync, because v1.17.6's gas changes (#35454, #35457, #35486, #35497) are written against it.

This changes nothing on Bor's networks. Everything #35318 touches is gated on Amsterdam, and AmsterdamBlock is nil on every Bor preset.

This is a hand-written port, not a merge. #35318's history is already in the tree from v1.17.5, and its content was reverted there, so the diff was replayed 3-way onto Bor's files.

What upstream changes

• IntrinsicGas returns uint64 and drops costPerStateByte. State gas is now charged at runtime, not as part of intrinsic gas.
• The block gas-pool reservation and the EIP-3860 init-code size check move from buyGas into preCheck.
• A new initRuntimeGasBudget sets up the running budget after the intrinsic charge.
• The top-level frames move into executeCreate and executeCall.
• EVM.Create and Create2 no longer return a creation flag. The account-creation state charge moves into chargeAccountCreation, called from opCreate and opCreate2.

Kept from Bor (the three items the v1.17.5 ledger required)

1. The Madhugiri arm of the EIP-7825 per-transaction gas cap. Upstream's replacement is !rules.IsAmsterdam && rules.IsOsaka, which would drop the cap Bor has had live since Madhugiri. The port keeps !rules.IsAmsterdam && (rules.IsOsaka || isMadhugiri).
2. Block-based, one-argument fork checks.
3. Bor's fee handling in execute. That's the burn to the burnt contract, the tip, and the fee transfer log, including its pre-gas-purchase balance snapshots. effectiveTip stays on signed big.Int, and upstream's NoBaseFee skip stays commented out, matching the existing TODO(raneet10).

Also:

• #35396 (tracer panic fix): its state_transition.go half is applied here. It fell with the deferral; its core/tracing/hooks.go half was already in.
• eth/tracers/parity.go, which is Bor-only, follows the new IntrinsicGas signature.
• Constants: EpochDuration, ExpByteGas, SloadGas and TierStepGas are removed as upstream did. None had a user in Bor.

Where to spend review time

File	What to check
core/state_transition.go	The port itself. preCheck: the EIP-7825 cap with the Madhugiri arm. execute: Bor's fee block and input1/input2, which must be captured before preCheck buys gas.
core/vm/evm.go, core/vm/instructions.go	createFramePreCheck / chargeAccountCreation. Before Amsterdam, create() runs the same checks in the same order.
core/vm/gas_table.go, core/vm/operations_acl.go	Every hunk is in an Amsterdam-only *8037 function. The PIP-88 twins are untouched.
core/txpool/validation.go	The mechanical IntrinsicGas change; Bor's txpool gates are unchanged.

Provenance

Each ported file was diffed against upstream v1.17.5, not just against the #35318 commit. Every remaining difference is an existing Bor divergence:

• Bor's MaxTxGas (2^25) and base-fee parameters;
• the declined stack arena;
• Bor's checkMaxCodeSize (the Ahmedabad cap);
• Bor's fee handling and fork-check shape.

No other upstream change was lost with the deferral.

Restored tests

The upstream EIP tests dropped in v1.17.5 are restored from upstream v1.17.5 and converted to Bor's block-based Amsterdam gate (AmsterdamTime becomes AmsterdamBlock).

• Restored, all passing:
  • core/eip8037_test.go
  • core/eip2780_test.go
  • core/eip7997_test.go
  • core/eip8038_test.go
  • core/vm/eip8037_test.go
• core/eip_testenv_test.go holds the part of upstream's eip7928_test.go test environment that these tests use. Remove it when eip7928_test.go is restored (POS-3737).
• Adaptations for Bor's MaxTxGas (2^25 against upstream's 2^24): TestEIP2780RuntimeOOGRevertsDelegations/with-reservoir uses 200 authorizations instead of 100, and TestValidationFloorCostCap uses 600,000 bytes of calldata instead of 300,000.
• The EIP-8037 test configs clear Bor (855630f). With Bor's forks, NewEVM selects Chicago's instruction set ahead of Amsterdam's, so the opcode tests would otherwise skip or pass without exercising EIP-8037. All of them now run and pass.
• tests.Forks has a block-based Amsterdam entry, so transaction fixtures with an Amsterdam result are validated.

Held back, and why this matters for enabling Amsterdam (POS-3738):

• The fork-ordering blocker itself: on a real Bor config, Chicago's instruction set shadows Amsterdam's, so EIP-8037/8038 opcode pricing wouldn't activate. The tests above run with Bor = nil for that reason.
• core/vm/eip8038_test.go is held back for the same reason.
• core/eip7708_test.go is held back because Bor's legacy transfer log is emitted alongside EIP-7708's.
• core/eip8246_test.go is held back because it fails at chain creation on Bor.

POS-3738 tracks all of these as blockers for scheduling Amsterdam. docs/upstream-merges/v1.17.5/needs-wiring.md is updated row by row.

