Decode block events with the metadata at that block

[vedhavyas]

We were pinning one metadata at chain head and using it to decode events from any block. That's wrong across runtime upgrades: mainnet went from spec 9 to 11 and the Balances pallet inserted event variants along the way, shifting Burned from index 11 to 12, Issued 15 to 17 and Rescinded 16 to 18. Decoding an older block with head metadata reads those variants as something else, consumes the wrong number of bytes, and every event after the first shifted one in that block is garbled or dropped. Since balances.Burned fires on essentially any block with a transfer, most of history was affected, and the staking paths swallow decode errors, so a from-genesis reindex would quietly under-populate share prices, nominators and the deposit tables rather than fail.

So now each block's events are decoded with the metadata that block actually ran. I read the runtime version at the block and cache one metadata per spec version, which is bounded by the number of runtimes the chain has had. events() no longer goes through subxt's events client (it can't take explicit metadata) and instead reads System::Events directly and decodes against the cached metadata; events_from_segments gets the same treatment. The client-level V14 pin stays for storage and constants.

This costs one extra state_getRuntimeVersion per block, which I think is the right trade for not silently losing events.

The two staking tests that were failing on this now pass against the same blocks they always read, and I made them surface the decode error instead of swallowing it, which is why this showed up as a bare "not found" before. I also added a test on block 6721910 that pins the mechanism: it has balances.Burned at position 1 and DomainEpochCompleted at 18, so under head metadata the whole tail of the block disappears.

[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.}

Want higher recall? High effort reviews run extra passes and find more bugs. A team admin can switch effort levels in the Cursor dashboard.

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit 61a1263. Configure here.}

[cursor]

Wrong metadata on upgrade blocks

Medium Severity

events() and events_from_segments resolve metadata via metadata.at(self.hash), which reads post-execution state. A block runs on the parent runtime, so on a CodeUpdated block this returns the new metadata while events were still encoded with the old one—the same variant-shift corruption this PR aims to fix, narrowed to each upgrade block.

Additional Locations (2)

• shared/src/subspace.rs#L273-L274
• shared/src/subspace.rs#L374-L391

 

^{Reviewed by Cursor Bugbot for commit 61a1263. Configure here.}