志宇 [Wed, 24 May 2023 03:37:26 +0000 (11:37 +0800)]
[chain_redesign] Remove old structures
Other changes:
* The `async-https` feature of `bdk_esplora` is no longer default.
* Rename `ObservedAs` to `ChainPosition`.
* Set temporary MSRV to 1.60.0 to compile all workspace members will all
features.
志宇 [Thu, 18 May 2023 02:02:23 +0000 (10:02 +0800)]
`bdk_electrum` API improvements and simplifications
* `ElectrumUpdate::missing_full_txs` now returns a `Vec<Txid>` so we
don't keep a reference to the passed-in `graph`.
* `ElectrumUpdate::finalize*` methods now takes in `missing` txids
instead of `full_txs`. `Client::batch_transaction_get` is called
within the methods.
Other changes:
* `wallet::ChangeSet` is now made public externally. This is required as
a wallet db should implement `PersistBackend<wallet::ChangeSet>`.
There are a number of improvements that can be done, but it is in a
decent state to be usable.
Possible improvements:
* Remove requirement to retry obtaining ALL data after reorg is
detected. Transactions can be anchored to a lower block (not block
tip), and an `assume_final_depth` value can be used.
* The logic to finalize an update with confirmation time can be improved
during reorgs to not require returning an error.
* [x] I've signed all my commits
* [x] I followed the [contribution guidelines](https://github.com/bitcoindevkit/bdk/blob/master/CONTRIBUTING.md)
* [x] I ran `cargo fmt` and `cargo clippy` before committing
base64 lowered the MSRV to 1.57.0 in version 0.21.2
### Checklists
#### All Submissions:
* [x] I've signed all my commits
* [x] I followed the [contribution guidelines](https://github.com/bitcoindevkit/bdk/blob/master/CONTRIBUTING.md)
* [x] I ran `cargo fmt` and `cargo clippy` before committing
* [x] I've signed all my commits
* [x] I followed the [contribution guidelines](https://github.com/bitcoindevkit/bdk/blob/master/CONTRIBUTING.md)
* [x] I ran `cargo fmt` and `cargo clippy` before committing
As noted in https://github.com/bitcoindevkit/bdk/issues/971#issuecomment-1542408941.
Instead of relying on a `OwnedIndexer` trait to filter for relevant txouts, we move the listing and balance methods from `IndexedTxGraph` to `TxGraph` and add an additional input (list of relevant outpoints) to these methods.
This provides a simpler implementation and a more flexible API.
Previously, regardless of whether writing to persistence backend is successful or not, the logic always cleared `self.staged`. This is changed to only clear `self.staged` after successful write.
Additionally, the written changeset (if any) is returned, and `PersistBackend::write_changes` will not be called if `self.staged` is empty.
### Notes to the reviewers
Yes, slightly more boilerplate to do the same things, but less code to maintain and a much more flexible API. Very worth it IMO.
### Changelog notice
### Checklists
#### All Submissions:
* [x] I've signed all my commits
* [x] I followed the [contribution guidelines](https://github.com/bitcoindevkit/bdk/blob/master/CONTRIBUTING.md)
* [x] I ran `cargo fmt` and `cargo clippy` before committing
#### New Features:
* [x] I've added tests for the new feature
* [x] I've added docs for the new feature
志宇 [Thu, 11 May 2023 03:49:33 +0000 (11:49 +0800)]
Fix and improve `Persist::commit` method
Previously, regardless of whether writing to persistence backend is
successful or not, the logic always cleared `self.staged`. This is
changed to only clear `self.staged` after successful write.
Additionally, the written changeset (if any) is returned, and
`PersistBackend::write_changes` will not be called if `self.staged` is
empty.
志宇 [Wed, 10 May 2023 06:14:29 +0000 (14:14 +0800)]
Improve `txout` listing and balance APIs for redesigned structures
Instead of relying on a `OwnedIndexer` trait to filter for relevant
txouts, we move the listing and balance methods from `IndexedTxGraph` to
`TxGraph` and add an additional input (list of relevant outpoints) to
these methods.
