test: add an in-process Tevm test lane, replacing Anvil where rc.151 supports it

[roninjin10]

Motivation

The suite starts an anvil child process for essentially every stateful test, behind a Prool proxy on a fixed port. That means a Foundry binary on PATH, a process per worker, port bookkeeping, and startup polling before a single assertion runs.

Tevm is an EVM that runs inside the Vitest worker. This PR wires it up as a first-class test lane and migrates as much of the suite to it as rc.151 correctly supports — while showing off what the Tevm test library can actually do.

Before / after

	Before (Anvil)	After (Tevm lane)
Node	anvil child process	in-process MemoryClient
Discovery	Prool proxy, fixed port	none — direct EIP-1193
Startup	spawn + poll until ready	createMemoryClient() returns immediately; tevmReady() awaits the fork anchor
Isolation	pool instance per worker	evm_snapshot baseline, reverted before every test
Global setup	setup.global.ts	none
Mining	noMining: true	miningConfig: { type: 'manual' }

Viem clients are still constructed with this repository's createClient, over a custom EIP-1193 transport that forwards to memoryClient.request. Migrated suites therefore still exercise this fork's actions, client, and transport code — Tevm only supplies the node. Tevm depends on a newer published Viem than this fork builds; that boundary is crossed in exactly one place, the adapter in test/src/tevm.ts.

What's here

A third Vitest project. tevm matches src/**/*.tevm.test.ts, uses test/setup.tevm.ts, and has no globalSetup. Everything it shares with core moved to test/setup.shared.ts.

test/src/tevm.ts — tevmMainnet, the Tevm counterpart of anvilMainnet. Pinned mainnet fork, manual mining, the familiar prefunded accounts, getClient() with the same shape as the Anvil fixture, and snapshot-based per-test isolation (a successful evm_revert consumes its id, so a replacement snapshot is taken immediately).

17 migrated action suites — getCode, getStorageAt, dropTransaction, dumpState, impersonateAccount, increaseTime, mine, revert, setBalance, setCode, setNextBlockTimestamp, setNonce, setStorageAt, snapshot, stopImpersonatingAccount, setBlockTimestampInterval, removeBlockTimestampInterval.

The complete HTTP transport suite. test/src/tevm-server.ts serves a MemoryClient over an ephemeral 127.0.0.1 listener (port 0) using createServer from tevm/server for the real-node paths; test/src/http-server.ts is a plain node:http server for the JSON-RPC error and header cases, which tevm/server answers with HTTP 400 rather than 200-with-error. No coverage was dropped.

Tevm test features shown off

src/tevm-showcase.tevm.test.ts, split into focused cases:

• in-process node — a MemoryClient and this repo's Viem client in one process; manual mining (transactions stay pending until tevmMine); snapshot & revert through the Anvil-compatible anvil_snapshot / anvil_revert RPC surface
• forking — a pinned mainnet fork, and a direct measurement of Tevm's lazy in-memory fork-state cache using a counting wrapper around the upstream provider
• high-level actions — tevmSetAccount, tevmDeal, tevmDeploy, tevmContract, tevmMine, impersonateAccount
• low-level TevmNode handlers — callHandler, contractHandler, dealHandler, deployHandler, setAccountHandler against a bare createTevmNode()
• published fixtures — SimpleContract, AdvancedContract, ErrorContract, TestERC20 from @tevm/test-utils
• every @tevm/test-matchers family — primitives (toBeAddress, toBeHex, toEqualAddress, toEqualHex), account & state (toBeInitializedAccount, toHaveState, toHaveStorageAt), events (toEmit with withEventArgs / withEventNamedArgs), traces (toCallContractFunction with withFunctionNamedArgs), reverts (toBeRevertedWithString, toBeRevertedWithError with withErrorNamedArgs), balances (toChangeBalance, toChangeTokenBalance)

Honesty about rc.151

Several matcher families run against non-fork createTevmNode() instances: on a forked node rc.151 resolves balances and storage through eth_getProof against the upstream historical block, so locally written values are invisible. test/README.md lists every gap found.

Six suites keep their original Anvil coverage and gain a .tevm sibling that asserts Tevm's actual behaviour rather than restating Anvil's — nothing is weakened:

Suite	Divergence the .tevm sibling documents
setStorageAt	eth_getStorageAt returns the storage word right-padded
getStorageAt	forked slots other than the lazily cached one read back as zero
impersonateAccount / stopImpersonatingAccount	Tevm auto-impersonates, so the "No Signer available" negatives cannot hold
setBlockTimestampInterval / removeBlockTimestampInterval	anvil_setBlockTimestampInterval is recorded but not applied to mined blocks

The fork is pinned to 22263621, the nearest blob-free block below Anvil's 22263623: rc.151's Common.copy() drops customCrypto, so a forked anchor containing EIP-4844 transactions cannot be deserialized. Choosing a blob-free anchor is preferred over filtering type-3 transactions out of an otherwise inconsistent block response — the fork sees exactly what the provider returns.

Anvil is not going anywhere: WebSocket and IPC transports, the Alto bundler / account-abstraction suites, the Optimism / zkSync / Sepolia fixtures, wallet-provider emulation, sendUnsignedTransaction, setAutomine and setIntervalMining all still need it, and pnpm install still runs contracts:build with forge.

Dependency hygiene

Tevm packages are pinned to exact 1.0.0-rc.151 (the moving latest / next tags resolve to older builds with a much smaller Anvil surface), and pnpm.packageExtensions pins the entire Tevm family onto one published Viem — @tevm/errors was resolving a second copy and is now pinned too. @tevm/server is deliberately not installed (its latest tag is an obsolete 0.0.1); createServer comes from tevm/server. @tevm/ts-plugin is not a dependency until direct Solidity imports are actually enabled.

src/tevm-resolution.tevm.test.ts guards all of that at runtime: it imports every Tevm entrypoint the suite uses, so a peer-resolution regression — Tevm silently binding the unbuilt workspace Viem — fails here, first and legibly, instead of as an opaque module error deep inside an unrelated suite.

No changeset: this change is entirely test infrastructure and devDependencies, with no effect on the published package.

Test evidence

The Tevm lane, against an archive mainnet RPC:

$ VITE_ANVIL_FORK_URL=<archive-rpc> pnpm exec vitest run -c test/vitest.config.ts --project tevm

 RUN  v4.1.10 /Users/williamcory/evmts-viem-wt-final

 Test Files  20 passed (20)
      Tests  59 passed (59)
   Duration  8.66s

The six suites whose original Anvil coverage was restored, on the core (Anvil) project:

$ pnpm exec vitest run -c test/vitest.config.ts --project core \
    src/actions/public/getStorageAt.test.ts \
    src/actions/test/{setStorageAt,impersonateAccount,stopImpersonatingAccount}.test.ts \
    src/actions/test/{setBlockTimestampInterval,removeBlockTimestampInterval}.test.ts

 Test Files  6 passed (6)
      Tests  7 passed (7)

biome check src test reports no new findings (the 9 warnings are pre-existing, all in src/tempo/). tsc -b reports no errors in any file this PR adds or modifies; the remaining errors are the pre-existing Voltaire-migration ones on main.

Credits

Produced by a multi-agent run: three independent implementation lanes (codex-sol, opus, kimi) were built and judged, and this branch is the winning lane with the best ideas from the others grafted in — the focused showcase split, the explicit rc.151 limitations matrix and retained-Anvil coverage, the blob-free fork block, dropping @tevm/ts-plugin, and the runtime resolution smoke tests.

🤖 Generated with Smithers multi-agent orchestration