feat: add EIP-8037 / TIP-1016 state gas support by rakita · Pull Request #3406 · bluealloy/revm

rakita · 2026-02-06T23:15:38Z

Summary

Implements EIP-8037 (TIP-1016) dual-limit gas accounting, separating execution gas (CPU work) from state gas (storage/account creation, code deposit). This introduces a reservoir model where state gas draws from a separate pool before spilling into regular gas.

Key Changes

Gas struct (interpreter/gas.rs): Added reservoir, regular_gas_budget, state_gas_spent fields with record_state_cost() that draws from reservoir first, then spills to regular gas. gas_left() excludes reservoir per EIP-8037.
GasParams (context/interface/cfg/gas_params.rs): New state gas cost IDs — sstore_set_state_gas, new_account_state_gas, code_deposit_state_gas, create_state_gas.
Charging sites (5 locations):
- SSTORE — new slot creation (host.rs)
- CREATE/CREATE2 — contract creation (contract.rs)
- Code deposit — per-byte code storage (frame.rs)
- CALL — new account with value transfer (call_helpers.rs)
- SELFDESTRUCT — new beneficiary account (host.rs)
Frame propagation (handler/frame.rs): Child frame state gas accumulates to parent on success; reservoir refill on halt/revert prevents state gas from permanently consuming regular gas budget.
Precompiles: All precompile gas charges use record_regular_cost() — no state gas for precompiles.
GAS opcode (system.rs): Returns remaining - reservoir per spec.
op-revm: Updated to handle new gas propagation in last_frame_result.
Inspector/EIP-3155: Logs state gas fields for tracing.

Architecture

Component	Details
Gas struct	`remaining` = `gas_left` + `reservoir`; `regular_gas_budget` is implicit cap for regular gas
Regular gas	Checked against `gas_left` only via `record_cost()`, cannot draw from reservoir
State gas	Drawn from `reservoir` first via `record_state_cost()`, spills to `gas_left` when exhausted
GAS opcode	Returns `gas_left()` only, excluding the state gas reservoir
Child frames	Inherit parent's `regular_gas_budget` and `reservoir` directly
EIP-8037	Higher gas costs for state-creating ops, auto-enabled for Amsterdam+ specs
TIP-1016	Dual-limit model with separate state gas reservoir, opt-in via `CfgEnv`

Cfg Integration

Cfg::state_gas_ids() returns Option<GasParams> — None disables state gas (backward compatible).
CfgEnv stores state gas config behind #[cfg(feature = "optional_state_gas_params")].

Test Suite

35+ test vectors in crates/ee-tests/src/eip8037.rs covering 7 categories:

SSTORE state gas — new slot, overwrite, zero-to-zero, multiple slots
CREATE/CREATE2 state gas — empty code, with code, with SSTORE, large code, OOG
CALL state gas — new/existing account, with/without value, SELFDESTRUCT
Regular gas cap enforcement — OOG, sufficient, reservoir interaction, block gas limit
Cross-frame propagation — child success/revert/halt, nested 3+ levels
Interactions — refund, GAS opcode, spend_all, precompiles, state_gas_spent tracking
Reservoir refill — revert vs halt behavior, boundary cases

Test Plan

cargo build --workspace passes
cargo nextest run --workspace — all tests pass
cargo nextest run -p ee-tests — all 35+ EIP-8037 tests pass
cargo clippy --workspace --all-targets --all-features — clean
cargo fmt --all — clean
no_std compatibility check
Ethereum state tests via revme statetest

codspeed-hq · 2026-02-06T23:22:13Z

Merging this PR will degrade performance by 5.58%

⚠️

Different runtime environments detected

Some benchmarks with significant performance changes were compared across different runtime environments,
which may affect the accuracy of the results.

Open the report in CodSpeed to investigate

⚡ 3 improved benchmarks
❌ 17 regressed benchmarks
✅ 157 untouched benchmarks

⚠️ Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Benchmark	`BASE`	`HEAD`	Efficiency
❌	`JUMP_50`	17.9 µs	18.6 µs	-3.34%
❌	`GASLIMIT_50`	18.4 µs	19 µs	-3.11%
❌	`DELEGATECALL_50`	86.6 µs	90 µs	-3.79%
❌	`GAS_50`	18.3 µs	18.9 µs	-3.12%
❌	`JUMPDEST_50`	15.6 µs	16.3 µs	-4.33%
❌	`MCOPY_50`	20.8 µs	21.5 µs	-3.01%
❌	`MSIZE_50`	18.4 µs	19 µs	-3.11%
❌	`MSTORE_50`	19.7 µs	20.4 µs	-3.31%
❌	`PUSH0_50`	18.3 µs	18.9 µs	-3.27%
❌	`PUSH1_50`	18.4 µs	19 µs	-3.26%
❌	`PUSH2_50`	18.5 µs	19.1 µs	-3.24%
❌	`PUSH8_50`	18.6 µs	19.2 µs	-3.07%
❌	`PUSH7_50`	18.8 µs	19.4 µs	-3.05%
❌	`STATICCALL_50`	89.8 µs	92.9 µs	-3.33%
❌	`subcall_1000_nested`	2.1 ms	2.3 ms	-5.58%
❌	`subcall_1000_transfer_1wei`	1.2 ms	1.3 ms	-4.15%
❌	`subcall_1000_same_account`	1.1 ms	1.2 ms	-5.34%
⚡	`blake2/12_rounds`	4.7 µs	4.5 µs	+4.06%
⚡	`blake2/2_rounds`	3.2 µs	3.1 µs	+4.12%
⚡	`blake2/4_rounds`	3.6 µs	3.4 µs	+5.49%

_{Comparing rakita/state-gas (f410840) with main (d060498)}

Add state gas accounting with reservoir model across interpreter, handler, context, and precompile crates. Includes comprehensive EIP-8037 test suite with 35+ test vectors covering SSTORE, CALL, CREATE, SELFDESTRUCT, reservoir refill, and gas opcode behavior.

