Repository navigation
Conversation
`AnvilProcessSteps.runStart` spawned anvil with only host/port/chain-id/state and switched interval mining on afterward over RPC. anvil has no RPC for the epoch slot count, so every `create` and flow cluster ran on anvil's default 32-slot epoch: `finalized` sat 64 blocks behind `latest`, and the batch operators — which read the outpost's outbound envelope at `finalized` — only saw each ETH envelope a minute after the tip that emitted it. Every depot epoch stretched to about two minutes waiting on that one hop. Ride `AnvilProcess.SlotsInAnEpoch` on the spawn (finality two blocks behind), exactly as the rendered start.sh already does for the `run` path, and keep `--block-time` off the spawn so the Hardhat deploy keeps its instamine. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
brianjohnson5972
left a comment
brianjohnson5972
left a comment
There was a problem hiding this comment.
Peer review alongside wire-ethereum#216 and wire-sysio#660. Looks good to merge. It is independent of the other two PRs and has no ordering constraint. The fix is correct and well tested. Inline: one FYI about production ETH finality latency, plus two test nits.
| AnvilProcess.StateFilename | ||
| ) | ||
| ), | ||
| slotsInAnEpoch: AnvilProcess.SlotsInAnEpoch |
There was a problem hiding this comment.
This looks right. anvil has no RPC to change --slots-in-an-epoch, so it has to be set at spawn. --block-time correctly stays off the spawn, so the Hardhat deploy keeps instamine, and this now matches the rendered start.sh on the run path.
FYI, not a change request for this PR: the measurement here (tip-to-read 74 s → 15 s, flow 842 s → 563 s) shows how much of each epoch is the single ETH→depot hop, read at finalized. On mainnet finalized trails latest by ~12.8 minutes (two epochs), so in production every depot epoch waits at least one ETH finality period, whatever epoch_duration_sec is set to. Worth flagging to whoever sizes production epochs and the depot's path-2 boundary.
| // ProcessManager is a set-once singleton: one sandbox for the whole block. | ||
| beforeAll(() => { | ||
| dir = Fs.mkdtempSync(Path.join(Os.tmpdir(), "anvilsteps-")) | ||
| ProcessManager.setClusterPath(dir) |
There was a problem hiding this comment.
Nit: ProcessManager.setClusterPath is set-once per process. That's fine here because jest gives each test file its own module registry, but a one-line note that this file must keep a single sandbox for the whole describe would stop someone adding a second describe with its own beforeAll from silently inheriting the first path.
| it("spawns anvil with the finality window and without interval mining", async () => { | ||
| const ctx = fixtureContext({ clusterPath: dir, dataPath: Path.join(dir, "data") }), | ||
| start = jest.fn(), | ||
| anvil = { start } as Partial<AnvilProcess> as AnvilProcess, |
There was a problem hiding this comment.
Nit: { start } as Partial<AnvilProcess> as AnvilProcess (here and at line 76) is fine in a test. If jest.mocked / a typed stub helper already exists in the repo's test utilities, prefer it so the double cast doesn't spread.
This branch has not been deployed
2 participants
Summary
One harness fix found while running the whole-envelope delivery branches end to end: pass
--slots-in-an-epochto the run-time anvil at spawn.AnvilProcessSteps.runStartspawned anvil with only host/port/chain-id/state and switched interval mining on afterwards over RPC. anvil has no RPC for the epoch slot count, so everycreateand flow cluster ran on the default 32-slot epoch:finalizedsat 64 blocks behindlatest, and the batch operators — which read the outpost's outbound envelope atfinalized— only saw each Ethereum envelope a minute after the tip that emitted it. Every depot epoch stretched to about two minutes waiting on that one hop.AnvilProcess.SlotsInAnEpochnow rides the spawn (finality two blocks behind), exactly as the renderedstart.shalready did for therunpath;--block-timestays off the spawn so the Hardhat deploy keeps instamine andplanEnableIntervalMiningstill turns mining on afterwards. Measured onflow-operator-collateral-deposit: tip-to-read 74s → 15s, flow 842s → 563s.Runner tests added in
AnvilProcessSteps.test.ts; repo gate (lint, build, jest 2268 tests) green.Cross-linked below: the wire-ethereum contracts and wire-sysio cranker this was run against.
Companion PRs
E2E gate
e2e-tests run 36780734850 —
success, full platform build plus every discovered flow, dispatched withBRANCH_WIRE_SYSIO,BRANCH_WIRE_ETHEREUM,BRANCH_WIRE_TOOLS_TSandBRANCH_WIRE_LIBRARIES_TS=feature/eth_opp_calldata_delivery(wire-solana and wire-cdt on their manifest branches).All 15 flows passed.
🤖 Generated with Claude Code