This fixes the following bug:
Due to complexities related to non-linear scanning, checkpoints are only
added to the wallet's commitment tree in cases where there are notes
discovered within a scanned block. At present, the `shardtree` API only
makes it possible to add multiple checkpoints of the same tree state
when adding checkpoints at the chain tip, and this functionality is not
used by `zcash_client_backend` because we perform checkpoint insertion
in batches that may contain gaps in the case that multiple blocks
contain no Sapling notes. While it would be possible to fix this by
altering the `shardtree` API to permit explicit insertion of multiple
checkpoints of the same tree state at a given note position, this fix
takes a simpler approach.
Instead of ensuring that a checkpoint exists at every block and
computing the required checkpoint depth directly from the minimum number
of confirmations required when attempting a spend, we alter the
`WalletCommitmentTrees` API to allow internal information of the note
commitment tree to be used to determine this checkpoint depth, given the
minimum number of commitments as an argument. This allows us to select a
usable checkpoint from the sparse checkpoint set that resulted from the
sparse insertion of checkpoints described above.
The `v_sapling_shard_scan_ranges` view pairs every scan range with every
shard range, such that each row shows an overlapping pair.
For the complete shards, this is an overlap check between two ranges,
which the previous query was performing correctly (if verbosely).
For the last incomplete shard, we have a half-open range that needs to
be handled separately. The previous query only handled the case where a
scan range was contained within the last shard, and did not handle the
case where the scan range contained the last shard.
This led to a puzzling bug, where `WalletDb::get_wallet_summary` was
sometimes treating any note received within the last shard as part of
the wallet's pending balance. If the wallet's scan queue contained a
range that encompassed the last incomplete shard, the bug in the
`v_sapling_shard_scan_ranges` view meant that it omitted any mention of
the last shard, which translated into these notes being considered
unmined when joining `sapling_received_notes` against the sub-view
`v_sapling_shards_scan_state`.
The bug was made harder to diagnose due to the previous commit's bug
that was causing scan ranges to not be correctly merged; this resulted
in smaller scan ranges that were more likely to be contained within the
last shard, making it visible in `v_sapling_shard_scan_ranges` and
enabling notes to be detected as mined.
The fixed view uses a simpler query that enables us to handle complete
and incomplete shards together.
Time spent investigating and fixing: 4.5 hours
Co-authored-by: Kris Nuttycombe <kris@nutty.land>
In the sandblasting block ranges, shard trees only cover a few hundred
blocks at most. When scanning block ranges much larger than this, it is
likely that when a note is discovered, its parent shard tree is entirely
contained within the scan range. In this situation, `extended_range`
would be set to `range`, and then because an extended range existed,
ranges with `FoundNote` priority would always be created - that in this
case are empty.
In an effectively-linear-scanning wallet situation, this leads to a
`SpanningTree` being constructed with adjacent `Scanned` ranges,
separated by empty `FoundNote` ranges, which it was unable to merge.
We address this by both preventing generation of the empty `FoundNote`
ranges, and by defensively fixing `SpanningTree::into_vec` to skip empty
ranges.
Co-authored-by: Kris Nuttycombe <kris@nutty.land>
This modifies `update_chain_tip` and `scan_complete` to ensure that
newly created scan ranged do not extend below the wallet birthday
height.
Fixes#947