… refill - Rename ResultGas::spent to regular_gas_spent to distinguish from state gas after EIP-8037 - Add total_gas_spent() method and deprecate old spent()/set_spent()/with_spent() accessors - Simplify reservoir refill logic: replace handler_reservoir_refill() with inline max() - Remove unnecessary reservoir manipulation in precompile path - Update op-revm handler to use simplified reservoir refill - Update Cargo.lock dependencies

…nd update test data

Consolidate the EIP-8037 auth list refund splitting logic from run_without_catch_error into apply_eip7702_auth_list, avoiding duplication between handler and inspector. pre_execution now takes init_and_floor_gas as a mutable reference and returns only the regular refund portion.

Introduce PrecompileFailure wrapper that carries both the error and an optional GasTracker, allowing precompiles that charge state gas (EIP-8037) to propagate gas info back to the caller on failure for proper reservoir refill.

Update all precompile Err returns to use .into() for converting PrecompileError to PrecompileFailure, enabling state gas tracking on precompile failures.

Commit 257d261 introduced a regression in gas accounting by incorrectly erasing both remaining and reservoir gas from total_gas_spent. The reservoir should not be erased since it's tracked separately via state_gas_spent. This fix reverts the erase_cost line from erase_cost(remaining + reservoir) to erase_cost(remaining), which correctly accounts for: - Regular gas: deducted from remaining calculation - State gas: tracked separately in state_gas_spent field Results: - State test failures reduced from 20 to 4 (pre-regression count) - All 408 workspace tests now pass - Updated EIP-8037 test data and assertions to match correct values

Refactor to improve code organization by moving the state gas deduction logic from the execution method into the first_frame_input method, where frame initialization concerns are naturally grouped together. This keeps frame setup logic cohesive and simplifies the execution method to focus on the execution loop rather than frame setup details. No functional change - same gas accounting and execution behavior.

Four fixes for EIP-8037 state gas interactions: 1. post_execution: Fix EIP-7623 floor gas check to account for EIP-8037 reservoir. Per EIP-8037, gas_used = tx.gas - gas_left - reservoir. The floor check now subtracts reservoir when computing gas used. 2. handler: Simplify reservoir handling in last_frame_result to always use the frame's final reservoir value directly. It already reflects child frame restorations via handle_reservoir_remaining_gas. 3. handler: Add EIP-8037 floor gas validation when gas_limit exceeds TX_MAX_GAS_LIMIT. The maximum gas_used is capped at TX_MAX_GAS_LIMIT since excess goes to reservoir, so floor_gas must not exceed that cap. 4. inspector: Remove double deduction of initial_state_gas in inspect_execution. The first_frame_input method already handles the deduction; the explicit deduction in the inspector was redundant.

… and EIP-7702 - tx_gas_used now subtracts unused reservoir gas (tx.gas - gas_left - state_gas_left) - Child revert/halt returns all state gas to parent reservoir (matching Python spec) - EIP-7702 state gas refund goes to reservoir instead of reducing initial_state_gas - Refund cap uses gas_used excluding reservoir - Precompile provider preserves reservoir across gas tracker replacement - Updated EIP-8037 test expectations and test data

…recovery Move require_non_staticcall! after gas charging in the CREATE/CREATE2 instruction handler. In the execution-specs, the static check is inside generic_create() after all gas (including state gas) has been charged. The previous ordering prevented state gas from being recorded on the frame, so on child error the parent could not recover it via handle_reservoir_remaining_gas.

This reverts commit 03986fc.

Compute total_gas_spent as limit - remaining - reservoir in build_result_gas to correctly exclude unused state gas. Use saturating_sub in Gas::spent/total_gas_spent to prevent underflow. Add std feature to ee-tests revm dependency.

Replace manual ResultGas construction with the shared build_result_gas function for consistency with the main execution path.

Gas had its own `limit` field duplicating `GasTracker::gas_limit`. Remove it and delegate `Gas::limit()` to `self.tracker.limit()`. Also fixes `Gas::new_spent` which previously set tracker gas_limit to 0 instead of the actual limit.

…reservoir support (#3541) * refactor(precompile): split PrecompileOutput and PrecompileError for state gas support Rename PrecompileOutput to PrecompileOutputEth (simple gas_used + bytes) for individual precompile functions. Introduce new PrecompileOutput with halt, gas_used, state_gas_used, reservoir, and bytes fields for provider-level results. Split PrecompileError: non-fatal errors become PrecompileHaltReason (expressed via PrecompileOutput::halt), while PrecompileError now only holds fatal errors (Fatal/FatalAny) that abort EVM execution. New type aliases: - PrecompileEthResult = Result<PrecompileOutputEth, PrecompileHaltReason> - PrecompileResult = Result<PrecompileOutput, PrecompileError> * refactor(precompile): replace Option<PrecompileHalt> with PrecompileStatus enum Introduce PrecompileStatus enum (Success, Revert, Halt) to replace the Option<PrecompileHalt> field in PrecompileOutput. Add PrecompileEthFn type alias for legacy precompile function signatures, with Precompile::execute() now returning PrecompileOutput via a From<PrecompileEthResult> conversion. Add helper methods: is_success, is_revert, halt_reason, into_halt_reason. * refactor(handler): extract precompile_output_to_interpreter_result helper Add a standalone function that converts PrecompileOutput into InterpreterResult, and use it in EthPrecompiles::run. Fmt and clippy fixes. * refactor(precompile): thread reservoir through PrecompileOutput and revert execute signature - Add `reservoir` parameter to PrecompileOutput::new, halt, revert constructors - Add `from_eth_result` constructor that takes reservoir - Revert Precompile::execute to return PrecompileEthResult - Move PrecompileOutput conversion to call site in EthPrecompiles - Remove unused into_halt_reason and From<PrecompileEthResult> impl * refactor(precompile): change Precompile fn_ from PrecompileEthFn to PrecompileFn (#3546) Change the Precompile struct's fn_ field from PrecompileEthFn (fn(&[u8], u64) -> PrecompileEthResult) to PrecompileFn (fn(&[u8], u64, u64) -> PrecompileOutput) so precompiles receive reservoir directly and return PrecompileOutput. Add call_eth_precompile helper to wrap existing PrecompileEthFn implementations into the new PrecompileFn signature. Each precompile module gets a thin wrapper using this helper. * refactor(precompile): add eth_precompile_fn! macro to eliminate wrapper boilerplate Replace 17 hand-written PrecompileEthFn-to-PrecompileFn wrapper functions with the new eth_precompile_fn! macro that generates them uniformly. * refactor(op-revm): use eth_precompile_fn! macro for precompile wrappers Replace 8 hand-written wrapper functions in op-revm with the eth_precompile_fn! macro. * style: apply cargo fmt * Rename Eth precompile structs * fix(op-revm): add missing reservoir argument to execute calls in tests

…ecompileStatus helpers PrecompileOutput now carries reservoir and state_gas_used, so precompile_output_to_interpreter_result no longer needs a separate reservoir parameter. Added convenience methods on PrecompileStatus and fixed unused import warning.

