Prior to the scan-before-sync changes, the wallet was able to assume
that the maximum scanned block height at the time of the spend was
within a few blocks of the chain tip. However, under linear scanning
after the spend-before-sync changes this invariant no longer holds,
resulting in a situation where in linear sync conditions the wallet
could attempt to create transactions with already-past expiry heights.
This change separates the notion of "chain tip" from "max scanned
height", relying upon the `scan_queue` table to maintain the wallet's
view of the consensus chain height and using information from the
`blocks` table only in situations where the latest and/or earliest
scanned height is required.
As part of this change, the `WalletRead` interface is also modified to
disambiguate these concepts.
In general, it is preferable to use globally relevant identifiers where
possible. This PR removes the `WalletRead::TxRef` associated type in
favor of using `TxId` directly for the transaction identifier, and
restricts the use of the `NoteRef` type to those scenarios where the
result of one query is intended to be used directly as the input to
another query.
Closes#834
The `add_checkpoint` method is intended to be idempotent. In the case
that we add a checkpoint at an already-checkpointed block height, we
should only raise an error in the case that the note commitment tree
position or the set of notes spent in the checkpointed block has
changed.
This implements a priority queue backed by the wallet database for scan
range ordering. The scan queue is updated on each call to `put_blocks`
or to `update_chain_tip`.
Instead of calling `put_block` for each block scanned,
`scan_cached_blocks` will now defer the block writes until the scan of a
batch is complete and will perform the block writes and note commitment
tree updates all within a single transaction.
This should ordinarily be fine in terms of memory consumption, because
the block data being saved is pruned to only that spend an output
information that is related to transactions in the wallet, which will
normally be sparse enough that the block range size that is appropriate
for a given platform to run within a batch scanner adequately bounds the
memory consumption of this pruned representation.
There are cases where we wish to return informaiton that is relevant to
a specific shielded protocol and `Transparent` is an invalid case. This
is a minor preparatory refactoring that makes this distinction
expressible.
In preparation for out-of-order range-based scanning, it is necessary
to ensure that the size of the Sapling note commitment tree is carried
along through the scan process and that stored blocks are always
persisted with the updated note commitment tree size.
`rusqlite` includes a mechanism for creating prepared statements that
automatically caches them and reuses the caches when possible. This
means that it's unnecessary for us to do our own caching, and also
offers a minor performance improvement in that we don't need to eagerly
prepare statements that we may not execute in the lifetime of a given
`WalletDb` object. It also improves code locality, because the prepared
statements are now adjacent in the code to the parameter assignment
blocks that correspond to those statements.
This also updates a number of `put_x` methods to use sqlite upsert
functionality via the `ON CONFLICT` clause, instead of having to perform
separate inserts and updates.
Memos may be absent for both sent and received notes in cases where only
compact block information has been used to populate the wallet database.
This fixes a potential crash in the case that we attempt to decode a
SQLite `NULL` as a byte array. It does, however, introduce a slight
semantic confusion that will need to be considered in the case of future
updates where a note may not have an associated memo; at present, the
only reason we might not have the memo is that we might not have
retrieved the full transaction information from the chain, but in the
future there might be other possible reasons for this absence.
Fixes#384
This is in preparation for extraction into the `incrementalmerkletree`
crate, which is not Sapling-specific and therefore cannot hard-code
the depths of these data structures.
Kris Nuttycombe
Jack Grigg
Kris Nuttycombe
str4d