[thomas-kroes]

Summary

Fixes #5 — any real reorg caused chain_reorged_impl to walk %elders all the way down and revert the DB to height 0, as seen on the height-112,500 mainnet fork.

Two defects combined (both confirmed against +get-elders in dumbnet's consensus.hoon):

• The elders peek returns the height of the oldest entry with ids ordered newest → oldest, but the local comparison window was anchored at that height, so every compared pair was offset by len − 1 blocks and a match was essentially impossible.
• Even on a match, break only exited the inner for; the outer while kept walking to genesis and the fall-through revert_to(0) wiped the database.

Fix

Every elder id is a block on the node's canonical chain, and block ids are unique hashes — so no height alignment is needed at all: the highest local block whose id appears in the elders batch is the common ancestor. Each batch is now checked with a single indexed id IN (...) lookup (primary key), and on a match we immediately revert_to(ancestor.height + 1). If a batch has no overlap, the walk continues from the oldest elder; empty or non-descending responses error out (UnableToPullElders) instead of looping forever. Reverting to genesis now only happens when the chains genuinely share no blocks.

Test plan

• New regression tests with a mock Scryable that mirrors +get-elders semantics (newest → oldest ids, oldest height returned, 24-entry cap, genesis stop):
  • shallow reorg resolved within one elders batch
  • deep reorg walking multiple batches (the 112,500-fork shape)
  • no common ancestor → full revert to genesis
• cargo test --features binary --all-targets (31 passed)
• cargo clippy --features binary --all-targets (no new warnings), cargo fmt --check

[cursor]

Bugbot is not enabled for your account, so this pull request was not reviewed.

Enable Bugbot in the Cursor dashboard to get automatic reviews on future PRs.