Update eip2537 benchmarks to use Precompile::execute() instead of removed precompile() method, and fix manual div_ceil clippy warning.

Remove references to removed getters (limit, spent, intrinsic_gas, remaining) and update table and derived values to reflect the actual methods (total_gas_spent, tx_gas_used, block_regular_gas_used, etc.).

- Fix unresolved links to non-existent `regular_gas_spent` and `with_regular_gas_spent` methods in deprecation notes - Fix typos: "substract" → "subtract", "enought" → "enough" - Fix stale comments referencing `regular_gas_spent` field

tempoxyz-bot

👁️ Cyclops Review

EIP-8037 dual gas model implementation. Core reservoir accounting matches the spec, but several downstream integration gaps exist — particularly in op-revm fee settlement, Amsterdam validation coverage, inspector gas reporting, and the precompile API boundary.

7 findings (2 high, 4 medium, 1 low)

Reviewer Callouts

⚡ Downstream reth/alloy-evm gas accounting: ExecutionResult::gas_used() is now receipt-domain only. Block builders using alloy-evm and reth still collapse receipt and block gas into a single accumulator. Once Amsterdam is enabled, valid blocks with state-gas transactions will produce incorrect header.gas_used. The in-tree harness was updated, but downstream integration needs block_regular_gas_used() / block_state_gas_used().
⚡ reth precompile cache: PrecompileOutput now carries halt status and live reservoir. reth's unchanged cache treats every Ok(output) as context-free success keyed by calldata/spec. Cached OOG halts (which set gas_used = 0) will always hit on later calls with sufficient gas, and cached success outputs replay stale reservoirs.
⚡ Gas accounting complexity: The saturating_sub → 0 pattern in first_frame_input() silently allows impossible negative regular-gas budgets instead of failing validation.

…eanups

…utput API Use `PrecompileOutput::from_eth_result` and `precompile_output_to_interpreter_result` instead of manually constructing InterpreterResult, aligning with the new gas handling.

…ecompileError> Distinguish fatal/unrecoverable precompile errors from non-fatal halts by returning a Result. PrecompileFn signature updated accordingly, and all call sites now propagate or unwrap the error.

The core revm handler already excludes reservoir gas when calculating beneficiary rewards and caller reimbursement, but op-revm's fee settlement did not account for it. Once Osaka/EIP-8037 is enabled, transactions with leftover reservoir gas would over-credit BASE_FEE_RECIPIENT and OPERATOR_FEE_RECIPIENT, and under-refund operator fees to the caller. - reward_beneficiary: use effective_used (gas.used() - reservoir) for base_fee_amount and operator_fee_cost calculations - operator_fee_refund: include gas.reservoir() when computing gas_used so the refund correctly covers unused reservoir gas - handler validation: use initial_regular_gas (excluding state gas) when checking floor_gas against TX_MAX_GAS_LIMIT cap

