[codex] add dynamic custom precompile gas meter

[codchen]

Summary

Adds dynamic gas accounting for evm-only custom precompiles, with the staking precompile using the shared meter through the existing store, balance-transfer, and log boundaries.

Changes

• Add a reusable precompile gas meter for base gas, SLOAD/SSTORE-style storage costs, keccak slot derivation, native value transfers, and log emission.
• Wrap storageBackedStore, nativeBalanceTransfer, and log emission in the custom precompile adapter so gas is charged based on the actual execution path.
• Reduce staking RequiredGas to only base/input gas, avoiding double charging now that state access is dynamically metered.
• Add executor coverage for dynamic store gas and update staking lifecycle tests to use realistic gas limits.

Validation

• go test ./giga/evmonly/...
• golangci-lint v2.8.0 run ./giga/evmonly/...

[codecov]

Codecov Report

❌ Patch coverage is 59.66851% with 73 lines in your changes missing coverage. Please review.
✅ Project coverage is 58.15%. Comparing base (accacbf) to head (7c6c44c).

Files with missing lines	Patch %	Lines
giga/evmonly/precompile_gas.go	68.46%	23 Missing and 18 partials ⚠️
giga/evmonly/precompile_adapter.go	34.69%	19 Missing and 13 partials ⚠️

Additional details and impacted files

@@                         Coverage Diff                          @@
##           codex/evmonly-staking-precompile    #3644      +/-   ##
====================================================================
- Coverage                             58.15%   58.15%   -0.01%     
====================================================================
  Files                                  2187     2185       -2     
  Lines                                178461   178564     +103     
====================================================================
+ Hits                                 103781   103836      +55     
- Misses                                65383    65413      +30     
- Partials                               9297     9315      +18     

Flag	Coverage Δ
sei-chain-pr	56.29% <59.66%> (-0.12%)	⬇️
sei-db	70.41% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
giga/evmonly/precompiles/staking/staking.go	23.91% <100.00%> (-0.40%)	⬇️
giga/evmonly/precompile_adapter.go	66.35% <34.69%> (-8.65%)	⬇️
giga/evmonly/precompile_gas.go	68.46% <68.46%> (ø)

... and 5 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.

[github-actions]

The latest Buf updates on your PR. Results from workflow Buf / buf (pull_request).

Build	Format	Lint	Breaking	Updated (UTC)
✅ passed	✅ passed	✅ passed	✅ passed	Jun 30, 2026, 6:12 AM

[cursor]

PR Summary

Medium Risk
Changes consensus-relevant gas and refund behavior for custom precompiles and staking; incorrect metering could cause unexpected OOG or undercharging, though coverage targets store OOG and staking e2e paths.

Overview
Introduces dynamic gas accounting for evm-only custom precompiles so state work is paid from the transaction gas budget instead of a flat upfront fee.

RunAndCalculateGas now uses a shared precompileGasMeter: it charges RequiredGas first, then meters Keccak for store key/slot derivation, EIP-2929/2200-style SLOAD/SSTORE (including refunds and access lists), native value transfers, and log emission. The adapter wires the meter into storageBackedStore, nativeBalanceTransfer, and a meteredLogSink; on exhaustion the precompile fails with ErrOutOfGas and store paths stop persisting changes.

The staking precompile drops fixed read/write gas in RequiredGas (now 700 + 16×input bytes only) so storage and transfers are not double-charged.

Tests assert higher gas use on successful store writes, OOG with no storage when gas is too low, and staking flows use realistic gas limits (8M).

^{Reviewed by Cursor Bugbot for commit bd57e4c. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit bd57e4c. Configure here.}

[seidroid]

The dynamic precompile gas meter is implemented carefully and faithfully mirrors go-ethereum's EIP-2929/2200/3529 SLOAD/SSTORE/refund accounting, with refund state being snapshot-safe and OOG correctly surfaced and discarded; tests cover the new metering. The only issue is a now-orphaned helper function and a behavioral note about read-method gas.

Findings: 0 blocking | 4 non-blocking | 1 posted inline

Blockers

• None at the file/PR level.

Non-blocking

• isTransaction in giga/evmonly/precompiles/staking/helpers.go was the only consumer of the removed read/write gas split and is now unused dead code (no remaining references in the package). Remove it — golangci-lint's unused check may flag this and fail CI even though the PR notes lint passed.
• Behavior change worth noting: read-only query methods (validators, delegatorDelegations, etc.) previously cost a flat readGas (3000) and now meter per-SLOAD dynamically. Read-heavy paged queries (up to pageLimit=100 validators, each a multi-chunk read) can consume substantially more gas than before; callers/eth_call gas caps must account for this. This is more accurate, just a notable semantics change.
• Codex and Cursor second-opinion passes both produced no material findings.
• 1 suggestion(s)/nit(s) flagged inline on specific lines.

[seidroid]

[suggestion] With the read/write gas split removed here, isTransaction in helpers.go is no longer referenced anywhere in the package and is now dead code. Consider removing it — the unused linter may otherwise fail CI.

[seidroid]

The PR adds a dynamic gas meter for evm-only custom precompiles whose SLOAD/SSTORE/access-list/keccak/log/transfer pricing faithfully mirrors go-ethereum (EIP-2200/2929/3529), with refund and access-list mutations correctly rolled back on OOG via RevertToSnapshot and good test coverage. The metering relies on GetCommittedState as the SSTORE "original" value, but nativeStateDB.base is block-start state that Finalise never advances, so SSTORE cost/refunds are mispriced for slots written across multiple transactions in the same block.

Findings: 1 blocking | 2 non-blocking | 1 posted inline

Blockers

• None at the file/PR level.
• 1 blocking issue(s) flagged inline on specific lines.

Non-blocking

• cursor-review.md was empty — the Cursor second-opinion pass produced no output, so its perspective is not reflected here. (REVIEW_GUIDELINES.md was also empty; reviewed without repo-specific guidance.)
• Minor: chargeNativeTransfer charges account-access (cold/warm) for both from and to. Standard EVM CALL gas does not recharge access for the caller/from (it is already warm), so transfers slightly overcharge relative to geth. Likely acceptable as a custom model, but worth confirming it is intentional.

[seidroid]

[blocker] GetCommittedState is used as the EIP-2200 "original" (transaction-start) value, but in this nativeStateDB the base map holds block-start state and is never advanced between transactions — Finalise (state_db.go:492) only clears selfdestruct/created flags and resets refund, and account()/baseAccount() populate base lazily from source (block start). The executor reuses one stateDB for the whole block (executor.go:80), so for a slot written in tx N and re-written/cleared in tx N+1 of the same block, original here is the block-start value rather than the tx N+1-start value.

This diverges from go-ethereum (where per-tx Finalise moves dirty storage into pending storage so GetCommittedState reflects prior txs in the block), causing SSTORE to take the wrong SET/RESET/dirty branch and emit incorrect AddRefund/SubRefund amounts (under/over-charging and bogus clear refunds). Before this PR the flat read/write gas hid the issue; the dynamic meter now makes it consensus-relevant for gas accounting. Consider tracking a separate transaction-start snapshot (advanced in Finalise) distinct from the block-start base used by ChangeSet.