Rework sqlite: Instead of only supported one schema (defined in
`bdk_sqlite`), we have a schema per changeset type for more flexiblity.
* rm `bdk_sqlite` crate (as we don't need `bdk_sqlite::Store` anymore).
* add `sqlite` feature on `bdk_chain` which adds methods on each
changeset type for initializing tables, loading the changeset and
writing.
Rework changesets: Some callers may want to use `KeychainTxOutIndex`
where `K` may change per descriptor on every run. So we only want to
persist the last revealed indices by `DescriptorId` (which uniquely-ish
identifies the descriptor).
* rm `keychain_added` field from `keychain_txout`'s changeset.
* Add `keychain_added` to `CombinedChangeSet` (which is renamed to
`WalletChangeSet`).
Rework persistence: add back some safety and convenience when persisting
our types. Working with changeset directly (as we were doing before) can
be cumbersome.
* Intoduce `struct Persisted<T>` which wraps a type `T` which stores
staged changes to it. This adds safety when creating and or loading
`T` from db.
* `struct Persisted<T>` methods, `create`, `load` and `persist`, are
avaliable if `trait PersistWith<Db>` is implemented for `T`. `Db`
represents the database connection and `PersistWith` should be
implemented per database-type.
* For async, we have `trait PersistedAsyncWith<Db>`.
* `Wallet` has impls of `PersistedWith<rusqlite::Connection>`,
`PersistedWith<rusqlite::Transaction>` and
`PersistedWith<bdk_file_store::Store>` by default.
Rework wallet-construction: Before, we had multiple methods for loading
and creating with different input-counts so it would be unwieldly to add
more parameters in the future. This also makes it difficult to impl
`PersistWith` (which has a single method for `load` that takes in
`PersistWith::LoadParams` and a single method for `create` that takes in
`PersistWith::CreateParams`).
* Introduce a builder pattern when constructing a `Wallet`. For loading
from persistence or `ChangeSet`, we have `LoadParams`. For creating a
new wallet, we have `CreateParams`.
Both `bdk_electrum` and `bdk_esplora` now report the exact block
that the transaction is in, which removes the need for having the
old `ConfirmationTimeHeightAnchor` and `ConfirmationHeightAnchor`.
This PR introduces a new, simpler anchor type that can be modified
to support additional data in the future.
Remove `PersistBackend`, `PersistBackendAsync`, `StageExt` and
`StageExtAsync`. Remove `async` feature flag and dependency. Update
examples and wallet.
The underlying SpkTxOutIndex should not use DescriptorIds to index
because this loses the ordering relationship of the spks so queries on
subranges of keychains work.
Along with that we enforce that there is a strict 1-to-1 relationship
between descriptors and keychains. Violating this leads to an error in
insert_descriptor now.
In general I try to make the translation layer between the SpkTxOutIndex
and the KeychainTxOutIndex thinner. Ergonomics of this will be improved
in next commit.
The test from the previous commit passes.
- The KeychainTxOutIndex's internal SpkIterator now uses DescriptorId
instead of K. The DescriptorId -> K translation is made at the
KeychainTxOutIndex level.
- The keychain::Changeset is now a struct, which includes a map for last
revealed indexes, and one for newly added keychains and their
descriptor.
API changes in bdk:
- Wallet::keychains returns a `impl Iterator` instead of `BTreeMap`
- Wallet::load doesn't take descriptors anymore, since they're stored in
the db
- Wallet::new_or_load checks if the loaded descriptor from db is the
same as the provided one
API changes in bdk_chain:
- `ChangeSet` is now a struct, which includes a map for last revealed
indexes, and one for keychains and descriptors.