* feat: implement EIP-8037 state gas (reservoir model) Add state gas accounting with reservoir model across interpreter, handler, context, and precompile crates. Includes comprehensive EIP-8037 test suite with 35+ test vectors covering SSTORE, CALL, CREATE, SELFDESTRUCT, reservoir refill, and gas opcode behavior. * rm statetest jsons * Rename spent to regular_gas_spent in ResultGas and simplify reservoir refill - Rename ResultGas::spent to regular_gas_spent to distinguish from state gas after EIP-8037 - Add total_gas_spent() method and deprecate old spent()/set_spent()/with_spent() accessors - Simplify reservoir refill logic: replace handler_reservoir_refill() with inline max() - Remove unnecessary reservoir manipulation in precompile path - Update op-revm handler to use simplified reservoir refill - Update Cargo.lock dependencies * Simplify EIP-8037 state gas: remove redundant fields from ResultGas and update test data * Move EIP-7702 state gas refund split into pre_execution Consolidate the EIP-8037 auth list refund splitting logic from run_without_catch_error into apply_eip7702_auth_list, avoiding duplication between handler and inspector. pre_execution now takes init_and_floor_gas as a mutable reference and returns only the regular refund portion. * Add PrecompileFailure type for state gas propagation on error Introduce PrecompileFailure wrapper that carries both the error and an optional GasTracker, allowing precompiles that charge state gas (EIP-8037) to propagate gas info back to the caller on failure for proper reservoir refill. * Convert precompile errors to PrecompileFailure via .into() Update all precompile Err returns to use .into() for converting PrecompileError to PrecompileFailure, enabling state gas tracking on precompile failures. * Fix EIP-8037 state gas accounting regression Commit 14bd5a61 introduced a regression in gas accounting by incorrectly erasing both remaining and reservoir gas from total_gas_spent. The reservoir should not be erased since it's tracked separately via state_gas_spent. This fix reverts the erase_cost line from erase_cost(remaining + reservoir) to erase_cost(remaining), which correctly accounts for: - Regular gas: deducted from remaining calculation - State gas: tracked separately in state_gas_spent field Results: - State test failures reduced from 20 to 4 (pre-regression count) - All 408 workspace tests now pass - Updated EIP-8037 test data and assertions to match correct values * Move state gas deduction into first_frame_input Refactor to improve code organization by moving the state gas deduction logic from the execution method into the first_frame_input method, where frame initialization concerns are naturally grouped together. This keeps frame setup logic cohesive and simplifies the execution method to focus on the execution loop rather than frame setup details. No functional change - same gas accounting and execution behavior. * Fix EIP-8037 gas accounting: floor gas, reservoir, inspector, validation Four fixes for EIP-8037 state gas interactions: 1. post_execution: Fix EIP-7623 floor gas check to account for EIP-8037 reservoir. Per EIP-8037, gas_used = tx.gas - gas_left - reservoir. The floor check now subtracts reservoir when computing gas used. 2. handler: Simplify reservoir handling in last_frame_result to always use the frame's final reservoir value directly. It already reflects child frame restorations via handle_reservoir_remaining_gas. 3. handler: Add EIP-8037 floor gas validation when gas_limit exceeds TX_MAX_GAS_LIMIT. The maximum gas_used is capped at TX_MAX_GAS_LIMIT since excess goes to reservoir, so floor_gas must not exceed that cap. 4. inspector: Remove double deduction of initial_state_gas in inspect_execution. The first_frame_input method already handles the deduction; the explicit deduction in the inspector was redundant. * Fix EIP-8037 reservoir accounting: tx_gas_used, child revert refunds, and EIP-7702 - tx_gas_used now subtracts unused reservoir gas (tx.gas - gas_left - state_gas_left) - Child revert/halt returns all state gas to parent reservoir (matching Python spec) - EIP-7702 state gas refund goes to reservoir instead of reducing initial_state_gas - Refund cap uses gas_used excluding reservoir - Precompile provider preserves reservoir across gas tracker replacement - Updated EIP-8037 test expectations and test data * Fix CREATE/CREATE2 static call check ordering for EIP-8037 state gas recovery Move require_non_staticcall! after gas charging in the CREATE/CREATE2 instruction handler. In the execution-specs, the static check is inside generic_create() after all gas (including state gas) has been charged. The previous ordering prevented state gas from being recorded on the frame, so on child error the parent could not recover it via handle_reservoir_remaining_gas. * Simplify last_frame_result: remove initial_reservoir parameter and fix reservoir handling - Remove unused `initial_reservoir` parameter from `last_frame_result` across all handlers - Fix precompile provider to save reservoir before overwriting gas tracker - Add deprecated `gas_used()` method on ExecutionResult pointing to `tx_gas_used()` * Revert CREATE/CREATE2 static call check to before gas charging Static call check doesn't need to be after gas charging - CREATE in a static context is always an error regardless of gas accounting. * Fix reservoir handling: restore state gas on revert/halt On revert/halt, state changes are rolled back so state gas should be refunded back to the reservoir. Only on success should the frame's final reservoir be used as-is. * Fix EIP-8037 state_gas_spent accounting for EIP-7702 auth list Two bugs in state_gas_spent reporting: 1. build_result_gas added full initial_state_gas without subtracting eip7702_reservoir_refund, overcounting state gas for existing authority accounts. This caused block_state_gas_used to be too high and block_regular_gas_used to be too low. 2. On revert/halt, last_frame_result set state_gas_spent before restoring it to the reservoir, leaving a stale execution value. Now correctly zeroed since state changes are rolled back. * Improve precompile reservoir handling and apply formatting fixes Refactor precompile provider to properly handle reservoir refill logic for both success and error/halt cases, ensuring state gas accounting is correct when precompiles don't track reservoir. Also apply rustfmt formatting across multiple files. * Fix precompile_provider compilation errors against current precompile API PrecompileOutput has gas_used (not gas: GasTracker) and PrecompileError is a flat enum (no .error/.gas subfields). Use record_regular_cost() to avoid deprecation warning. * Simplify precompile interface: remove gas_limit from PrecompileOutput, flatten PrecompileError Remove gas_limit field from PrecompileOutput (only gas_used needed) and eliminate the PrecompileError wrapper layer so errors are returned directly as the enum variant without .into() conversion. * Fix EIP-8037 reservoir refill tests and complete precompile interface cleanup Update 4 EIP-8037 tests to match current reservoir