Executed tests

• go build ./... is clean.
• go test -run XXXNOMATCHXXX ./... compiles every test package.
• go vet on core, core/vm and eth/tracers shows only the known pre-existing lock-copy finding. gofmt is clean.
• go test -short passes on core, core/vm/..., core/txpool/..., core/tracing/..., eth/tracers/..., internal/ethapi/..., tests, miner and params.
  • cmd/evm fails TestT8n, TestEVMTracing, TestEvmRun and TestEvmRunRegEx with the same failure messages as the unmodified core/vm, params, eth: merge geth v1.17.5 (v1.17.5 sync) #2354 tip. That's the known pre-existing set; CI excludes cmd/.
• Mainnet replay, TestV2BlockSTMAllBlocks (BOR_BLOCKSTM_TEST=1): identical results to the unmodified core/vm, params, eth: merge geth v1.17.5 (v1.17.5 sync) #2354 tip. 238/241 blocks are consistent between serial and BlockSTM V2 execution on both trees. The same three blocks fail on both with "code is not found": fixture data the test fetches from an RPC endpoint (ALCHEMY_URL) when available. None of the failures is a state-root mismatch.
• TestV2WitnessRegeneration{AllBlocks,PipelinedSRCAllBlocks,PipelinedSRCChained} pass in non-short mode. Chained covers 222/222 consecutive mainnet pairs.

Rollout notes

• Not consensus-affecting as merged. Every behaviour change is behind IsAmsterdam, which is nil on every Bor preset. Before Amsterdam, the EIP-7825 cap, the intrinsic gas values and the fee handling are unchanged. The replay above is the empirical check.
• RPC-visible: error ordering in preCheck. The gas-pool check and the init-code size check now run before the balance check, as upstream has them. A transaction that fails several checks may report a different error first. Gas-tracer events keep the same sequence, because #35396 re-emits the initial-balance event.
• No coordinated upgrade is needed, and there's no operator-facing change.

🤖 Generated with Claude Code

[pratikspatil024]

@claude review

[Copilot]

Copilot review overview

🟡 Changes recommended

Amsterdam runtime creation bypasses required checks, and parts of the restored test harness do not exercise the intended fork behavior.

Review effort: Balanced
Findings: 2

Open (2)

• Bor fork ordering prevents Amsterdam pricing tests · New
• Test Forks lacks Amsterdam configuration · New

What changed in this PR

Adopts updated EIP-2780/EIP-8037 gas accounting behind the dormant Amsterdam fork.

Changes:

• Moves state-dependent gas charges from intrinsic gas to runtime execution.
• Refactors CREATE/CALL gas accounting and tracing.
• Restores Amsterdam EIP tests and fixture support.

File	Description
cmd/​evm/​internal/​t8ntool/​transaction.go	Adapts intrinsic-gas API.
core/​bench_test.go	Updates benchmark gas calculation.
core/​eip2780_test.go	Adds EIP-2780 coverage.
core/​eip7997_test.go	Adds factory-transition tests.
core/​eip8037_test.go	Adds state-transition gas tests.
core/​eip8038_test.go	Adds authorization gas tests.
core/​eip_testenv_test.go	Supplies shared EIP test fixtures.
core/​error.go	Adds runtime out-of-gas error.
core/​state_transition.go	Implements runtime gas charging.
core/​state_transition_test.go	Updates intrinsic-gas expectations.
core/​tracing/​hooks.go	Adds gas-change reasons.
core/​tracing/​gen_gas_change_reason_stringer.go	Regenerates reason strings.
core/​txpool/​validation.go	Adapts transaction validation.
core/​vm/​common.go	Minor formatting cleanup.
core/​vm/​eip8037_test.go	Adds VM gas-accounting tests.
core/​vm/​evm.go	Refactors contract creation.
core/​vm/​gas_table.go	Moves creation charges to runtime.
core/​vm/​gascosts.go	Updates gas-budget operations.
core/​vm/​instructions.go	Charges conditional creation costs.
core/​vm/​interface.go	Minor interface formatting.
core/​vm/​interpreter.go	Uses exported regular-gas charging.
core/​vm/​operations_acl.go	Applies Amsterdam warm-access pricing.
core/​vm/​runtime/​runtime.go	Adapts the CREATE return signature.
docs/​upstream-merges/​v1.17.5/​needs-wiring.md	Records adoption and remaining blockers.
eth/​tracers/​parity.go	Adapts parity intrinsic-gas tracing.
params/​protocol_params.go	Updates Amsterdam gas constants.
tests/​gen_sttransaction.go	Supports arbitrary-precision fixture nonces.
tests/​state_test_util.go	Validates fixture nonces against EIP-2681.
tests/​transaction_test_util.go	Adds Osaka and Amsterdam cases.

Files not reviewed (2)

• core/tracing/gen_gas_change_reason_stringer.go: Generated file
• tests/gen_sttransaction.go: Generated file

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[claude]

Nothing blocking. The comments below are optional suggestions. There is no need to push a fix for them before merging.

