PLT-457: StorageRWv1 contract + pinned ABI + binding

[bdchatham]

Implements PLT-457 — the contention contract for the sei-load workload modeler. Contract + binding only; no scenario wiring (PLT-461/465).

What

• generator/contracts/StorageRWv1.sol — mapping-backed store; write/read/rmw over a caller-selected slot. recordcount/keyspace is purely generator-side (no baked array size) → resize with no redeploy.
• read is a state-touching tx, not view (folds the loaded value into readAccumulator so the SLOAD can't be elided and the SUT pays for it).
• Trailing bytes calldata _pad on each function lets callers vary tx size independently of the key.
• Generated generator/bindings/StorageRWv1.go, committed.

Locked decisions honored

Mapping-backed; read non-view; versioned StorageRWv1; added alongside ERC20Conflict (untouched). Function selectors write(uint256,uint256,bytes) / read(uint256,bytes) / rmw(uint256,bytes) are the pinned one-way door.

Toolchain

solc 0.8.19 (repo-pinned) + abigen pinned to the repo's go-ethereum version. go build ./... passes. On CI/Linux the binding regenerates via make setup-node && make generate.

Decision brief: designs/sei-load-workload-modeler/PLT-457-storagerwv1-contract.md (sei-protocol/bdchatham-designs).

🤖 Generated with Claude Code

[cursor]

PR Summary

Medium Risk
New load-test contract bytecode is additive, but Makefile compile flag changes alter deployment bytecode for all contracts and could affect gas benchmarks if redeployed without coordination.

Overview
Adds StorageRWv1 for sei-load contention testing: a mapping-backed store, caller-chosen slot, and write / state-touching read (via readAccumulator) / rmw, plus optional _pad calldata. Commits generator/contracts/StorageRWv1.sol and generator/bindings/StorageRWv1.go; no scenario wiring in this PR.

The Makefile now pins solc 0.8.19 (sha256-verified download), EVM paris, --metadata-hash none, and abigen to the go.mod go-ethereum version. New targets generate-bindings, check-bindings, and install-abigen enforce committed bindings match contracts. Existing binding Bin fields change only from that reproducible compile (selectors/ABI unchanged).

A new Bindings Check GitHub Actions workflow runs make check-bindings on push/PR to main.

^{Reviewed by Cursor Bugbot for commit cd43b29. Bugbot is set up for automated code reviews on this repo. Configure here.}

[bdchatham]

Expert review — converged (no remaining concerns)

Reviewed by independent specialists (not the authoring agent), two rounds:

• Security / correctness: ✅ resolved a blocking defect — read's readAccumulator += store[slot] used 0.8 checked math and could overflow→revert (Panic 0x11) under caller-controlled write values, permanently bricking all subsequent reads and silently collapsing goodput. Now wrapped in unchecked (verified at the bytecode level: plain ADD, no PANIC path; rmw likewise for gas-profile parity). Confirmed unchecked does not alter the read/rmw/write selectors and introduces no new safety surface (values are never asserted on).
• Idiom (Solidity + binding): ✅ resolved — added uint256 public constant VERSION = 1 (supports the locked v2-coexistence decision); binding regenerated via the repo's solc 0.8.19 + pinned abigen (pipeline-consistent, file mode 0644). Named bytes calldata _pad vetted as the correct readable choice.

Scenario/factory wiring is intentionally out of scope (deferred to PLT-461/465); the contract + pinned binding are what this ticket delivers. go build ./... green.

[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 aa93ebf. Configure here.}

[bdchatham]

Systems review — APPROVE (+ EVM-target decision, researched)

Final systems-engineer pass: APPROVE, no blocking items. Two strongly-recommended solc-flag hardenings (the angle security/idiom structurally don't cover — what the compiler stamps into the deployed artifact) are now applied:

• --evm-version paris (explicit). Researched against sei-chain (x/evm/types/config.go, HEAD 2026-06-07): Sei runs a geth fork v1.15.7-sei-16 with Shanghai + Cancun + Prague active from genesis ("Pectra w/o blobs"), so paris-targeted bytecode is a strict subset — unconditionally safe to deploy. Crucially, --evm-version is a compile-time opcode-availability knob, not a runtime-gas knob: runtime gas is charged by the chain's active fork regardless of target, so paris-targeting does not distort measured gas. paris is solc 0.8.19's ceiling anyway, and StorageRWv1 (only SLOAD/SSTORE/ADD/mapping) emits identical bytecode under paris/shanghai/cancun. Pinning it explicitly prevents a future solc bump from silently changing emitted opcodes (the "measurement discontinuity" risk). Consistent with all existing repo contracts.
• --metadata-hash none. Removes the path-dependent IPFS metadata trailer → reproducible bytecode across repo layout / build host, and avoids phantom "drift" on a future repo reorg.

Both flags regenerate all 6 bindings — verified ABIs are byte-identical (only bytecode changed) and all selectors are preserved.

Deferred (not now): matching Cancun via a solc bump (≥0.8.24) buys nothing measurable for the current SLOAD/SSTORE contracts; un-defer only if we add a transient-storage (TSTORE/TLOAD) benchmark variant or want to measure Cancun-specific opcodes as first-class instruments.