This provides a simpler implementation and a more flexible API.
* Introduce a more generic version of the `keychain::persist::*` structures that only needs a single generic for the changeset type.
Additional changes:
* The `Append` trait has a new method `is_empty`.
* Introduce `Store` structure for `bdk_file_store` (which implements `PersistBackend`).
### Changelog notice
### Checklists
#### All Submissions:
* [x] I've signed all my commits
* [x] I followed the [contribution guidelines](https://github.com/bitcoindevkit/bdk/blob/master/CONTRIBUTING.md)
* [x] I ran `cargo fmt` and `cargo clippy` before committing
#### New Features:
* [x] I've added tests for the new feature
* [x] I've added docs for the new feature
[chain_redesign] Various `LocalChain` improvements
* Introduce `LocalChain::inner` method to get the inner map of block
height to hash.
* Replace `LocalChain::get_block` (which outputted `BlockId`, thus able
to return invalid representation) with `get_blockhash` that just
returns a `BlockHash`.
* Remove `TODO` comments that should be github tickets.
[chain_redesign] `BlockId` should not implement `Anchor`
If `BlockId` implements `Anchor`, the meaning is ambiguous. We cannot
tell whether it means the tx is anchors at the block, or whether it also
means the tx is confirmed at that block.
Instead, `ConfirmationHeightAnchor` and `ConfirmationTimeAnchor` structs
are introduced as non-ambiguous `Anchor` implementations.
Additionally, `TxGraph::relevant_heights` is removed because it is also
ambiguous. What heights are deemed relevant? A simpler and more flexible
method `TxGraph::all_anchors` is introduced instead.
[chain_redesign] Change behavior of `try_get_chain_position`
`TxGraph::try_get_chain_position` used to always exclude unconfirmed
transactions with last_seen value of 0. However, what is the point of
including a transaction in the graph if it cannot be part of the chain
history? Additionally, maybe sometimes we don't wish to use the
last_seen field at all.
The new behavior will consider unconfirmed transactions with last_seen
of 0.
The initial `bdk_chain` structures allowed updating to be done without blocking IO (alongside many other benefits). However, the requirement to have transactions "perfectly positioned" in our `SparseChain` increased complexity of the syncing API. Updates needed to be meticulously crafted to properly connect with the original `SparseChain`. Additionally, it forced us to keep a local copy of the "best chain" data (which may not always be needed depending on the chain source).
The redesigned structures, as introduced by this PR, fixes these shortcomings.
1. Instead of `SparseChain`, we introduce the ability to `Anchor` a transaction to multiple blocks that may or may not be in the same chain history. We expand `TxGraph` to records these anchors (while still maintaining the *monotone* nature of `TxGraph`). When updating our new `TxGraph`, we don't need to complicated *are-we-connected* checks that we need for `SparseChain`.
2. The chain source, in combination with our new`TxGraph` is all that we need to determine the "chain position" of a transaction. The chain source only needs to implement a single trait `ChainOracle`. This typically does not need to be updated (as it is remote), although we can have a special `ChainOracle` implementation that is stored locally. This only needs block height and hash information, reducing the scope of *non-monotine* structures that need to be updated.
**What is done:**
* `TxGraph` includes anchors (ids of blocks that the tx is seen in) and last-seem unix timestamp (for determining which non-confirmed tx we should consider as part of "best chain" if there are conflicts). This structure continues to be fully "monotone"; we can introduce data in any manner and not risk resulting in an inconsistent state.
* `LocalChain` includes the "checkpoint" logic of `SparseChain` but removes the `txid` data. `LocalChain` implements the `ChainOracle` trait. Any blockchain-source can also implement the `ChainOracle` trait.
* `IndexedTxGraph` is introduced and contains two fields; an internal `TxGraph` struct and a `TxIndex` implementation. These two fields will be updated atomically and can replace the functionality of `keychain::Tracker`.