[codecov]

Codecov Report

❌ Patch coverage is 80.11204% with 71 lines in your changes missing coverage. Please review.
✅ Project coverage is 56.42%. Comparing base (679b805) to head (ada1fc7).
⚠️ Report is 4 commits behind head on upstream-merge-v1.17.4.

Files with missing lines	Patch %	Lines
core/state_transition.go	88.36%	17 Missing and 10 partials ⚠️
core/vm/evm.go	69.04%	11 Missing and 2 partials ⚠️
tests/state_test_util.go	0.00%	10 Missing ⚠️
core/txpool/validation.go	20.00%	3 Missing and 1 partial ⚠️
core/vm/instructions.go	83.33%	2 Missing and 2 partials ⚠️
tests/gen_sttransaction.go	0.00%	4 Missing ⚠️
tests/transaction_test_util.go	0.00%	4 Missing ⚠️
core/vm/gascosts.go	25.00%	3 Missing ⚠️
core/tracing/gen_gas_change_reason_stringer.go	0.00%	1 Missing ⚠️
core/vm/operations_acl.go	0.00%	1 Missing ⚠️

❌ Your patch check has failed because the patch coverage (80.11%) is below the target coverage (90.00%). You can increase the patch coverage or adjust the target coverage.

Additional details and impacted files

@@                    Coverage Diff                     @@
##           upstream-merge-v1.17.4    #2455      +/-   ##
==========================================================
+ Coverage                   56.25%   56.42%   +0.16%     
==========================================================
  Files                         944      944              
  Lines                      174742   174836      +94     
==========================================================
+ Hits                        98295    98645     +350     
+ Misses                      70396    70158     -238     
+ Partials                     6051     6033      -18     

Files with missing lines	Coverage Δ
core/tracing/hooks.go	0.00% <ø> (ø)
core/vm/common.go	59.67% <ø> (ø)
core/vm/gas_table.go	44.15% <100.00%> (+9.47%)	⬆️
core/vm/interpreter.go	65.89% <100.00%> (ø)
core/vm/runtime/runtime.go	83.89% <100.00%> (ø)
eth/tracers/parity.go	81.40% <100.00%> (ø)
params/protocol_params.go	84.21% <ø> (ø)
tests/init.go	0.00% <ø> (ø)
core/tracing/gen_gas_change_reason_stringer.go	0.00% <0.00%> (ø)
core/vm/operations_acl.go	46.44% <0.00%> (+2.13%)	⬆️
... and 8 more

... and 25 files with indirect coverage changes

Files with missing lines	Coverage Δ
core/tracing/hooks.go	0.00% <ø> (ø)
core/vm/common.go	59.67% <ø> (ø)
core/vm/gas_table.go	44.15% <100.00%> (+9.47%)	⬆️
core/vm/interpreter.go	65.89% <100.00%> (ø)
core/vm/runtime/runtime.go	83.89% <100.00%> (ø)
eth/tracers/parity.go	81.40% <100.00%> (ø)
params/protocol_params.go	84.21% <ø> (ø)
tests/init.go	0.00% <ø> (ø)
core/tracing/gen_gas_change_reason_stringer.go	0.00% <0.00%> (ø)
core/vm/operations_acl.go	46.44% <0.00%> (+2.13%)	⬆️
... and 8 more

... and 25 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[claude]

Claude Code Review

This repository is configured for manual code reviews. Comment @claude review for a one-time review, or @claude review always to subscribe this PR to a review on every future push.

_{Tip: disable this comment in your organization's Code Review settings.}

[lucca30]

Hi Pratik, approving this one. I compared the port with upstream v1.17.5: state_transition.go only has our known differences (Madhugiri cap arm, fee burn/tip/log, block based fork checks), and gas_table.go and gascosts.go are identical. Before Amsterdam, IntrinsicGas, the create() prechecks and the gas budget are the same as before. The new check order in preCheck changes only which error comes first, and miner/BlockSTM don't depend on the error type, so no impact there.

One small ask: the 2 claude[bot] points (no tracer frame when the CREATE precheck fails, and runtime.Create skipping the precheck under Amsterdam) are upstream verbatim, ok. But can you add them to POS-3738 / needs-wiring so we check them again when we schedule Amsterdam?

Nit: Forks["Amsterdam"] in tests/init.go has no BPO1/BPO2 blob config like upstream. Fine if it is on purpose (same as our Osaka), right?

[pratikspatil024]

@lucca30 thanks for the review.

• The two claude[bot] points: both are now needs-wiring rows pointing at POS-3738 (754d732), with the review links, and I added them to POS-3738 so we recheck them when Amsterdam is scheduled. On the runtime.Create one, I confirmed block processing is unaffected: the state transition checks CanTransfer and the nonce ceiling before its evm.Create, and vm/runtime.Create is the only other direct caller.
• Forks["Amsterdam"] without BPO1/BPO2: yes, on purpose, same as our Osaka entry. Bor's ChainConfig has no BPO forks at all (no BPO1Block or IsBPO*), so there's nothing to hang upstream's BPO blob entries on.