refill behavior: on revert/halt at the top level, state_gas_spent is zeroed and state gas is restored to the reservoir (state changes rolled back). Regenerate testdata. Also complete the PrecompileOutput/PrecompileError simplification from the previous commit: remove reverted field, flatten error matching in tests. * temporary remove all ee-tests * Fix state gas accounting: include reservoir in total_gas_spent and prevent underflow - Remove saturating_sub of reservoir from total_gas_spent in build_result_gas - Use saturating_sub in block_regular_gas_used to prevent underflow * Apply formatting fixes and use saturating arithmetic for state gas spent * Revert "temporary remove all ee-tests" This reverts commit 03986fcd15d861b32ce2e56186ffd9ebdbfa3871. * Fix total_gas_spent to exclude reservoir and use saturating arithmetic Compute total_gas_spent as limit - remaining - reservoir in build_result_gas to correctly exclude unused state gas. Use saturating_sub in Gas::spent/total_gas_spent to prevent underflow. Add std feature to ee-tests revm dependency. * Use build_result_gas helper in system call gas handling Replace manual ResultGas construction with the shared build_result_gas function for consistency with the main execution path. * bump rust to 1.91 * Remove redundant gas_limit from Gas, delegate to GasTracker Gas had its own `limit` field duplicating `GasTracker::gas_limit`. Remove it and delegate `Gas::limit()` to `self.tracker.limit()`. Also fixes `Gas::new_spent` which previously set tracker gas_limit to 0 instead of the actual limit. * refactor(precompile): restructure PrecompileOutput for state gas and reservoir support (bluealloy/revm#3541) * refactor(precompile): split PrecompileOutput and PrecompileError for state gas support Rename PrecompileOutput to PrecompileOutputEth (simple gas_used + bytes) for individual precompile functions. Introduce new PrecompileOutput with halt, gas_used, state_gas_used, reservoir, and bytes fields for provider-level results. Split PrecompileError: non-fatal errors become PrecompileHaltReason (expressed via PrecompileOutput::halt), while PrecompileError now only holds fatal errors (Fatal/FatalAny) that abort EVM execution. New type aliases: - PrecompileEthResult = Result<PrecompileOutputEth, PrecompileHaltReason> - PrecompileResult = Result<PrecompileOutput, PrecompileError> * refactor(precompile): replace Option<PrecompileHalt> with PrecompileStatus enum Introduce PrecompileStatus enum (Success, Revert, Halt) to replace the Option<PrecompileHalt> field in PrecompileOutput. Add PrecompileEthFn type alias for legacy precompile function signatures, with Precompile::execute() now returning PrecompileOutput via a From<PrecompileEthResult> conversion. Add helper methods: is_success, is_revert, halt_reason, into_halt_reason. * refactor(handler): extract precompile_output_to_interpreter_result helper Add a standalone function that converts PrecompileOutput into InterpreterResult, and use it in EthPrecompiles::run. Fmt and clippy fixes. * refactor(precompile): thread reservoir through PrecompileOutput and revert execute signature - Add `reservoir` parameter to PrecompileOutput::new, halt, revert constructors - Add `from_eth_result` constructor that takes reservoir - Revert Precompile::execute to return PrecompileEthResult - Move PrecompileOutput conversion to call site in EthPrecompiles - Remove unused into_halt_reason and From<PrecompileEthResult> impl * refactor(precompile): change Precompile fn_ from PrecompileEthFn to PrecompileFn (bluealloy/revm#3546) Change the Precompile struct's fn_ field from PrecompileEthFn (fn(&[u8], u64) -> PrecompileEthResult) to PrecompileFn (fn(&[u8], u64, u64) -> PrecompileOutput) so precompiles receive reservoir directly and return PrecompileOutput. Add call_eth_precompile helper to wrap existing PrecompileEthFn implementations into the new PrecompileFn signature. Each precompile module gets a thin wrapper using this helper. * refactor(precompile): add eth_precompile_fn! macro to eliminate wrapper boilerplate Replace 17 hand-written PrecompileEthFn-to-PrecompileFn wrapper functions with the new eth_precompile_fn! macro that generates them uniformly. * refactor(op-revm): use eth_precompile_fn! macro for precompile wrappers Replace 8 hand-written wrapper functions in op-revm with the eth_precompile_fn! macro. * style: apply cargo fmt * Rename Eth precompile structs * fix(op-revm): add missing reservoir argument to execute calls in tests * refactor(precompile): move reservoir into PrecompileOutput and add PrecompileStatus helpers PrecompileOutput now carries reservoir and state_gas_used, so precompile_output_to_interpreter_result no longer needs a separate reservoir parameter. Added convenience methods on PrecompileStatus and fixed unused import warning. * fix: update bench and test code for new Precompile API Update eip2537 benchmarks to use Precompile::execute() instead of removed precompile() method, and fix manual div_ceil clippy warning. * docs: fix ResultGas doc comments to match current API Remove references to removed getters (limit, spent, intrinsic_gas, remaining) and update table and derived values to reflect the actual methods (total_gas_spent, tx_gas_used, block_regular_gas_used, etc.). * docs: fix broken doc links and typos in ResultGas - Fix unresolved links to non-existent `regular_gas_spent` and `with_regular_gas_spent` methods in deprecation notes - Fix typos: "substract" → "subtract", "enought" → "enough" - Fix stale comments referencing `regular_gas_spent` field * feat: track state_gas_spent and reservoir in GasInspector * fix: validate max(intrinsic_gas, floor_gas) against TX_MAX_GAS_LIMIT cap * chore: add must_use annotations to gas recording methods and minor cleanups * refactor: update custom_precompile_journal example to use PrecompileOutput API Use `PrecompileOutput::from_eth_result` and `precompile_output_to_interpreter_result` instead of manually constructing InterpreterResult, aligning with the new gas handling. * refactor: make Precompile::execute return Result<PrecompileOutput, PrecompileError> Distinguish fatal/unrecoverable precompile errors from non-fatal halts by returning a Result. PrecompileFn signature updated accordingly, and all call sites now propagate or unwrap the error. * chore: bump MSRV to 1.91 in CI * fix: exclude EIP-8037 reservoir gas from op-revm fee settlement The core revm handler already excludes reservoir gas when calculating beneficiary rewards and caller reimbursement, but op-revm's fee settlement did not account for it. Once Osaka/EIP-8037 is enabled, transactions with leftover reservoir gas would over-credit BASE_FEE_RECIPIENT and OPERATOR_FEE_RECIPIENT, and under-refund operator fees to the caller. - reward_beneficiary: use effective_used (gas.used() - reservoir) for base_fee_amount and operator_fee_cost calculations - operator_fee_refund: include gas.reservoir() when computing gas_used so the refund correctly covers unused reservoir gas - handler validation: use initial_regular_gas (excluding state gas) when checking floor_gas against TX_MAX_GAS_LIMIT cap

