fix: read a block's event metadata from its parent state

[mrq1911]

Follow-up to #1503, which is already merged and tagged v49.2.1. Fixes a cache-poisoning bug found in a security review of that PR.

The bug

layout() keyed the event-layout cache on the spec version from at's post-state, but fetched metadata by calling the runtime API at at — which executes whatever :code sits in at's post-state.

block N-1 ─┬─ :code = old
block N    ├─ executed by OLD runtime; cumulus on_initialize writes :code = NEW
           │   LastRuntimeUpgrade = old  (frame_executive only bumps it when the new runtime first runs)
block N+1  └─ executed by NEW runtime; LastRuntimeUpgrade = new

At block N the metadata call returns the new runtime's layout while the spec key is still old, so the new layout gets cached under the old spec version. Every later old-spec block then decodes with the wrong metadata — and since record spans are measured from that metadata, a wrong layout can mis-measure and swallow neighbouring records, which is the exact failure #1503 exists to eliminate. It persists until restart.

The fix

Keep the spec key from at — LastRuntimeUpgrade in a block's post-state exactly names the runtime that executed it, which is the layout needed to decode its events — and take the metadata from the parent, whose post-state :code is that runtime:

layout for N   → spec old (from N)   + metadata from N-1 (old code)  ✓
layout for N+1 → spec new (from N+1) + metadata from N   (new code)  ✓

Note taking the spec key from the parent instead would mismatch in the other direction: at N+1 the parent's post-state already has the new :code but LastRuntimeUpgrade still reads old.

Falls back to at at genesis and when the parent header is unknown, so it is never worse than before.

Tests

• metadata_comes_from_the_parent_state — the regression itself
• genesis_and_unknown_blocks_fall_back_to_themselves — no reads at the zero hash

18 node synth tests pass; formatting clean. The same fix is on #1504 as 5eb4e9d7f.

Once this lands, v49.2.1 moves onto it — no GitHub release or Docker image was published against the old commit, so nothing downstream is pinned.

[github-actions]

Weight Diff Report

⚠️ 2 pallet(s) have changes exceeding ±10% threshold

35 extrinsic(s) changed across 2 pallet(s). New: 4. Removed: 2.

pallet_omnipool_liquidity_mining

Extrinsic	RefTime	Proof Size	Reads	Writes
create_global_farm	-0.1% (119.1M → 118.9M)	—	8 → 7 (-1)	—
update_global_farm	+0.8% (164.2M → 165.6M)	—	9 → 8 (-1)	—
terminate_global_farm	+1.5% (137.6M → 139.6M)	—	8 → 7 (-1)	—
create_yield_farm	-0.6% (186.5M → 185.5M)	—	12 → 11 (-1)	—
update_yield_farm	-0.1% (191.7M → 191.6M)	—	12 → 11 (-1)	—
stop_yield_farm	-0.1% (184.5M → 184.4M)	—	11 → 10 (-1)	—
resume_yield_farm	+0.2% (188.5M → 188.9M)	—	12 → 11 (-1)	—
terminate_yield_farm	-0.9% (135.8M → 134.6M)	—	9 → 8 (-1)	—
deposit_shares	-0.3% (282.1M → 281.3M)	—	20 → 19 (-1)	—
redeposit_shares	-0.1% (242.5M → 242.3M)	—	18 → 17 (-1)	—
claim_rewards	+1.5% (244.9M → 248.5M)	—	14 → 13 (-1)	—
withdraw_shares	+0.2% (336.4M → 337.0M)	—	19 → 18 (-1)	—
join_farms	-2.3% (172.4M → 168.4M)	—	16 → 15 (-1)	—
add_liquidity_and_join_farms	-2.2% (517.9M → 506.5M)	—	46 → 45 (-1)	25 → 26 (+1)
exit_farms	+0.8% (107.7M → 108.6M)	—	13 → 12 (-1)	—
add_liquidity_stableswap_omnipool_and_join_farms	⚠️ -22.5% (2.7B → 2.1B)	—	80 → 77 (-3)	—
remove_liquidity_stableswap_omnipool_and_exit_farms	⚠️ -35.2% (1.7B → 1.1B)	—	64 → 63 (-1)	—
price_adjustment_get	-1.8% (177.3M → 174.2M)	—	—	—

pallet_stableswap

Extrinsic	RefTime	Proof Size	Reads	Writes
create_pool	+0.6% (54.4M → 54.7M)	—	—	—
create_pool_with_pegs	-0.8% (100.2M → 99.4M)	—	—	—
add_assets_liquidity	-0.5% (2.8B → 2.8B)	—	54 → 53 (-1)	18 → 17 (-1)
add_liquidity_shares	-1.1% (2.1B → 2.1B)	—	38 → 37 (-1)	10 → 9 (-1)
remove_liquidity_one_asset	-1.3% (2.1B → 2.1B)	—	35 → 34 (-1)	9 → 8 (-1)
remove_liquidity	-0.4% (1.7B → 1.7B)	—	56 → 55 (-1)	14 → 13 (-1)
withdraw_asset_amount	-4.3% (2.6B → 2.5B)	—	36 → 35 (-1)	8 → 7 (-1)
sell	-1.0% (2.1B → 2.1B)	—	—	—
buy	-1.1% (2.0B → 2.0B)	—	—	—
set_asset_tradable_state	+6.5% (23.1M → 24.6M)	—	1 → 2 (+1)	—
update_pool_fee	-2.0% (21.4M → 21.0M)	—	—	—
update_amplification	-2.3% (22.4M → 21.9M)	—	—	—
update_asset_peg_source	-2.9% (28.7M → 27.9M)	—	—	—
update_pool_max_peg_update	-1.8% (26.1M → 25.6M)	—	—	—
router_execution_sell	+0.2% (1.2B → 1.2B)	⚠️ +18.6% (13990 → 16590)	18 → 19 (+1)	—
router_execution_buy	-0.5% (588.8M → 585.6M)	—	—	—
calculate_spot_price_with_fee	+3.5% (494.3M → 511.8M)	⚠️ +18.6% (13990 → 16590)	17 → 18 (+1)	—

New extrinsics (4)

Pallet	Extrinsic	RefTime	Proof	Reads	Writes
cumulus_pallet_parachain_system	block_weight_tx_extension_max_weight	0	0	0	0
cumulus_pallet_parachain_system	block_weight_tx_extension_stays_fraction_of_core	0	0	0	0
cumulus_pallet_parachain_system	block_weight_tx_extension_full_core	0	0	0	0
pallet_xcm	weigh_message	100.0K	0	0	0

Removed extrinsics (2)

Pallet	Extrinsic	RefTime	Proof	Reads	Writes
pallet_evm_accounts	set_ntt_minter	23.3M	0	0	1
pallet_evm_accounts	clear_ntt_minter	23.2M	0	0	1

---

Threshold: ±10%. Base Weight::from_parts(ref_time, proof_size) compared; per-unit components ignored.

[github-actions]

Crate versions that have not been updated:

• hydradx: v16.0.1

Runtime version has not been increased.