fix: update pnpm lockfile and dispatch epoch_boundary sanity runner

[lodekeeper]

Opening this against your feat/nightly-workflow branch so you don't need to patch it yourself (per @nflaig's ask on ChainSafe#9221).

Two changes:

1. pnpm-lock.yaml — the upstream Build job on feat: support nightly spec tests ChainSafe/lodestar#9221 was failing with ERR_PNPM_OUTDATED_LOCKFILE. dotenv@^16.4.5 was added to packages/beacon-node/package.json but the lockfile wasn't regenerated. Three-line addition in the beacon-node importer section.

2. test/spec/presets/sanity.test.ts — when I ran pnpm test:spec:minimal against freshly-downloaded nightly artifacts, the gloas epoch_boundary/pyspec_tests suite (new in consensus-specs Schedule capella on sepolia ChainSafe/lodestar#5127) threw Unknown sanity test epoch_boundary from the dispatcher. The fixture format is identical to blocks (pre.ssz_snappy + post.ssz_snappy + blocks_N.ssz_snappy + meta.blocks_count), so routing it through the existing sanityBlocks runner is sufficient.

Verification

• pnpm install --frozen-lockfile --prefer-offline succeeds from the updated lockfile.
• Nightly artifact download end-to-end (unstable branch -> pretest -> extract 3 .tar.gz into spec-tests-nightly) — OK.
• pnpm test:spec:minimal result with both fixes applied:
  • Test Files: 1 failed | 13 passed
  • Tests: 5 failed | 56128 passed | 95 skipped
  • All 5 failures are in gloas/sanity/epoch_boundary/pyspec_tests (epoch_boundary_full_parent_* + switch_to_compounding_across_epoch_boundary) with Block parent root ... does not match state latest block at processBlockHeader.ts:34. These are pre-existing Lodestar state-transition gaps surfaced by the new gloas tests (exactly what the nightly workflow is designed to catch), not regressions introduced by this PR — needs separate investigation once nightly CI is landed.

Commits

• b4c68811cb — chore: update pnpm-lock.yaml for dotenv devDependency
• f84920bd23 — test: add epoch_boundary sanity runner case for nightly spec tests

🤖 Generated with AI assistance

[nflaig]

@brech1 I think this is because of the epoch boundary tests I added, maybe these should be under blocks?

[lodekeeper]

@nflaig — I dug into this, and the routing epoch_boundary vs blocks isn't actually the issue. The fixture layout is identical (pre / post / blocks_N.ssz_snappy / meta.blocks_count), so whether Lodestar dispatches via case "epoch_boundary": or moves the tests under the blocks/ directory, the consumer-side replay is the same.

The real cause of the 5 failures is in the test generator itself: set_parent_block_full(spec, state) is called between signed_block_1 = state_transition_and_sign_block(...) and block_2 = _build_..., i.e. after yield "pre", state. That mutation touches state.latest_block_hash (+ state.latest_execution_payload_bid.block_hash), which changes hash_tree_root(state), which in turn ends up written into state.latest_block_header.state_root by the very next process_slot. Python computes block_2.parent_root = hash_tree_root(state.latest_block_header) using that mutated state root. But the fixture has no way to re-apply that mutation on replay — no envelope file, no second state, nothing. So consumers hit processBlockHeader.ts:34 with a parent-root mismatch.

Evidence for this interpretation:

• All 3 passing test_epoch_boundary_empty_parent_* only call set_parent_block_full before yield "pre", so the mutation is baked into the pre-state and fully captured.
• All 5 failing tests (test_epoch_boundary_full_parent_* + test_switch_to_compounding_across_epoch_boundary + test_epoch_boundary_full_parent_all_requests_gap_5_epochs) call it a second time between blocks 1 and 2, which is the one that can't be serialized.

Fix direction is in consensus-specs, not Lodestar. Either (a) teach the fixture format to carry an execution_payload_envelopes_N.ssz_snappy and teach the sanity runner to deliver it between blocks, or (b) restructure the tests so the FULL-parent path is exercised by actually processing a real envelope through spec.process_execution_payload instead of a test helper that mutates state out-of-band. Option (b) is smaller.

Happy to open a PR against consensus-specs once you confirm the direction.