* feat: add EIP-8037 / TIP-1016 state gas support (bluealloy/revm#3406) * feat: implement EIP-8037 state gas (reservoir model) Add state gas accounting with reservoir model across interpreter, handler, context, and precompile crates. Includes comprehensive EIP-8037 test suite with 35+ test vectors covering SSTORE, CALL, CREATE, SELFDESTRUCT, reservoir refill, and gas opcode behavior. * rm statetest jsons * Rename spent to regular_gas_spent in ResultGas and simplify reservoir refill - Rename ResultGas::spent to regular_gas_spent to distinguish from state gas after EIP-8037 - Add total_gas_spent() method and deprecate old spent()/set_spent()/with_spent() accessors - Simplify reservoir refill logic: replace handler_reservoir_refill() with inline max() - Remove unnecessary reservoir manipulation in precompile path - Update op-revm handler to use simplified reservoir refill - Update Cargo.lock dependencies * Simplify EIP-8037 state gas: remove redundant fields from ResultGas and update test data * Move EIP-7702 state gas refund split into pre_execution Consolidate the EIP-8037 auth list refund splitting logic from run_without_catch_error into apply_eip7702_auth_list, avoiding duplication between handler and inspector. pre_execution now takes init_and_floor_gas as a mutable reference and returns only the regular refund portion. * Add PrecompileFailure type for state gas propagation on error Introduce PrecompileFailure wrapper that carries both the error and an optional GasTracker, allowing precompiles that charge state gas (EIP-8037) to propagate gas info back to the caller on failure for proper reservoir refill. * Convert precompile errors to PrecompileFailure via .into() Update all precompile Err returns to use .into() for converting PrecompileError to PrecompileFailure, enabling state gas tracking on precompile failures. * Fix EIP-8037 state gas accounting regression Commit 14bd5a61 introduced a regression in gas accounting by incorrectly erasing both remaining and reservoir gas from total_gas_spent. The reservoir should not be erased since it's tracked separately via state_gas_spent. This fix reverts the erase_cost line from erase_cost(remaining + reservoir) to erase_cost(remaining), which correctly accounts for: - Regular gas: deducted from remaining calculation - State gas: tracked separately in state_gas_spent field Results: - State test failures reduced from 20 to 4 (pre-regression count) - All 408 workspace tests now pass - Updated EIP-8037 test data and assertions to match correct values * Move state gas deduction into first_frame_input Refactor to improve code organization by moving the state gas deduction logic from the execution method into the first_frame_input method, where frame initialization concerns are naturally grouped together. This keeps frame setup logic cohesive and simplifies the execution method to focus on the execution loop rather than frame setup details. No functional change - same gas accounting and execution behavior. * Fix EIP-8037 gas accounting: floor gas, reservoir, inspector, validation Four fixes for EIP-8037 state gas interactions: 1. post_execution: Fix EIP-7623 floor gas check to account for EIP-8037 reservoir. Per EIP-8037, gas_used = tx.gas - gas_left - reservoir. The floor check now subtracts reservoir when computing gas used. 2. handler: Simplify reservoir handling in last_frame_result to always use the frame's final reservoir value directly. It already reflects child frame restorations via handle_reservoir_remaining_gas. 3. handler: Add EIP-8037 floor gas validation when gas_limit exceeds TX_MAX_GAS_LIMIT. The maximum gas_used is capped at TX_MAX_GAS_LIMIT since excess goes to reservoir, so floor_gas must not exceed that cap. 4. inspector: Remove double deduction of initial_state_gas in inspect_execution. The first_frame_input method already handles the deduction; the explicit deduction in the inspector was redundant. * Fix EIP-8037 reservoir accounting: tx_gas_used, child revert refunds, and EIP-7702 - tx_gas_used now subtracts unused reservoir gas (tx.gas - gas_left - state_gas_left) - Child revert/halt returns all state gas to parent reservoir (matching Python spec) - EIP-7702 state gas refund goes to reservoir instead of reducing initial_state_gas - Refund cap uses gas_used excluding reservoir - Precompile provider preserves reservoir across gas tracker replacement - Updated EIP-8037 test expectations and test data * Fix CREATE/CREATE2 static call check ordering for EIP-8037 state gas recovery Move require_non_staticcall! after gas charging in the CREATE/CREATE2 instruction handler. In the execution-specs, the static check is inside generic_create() after all gas (including state gas) has been charged. The previous ordering prevented state gas from being recorded on the frame, so on child error the parent could not recover it via handle_reservoir_remaining_gas. * Simplify last_frame_result: remove initial_reservoir parameter and fix reservoir handling - Remove unused `initial_reservoir` parameter from `last_frame_result` across all handlers - Fix precompile provider to save reservoir before overwriting gas tracker - Add deprecated `gas_used()` method on ExecutionResult pointing to `tx_gas_used()` * Revert CREATE/CREATE2 static call check to before gas charging Static call check doesn't need to be after gas charging - CREATE in a static context is always an error regardless of gas accounting. * Fix reservoir handling: restore state gas on revert/halt On revert/halt, state changes are rolled back so state gas should be refunded back to the reservoir. Only on success should the frame's final reservoir be used as-is. * Fix EIP-8037 state_gas_spent accounting for EIP-7702 auth list Two bugs in state_gas_spent reporting: 1. build_result_gas added full initial_state_gas without subtracting eip7702_reservoir_refund, overcounting state gas for existing authority accounts. This caused block_state_gas_used to be too high and block_regular_gas_used to be too low. 2. On revert/halt, last_frame_result set state_gas_spent before restoring it to the reservoir, leaving a stale execution value. Now correctly zeroed since state changes are rolled back. * Improve precompile reservoir handling and apply formatting fixes Refactor precompile provider to properly handle reservoir refill logic for both success and error/halt cases, ensuring state gas accounting is correct when precompiles don't track reservoir. Also apply rustfmt formatting across multiple files. * Fix precompile_provider compilation errors against current precompile API PrecompileOutput has gas_used (not gas: GasTracker) and PrecompileError is a flat enum (no .error/.gas subfields). Use record_regular_cost() to avoid deprecation warning. * Simplify precompile interface: remove gas_limit from PrecompileOutput, flatten PrecompileError Remove gas_limit field from PrecompileOutput (only gas_used needed) and eliminate the PrecompileError wrapper layer so errors are returned directly as the enum variant without .into() conversion. * Fix EIP-8037 reservoir refill tests and complete precompile interface cleanup Update 4 EIP-8037 tests to match current reservoir refill behavior: on revert/halt at the top level, state_gas_spent is zeroed and state gas is restored to the reservoir (state changes rolled back). Regenerate testdata. Also complete the PrecompileOutput/PrecompileError simplification from the previous commit: remove reverted field, flatten error matching in tests. * temporary remove all ee-tests * Fix state gas accounting: include reservoir in total_gas_spent and prevent underflow - Remove saturating_sub of reservoir from total_gas_spent in