- `KeychainTxOutIndex::inner` returns a `SpkIterator<(DescriptorId, u32)>`
- `KeychainTxOutIndex::outpoints` returns a `impl Iterator` instead of `&BTreeSet`
- `KeychainTxOutIndex::keychains` returns a `impl Iterator` instead of
`&BTreeMap`
- `KeychainTxOutIndex::txouts` doesn't return a ExactSizeIterator
anymore
- `KeychainTxOutIndex::unbounded_spk_iter` returns an `Option`
- `KeychainTxOutIndex::next_index` returns an `Option`
- `KeychainTxOutIndex::last_revealed_indices` returns a `BTreeMap`
instead of `&BTreeMap`
- `KeychainTxOutIndex::reveal_to_target` returns an `Option`
- `KeychainTxOutIndex::reveal_next_spk` returns an `Option`
- `KeychainTxOutIndex::next_unused_spk` returns an `Option`
- `KeychainTxOutIndex::add_keychain` has been renamed to
`KeychainTxOutIndex::insert_descriptor`, and now it returns a
ChangeSet
The intention is to remove the `Update::introduce_older_blocks`
parameter and update the local chain directly with `CheckPoint`.
This simplifies the API and there is a way to do this efficiently.
e6433fb2c1526cff37e3a17cc71986759ebac30f feat(persist): Add stage_and_commit to Persist (LLFourn)
0bee46e75bc99112abc15de59984e6059ff874e5 fix(store): Remove lifetime (LLFourn)
Pull request description:
Remove gratuitous use of lifetimes in the main persistence struct `Store`. Having lifetimes on this means that you have to keep the magic bytes alive longer than the database which is particularly offensive if you have to send the database to another thread. On top of that the bytes aren't even read.
ACKs for top commit:
evanlinjin:
ACK e6433fb2c1526cff37e3a17cc71986759ebac30f
Tree-SHA512: 7f6d9d60951a8ceaee30719d0771e15632c6fad0702294af15409c5df492669a07299874ef5ee34e3d75bdecbbd41df29bced3ff16b2360d5d5c7687ef677ffc
In the example_cli we were not always committing (seemingly by mistake).
This then caused all the examples to have to compensate by manually
committing.
Previously, emissions are purely blocks + the block height. This means
emitted blocks can only connect to previous-adjacent blocks. Hence, sync
must start from genesis and include every block.
The wallet is currently created without setting any lookahead value for
the keychain. This implicitly makes it a lookahead of 0. As this is a
high-level interface we should avoid footguns and aim for a reasonable
default.
Instead of simply patching it for wallet, we alter `KeychainTxOutIndex`
to have a default lookahead value. Additionally, instead of a
per-keychain lookahead, the constructor asks for a `lookahead` value.
This avoids the footguns of having methods which allows the caller the
decrease the `lookahead` (and therefore panicing). This also simplifies
the API.
Co-authored-by: Antoine Poisot <darosior@protonmail.com>
Co-authored-by: 志宇 <hello@evanlinjin.me>
* avoid holding mutex lock over io
* document `CHANNEL_BOUND` const
* use the `relevant` variant of `batch_insert_unconfirmed`
* print elapsed time in stdout for various updates
* `bdk_chain` dependency is added. In the future, we will introduce a
separate `bdk_core` crate to contain shared types.
* replace `Emitter::new` with `from_height` and `from_checkpoint`
* `from_height` emits from the given start height
* `from_checkpoint` uses the provided cp to find agreement point
* introduce logic that ensures emitted blocks can connect with
receiver's `LocalChain`
* in our rpc example, we can now `expect()` chain updates to always
since we are using checkpoints and receiving blocks in order
For `IndexedTxGraph`:
- Remove `InsertTxItem` type (this is too complex).
- `batch_insert_relevant` now uses a simple tuple `(&tx, anchors)`.
- `batch_insert` is now also removed, as the same functionality can be
done elsewhere.
- Add internal helper method `index_tx_graph_changeset` so we don't need
to create a seprate `TxGraph` update in each method.
- `batch_insert_<relevant>_unconfirmed` no longer takes in an option of
last_seen.
- `batch_insert_unconfirmed` no longer takes a reference of a
transaction (since we apply all transactions anyway, so there is no
need to clone).
For `TxGraph`:
- Add `batch_insert_unconfirmed` method.