build_result_gas - Use saturating_sub in block_regular_gas_used to prevent underflow * Apply formatting fixes and use saturating arithmetic for state gas spent * Revert "temporary remove all ee-tests" This reverts commit 03986fcd15d861b32ce2e56186ffd9ebdbfa3871. * Fix total_gas_spent to exclude reservoir and use saturating arithmetic Compute total_gas_spent as limit - remaining - reservoir in build_result_gas to correctly exclude unused state gas. Use saturating_sub in Gas::spent/total_gas_spent to prevent underflow. Add std feature to ee-tests revm dependency. * Use build_result_gas helper in system call gas handling Replace manual ResultGas construction with the shared build_result_gas function for consistency with the main execution path. * bump rust to 1.91 * Remove redundant gas_limit from Gas, delegate to GasTracker Gas had its own `limit` field duplicating `GasTracker::gas_limit`. Remove it and delegate `Gas::limit()` to `self.tracker.limit()`. Also fixes `Gas::new_spent` which previously set tracker gas_limit to 0 instead of the actual limit. * refactor(precompile): restructure PrecompileOutput for state gas and reservoir support (bluealloy/revm#3541) * refactor(precompile): split PrecompileOutput and PrecompileError for state gas support Rename PrecompileOutput to PrecompileOutputEth (simple gas_used + bytes) for individual precompile functions. Introduce new PrecompileOutput with halt, gas_used, state_gas_used, reservoir, and bytes fields for provider-level results. Split PrecompileError: non-fatal errors become PrecompileHaltReason (expressed via PrecompileOutput::halt), while PrecompileError now only holds fatal errors (Fatal/FatalAny) that abort EVM execution. New type aliases: - PrecompileEthResult = Result<PrecompileOutputEth, PrecompileHaltReason> - PrecompileResult = Result<PrecompileOutput, PrecompileError> * refactor(precompile): replace Option<PrecompileHalt> with PrecompileStatus enum Introduce PrecompileStatus enum (Success, Revert, Halt) to replace the Option<PrecompileHalt> field in PrecompileOutput. Add PrecompileEthFn type alias for legacy precompile function signatures, with Precompile::execute() now returning PrecompileOutput via a From<PrecompileEthResult> conversion. Add helper methods: is_success, is_revert, halt_reason, into_halt_reason. * refactor(handler): extract precompile_output_to_interpreter_result helper Add a standalone function that converts PrecompileOutput into InterpreterResult, and use it in EthPrecompiles::run. Fmt and clippy fixes. * refactor(precompile): thread reservoir through PrecompileOutput and revert execute signature - Add `reservoir` parameter to PrecompileOutput::new, halt, revert constructors - Add `from_eth_result` constructor that takes reservoir - Revert Precompile::execute to return PrecompileEthResult - Move PrecompileOutput conversion to call site in EthPrecompiles - Remove unused into_halt_reason and From<PrecompileEthResult> impl * refactor(precompile): change Precompile fn_ from PrecompileEthFn to PrecompileFn (bluealloy/revm#3546) Change the Precompile struct's fn_ field from PrecompileEthFn (fn(&[u8], u64) -> PrecompileEthResult) to PrecompileFn (fn(&[u8], u64, u64) -> PrecompileOutput) so precompiles receive reservoir directly and return PrecompileOutput. Add call_eth_precompile helper to wrap existing PrecompileEthFn implementations into the new PrecompileFn signature. Each precompile module gets a thin wrapper using this helper. * refactor(precompile): add eth_precompile_fn! macro to eliminate wrapper boilerplate Replace 17 hand-written PrecompileEthFn-to-PrecompileFn wrapper functions with the new eth_precompile_fn! macro that generates them uniformly. * refactor(op-revm): use eth_precompile_fn! macro for precompile wrappers Replace 8 hand-written wrapper functions in op-revm with the eth_precompile_fn! macro. * style: apply cargo fmt * Rename Eth precompile structs * fix(op-revm): add missing reservoir argument to execute calls in tests * refactor(precompile): move reservoir into PrecompileOutput and add PrecompileStatus helpers PrecompileOutput now carries reservoir and state_gas_used, so precompile_output_to_interpreter_result no longer needs a separate reservoir parameter. Added convenience methods on PrecompileStatus and fixed unused import warning. * fix: update bench and test code for new Precompile API Update eip2537 benchmarks to use Precompile::execute() instead of removed precompile() method, and fix manual div_ceil clippy warning. * docs: fix ResultGas doc comments to match current API Remove references to removed getters (limit, spent, intrinsic_gas, remaining) and update table and derived values to reflect the actual methods (total_gas_spent, tx_gas_used, block_regular_gas_used, etc.). * docs: fix broken doc links and typos in ResultGas - Fix unresolved links to non-existent `regular_gas_spent` and `with_regular_gas_spent` methods in deprecation notes - Fix typos: "substract" → "subtract", "enought" → "enough" - Fix stale comments referencing `regular_gas_spent` field * feat: track state_gas_spent and reservoir in GasInspector * fix: validate max(intrinsic_gas, floor_gas) against TX_MAX_GAS_LIMIT cap * chore: add must_use annotations to gas recording methods and minor cleanups * refactor: update custom_precompile_journal example to use PrecompileOutput API Use `PrecompileOutput::from_eth_result` and `precompile_output_to_interpreter_result` instead of manually constructing InterpreterResult, aligning with the new gas handling. * refactor: make Precompile::execute return Result<PrecompileOutput, PrecompileError> Distinguish fatal/unrecoverable precompile errors from non-fatal halts by returning a Result. PrecompileFn signature updated accordingly, and all call sites now propagate or unwrap the error. * chore: bump MSRV to 1.91 in CI * fix: exclude EIP-8037 reservoir gas from op-revm fee settlement The core revm handler already excludes reservoir gas when calculating beneficiary rewards and caller reimbursement, but op-revm's fee settlement did not account for it. Once Osaka/EIP-8037 is enabled, transactions with leftover reservoir gas would over-credit BASE_FEE_RECIPIENT and OPERATOR_FEE_RECIPIENT, and under-refund operator fees to the caller. - reward_beneficiary: use effective_used (gas.used() - reservoir) for base_fee_amount and operator_fee_cost calculations - operator_fee_refund: include gas.reservoir() when computing gas_used so the refund correctly covers unused reservoir gas - handler validation: use initial_regular_gas (excluding state gas) when checking floor_gas against TX_MAX_GAS_LIMIT cap * refactor: move EIP-8037 gas cap validation into validate_initial_tx_gas (bluealloy/revm#3552) Extract `initial_regular_gas()` helper on InitialAndFloorGas and move the EIP-8037 TX_MAX_GAS_LIMIT validation from the handler into `validate_initial_tx_gas`, consolidating validation logic. Also clean up formatting across precompile and op-revm crates. * chore: release (bluealloy/revm#3472) Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> * chore: v106 release prep (bluealloy/revm#3554) * docs: add v106 migration guide entry * chore: v106 release prep — version bumps and changelogs * add revm version * fix(op-revm): update validate_against_state_and_deduct_caller signature Adds the new `&mut InitialAndFloorGas` parameter introduced in ethereum-optimism#3577 to the OpHandler trait impl, and passes `&mut Default::default()` at all test call sites. * chore: v107 release prep * chore(op-revm): fmt --------- Co-authored-by: rakita <rakita@users.noreply.github.com> Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: rakita <dragan0rakita@gmail.com>

rakita force-pushed the rakita/state-gas branch from 2a4830a to 2eb6654 Compare February 10, 2026 16:38

rakita changed the base branch from main to expose-gas February 10, 2026 16:42

rakita force-pushed the rakita/state-gas branch from 6ec22a9 to cf91d0f Compare February 11, 2026 13:30

Base automatically changed from expose-gas to main February 12, 2026 13:20

rakita force-pushed the rakita/state-gas branch from 2ade6e8 to f88aa3c Compare February 16, 2026 11:48

rakita commented Feb 16, 2026

View reviewed changes

Comment thread crates/context/interface/src/cfg/gas_params.rs Outdated

rakita commented Feb 17, 2026

View reviewed changes

Comment thread crates/context/interface/src/cfg/gas.rs

rakita commented Feb 17, 2026

View reviewed changes

Comment thread crates/context/interface/src/cfg/gas.rs Outdated

rakita changed the title ~~feat: add TIP-1016 state gas support~~ feat: add EIP-8037 / TIP-1016 state gas support Mar 4, 2026

rakita changed the base branch from main to devnet3 March 4, 2026 09:05

Base automatically changed from devnet3 to main March 13, 2026 10:45

rakita force-pushed the rakita/state-gas branch from f904eac to ee5f210 Compare March 13, 2026 10:50

rakita added 2 commits March 13, 2026 12:43

Merge remote-tracking branch 'origin/main' into state-gas

d154685

rm statetest jsons

556602b

rakita commented Mar 13, 2026

View reviewed changes

Comment thread crates/context/interface/src/cfg/gas_params.rs

rakita commented Mar 13, 2026

View reviewed changes

Comment thread crates/context/interface/src/cfg/gas_params.rs

rakita added 11 commits March 16, 2026 14:00

Simplify EIP-8037 state gas: remove redundant fields from ResultGas a…

257d261

…nd update test data

Merge remote-tracking branch 'origin/main' into state-gas

a9b3634

Convert precompile errors to PrecompileFailure via .into()

03b7c8d

Update all precompile Err returns to use .into() for converting PrecompileError to PrecompileFailure, enabling state gas tracking on precompile failures.

jenpaff requested a review from klkvr March 23, 2026 18:01

rakita and others added 11 commits March 25, 2026 10:33

Apply formatting fixes and use saturating arithmetic for state gas spent

f77d65f

Revert "temporary remove all ee-tests"

ef18e6a

This reverts commit 03986fc.

Use build_result_gas helper in system call gas handling

7e01885

Replace manual ResultGas construction with the shared build_result_gas function for consistency with the main execution path.

Merge remote-tracking branch 'origin/main' into rakita/state-gas

4b975de

bump rust to 1.91

8891655

Merge remote-tracking branch 'origin/main' into rakita/state-gas

a357a60

fix: update bench and test code for new Precompile API

28c73bb

Update eip2537 benchmarks to use Precompile::execute() instead of removed precompile() method, and fix manual div_ceil clippy warning.

rakita added the cyclops label Apr 9, 2026

rakita added 4 commits April 9, 2026 14:15

docs: fix ResultGas doc comments to match current API

d3b4555

Remove references to removed getters (limit, spent, intrinsic_gas, remaining) and update table and derived values to reflect the actual methods (total_gas_spent, tx_gas_used, block_regular_gas_used, etc.).

Merge remote-tracking branch 'origin/main' into state-gas

0064a83

feat: track state_gas_spent and reservoir in GasInspector

99b3589

tempoxyz-bot reviewed Apr 9, 2026

View reviewed changes

rakita added 6 commits April 9, 2026 18:28

fix: validate max(intrinsic_gas, floor_gas) against TX_MAX_GAS_LIMIT cap

c8460de

chore: add must_use annotations to gas recording methods and minor cl…

df67e80

…eanups

refactor: update custom_precompile_journal example to use PrecompileO…

0634dfa

…utput API Use `PrecompileOutput::from_eth_result` and `precompile_output_to_interpreter_result` instead of manually constructing InterpreterResult, aligning with the new gas handling.

refactor: make Precompile::execute return Result<PrecompileOutput, Pr…

577c903

…ecompileError> Distinguish fatal/unrecoverable precompile errors from non-fatal halts by returning a Result. PrecompileFn signature updated accordingly, and all call sites now propagate or unwrap the error.

chore: bump MSRV to 1.91 in CI

3f340d8

rakita merged commit f158464 into main Apr 10, 2026
29 of 31 checks passed

github-actions Bot mentioned this pull request Apr 9, 2026

chore: release #3472

Merged

rakita deleted the rakita/state-gas branch April 14, 2026 18:07

rakita restored the rakita/state-gas branch April 14, 2026 18:07

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

feat: add EIP-8037 / TIP-1016 state gas support#3406

feat: add EIP-8037 / TIP-1016 state gas support#3406
rakita merged 46 commits into
mainfrom
rakita/state-gas

rakita commented Feb 6, 2026 •

edited

Loading

Uh oh!

codspeed-hq Bot commented Feb 6, 2026 •

edited

Loading

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

tempoxyz-bot left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

rakita commented Feb 6, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Key Changes

Architecture

Cfg Integration

Test Suite

Test Plan

Uh oh!

codspeed-hq Bot commented Feb 6, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Merging this PR will degrade performance by 5.58%

Performance Changes

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

tempoxyz-bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

rakita commented Feb 6, 2026 •

edited

Loading

codspeed-hq Bot commented Feb 6, 2026 •

edited

Loading