**What is in-progress:**
* ~@evanlinjin: The `TxIndex` trait should not have `is_{}_relevant` methods as we cannot guarantee this across all transaction indexes. We should introduce extension traits for these (https://github.com/bitcoindevkit/bdk/pull/926#discussion_r1159286393).~
* ~@evanlinjin: `BlockAnchor` should be defined as "if this block is in chain, then this tx must be in chain". Therefore, the anchor does not provide us with the exact confirmation height of the tx. We need to introduce an extension trait `ExactConfirmationHeightAnchor` for certain operations (https://github.com/bitcoindevkit/bdk/pull/926#discussion_r1155480352).~
**What will be done external to this PR:**
* @rajarshimaitra: Persisting `indexed_tx_graph::Additions` (#937).
* @notmandatory: Update examples to use redesigned structures.
* Update `bdk::Wallet` to use the redesigned structures.
### Changelog notice
* Initial implementation of the `bdk_chain` redesign (as mentioned in #895). Currently, only the `bdk_chain` core structures are implemented. Persistence and updates to the current examples and `bdk::Wallet` will be done in separate PRs.
### Checklists
#### All Submissions:
* [x] I've signed all my commits
* [x] I followed the [contribution guidelines](https://github.com/bitcoindevkit/bdk/blob/master/CONTRIBUTING.md)
* [x] I ran `cargo fmt` and `cargo clippy` before committing
#### New Features:
* [x] I've added tests for the new feature
* [x] I've added docs for the new feature
ACKs for top commit:
rajarshimaitra:
ACK https://github.com/bitcoindevkit/bdk/pull/926/commits/b799a5728b7e08a18059ffee7fe7ee6078354977
danielabrozzoni:
Partial ACK b799a5728b7e08a18059ffee7fe7ee6078354977 - good job! I haven't looked at the tests or at the methods implementations in depth, but I have reviewed the architecture and it looks good. I have a few non-blocking questions.
* `IndexedTxGraph::try_balance` should include "confirmed and spendable"
into `confirmed` balance.
* `TxGraph::try_list_chain_unspents` filter logic should be reversed.
* `Additions` now implements `Append` and uses `Append` to implement
`append()`.
* `append()` logic enforces that `last_seen` values should only
increase.
* Test written for `append()` with `last_seen` behaviour.
[bdk_chain_redesign] Make `insert_relevant_txs` topologically-agnostic
The `insert_relevant_txs` test has also been changed to used
`KeychainTxOutIndex` so that index additions can be checked
(`SpkTxOutIndex` has no additions).
Additionally, generic bounds of some `IndexedTxGraph` list methods have
been fixed.
The `ConfirmationHeight` trait has been removed in favour of a second
method on the `Anchor` trait: `confirmation_height_upper_bound()`.
Methods `try_balance_at()` and `balance_at()` of `IndexedTxGraph` have
been removed as they do not provide additional functionality.
`IndexedTxGraph::insert_relevant_txs` now uses two loops, the first loop
indexes all transactions first. This is done as some indexes require
ancestor transactions to be indexed first and we cannot guarantee that
the input transactions are in topological order.
[bdk_chain_redesign] Rm unnecessary code and premature optimisation
* Remove `chain_oracle::CacheBackend` for now as it is not used.
* `SparseChain` does not need to implement `ChainOracle`.
* Remove filter predicate for `list..` methods of `TxGraph` and
`IndexedTxGraph` as this is premature optimisation.
* `Append` can be implemented for all `BTreeMap`s and `BTreeSet`s,
instead of only `local_chain::ChangeSet`.
[bdk_chain_redesign] Rm `HashSet` from `TxGraph::relevant_heights`
The `HashSet` was used for iterating without duplicate items. However,
since `anchors` is a `BTreeSet`, heights are in order. So a single
variable tracking last height will be sufficient.
This is mostly copying over the relevant tests from `SparseChain`.
Changes are made to `local_chain::ChangeSet` to re-add the ability to
remove blocks.
Previously, I have misunderstood the definition of anchor. If a tx is
anchored in a block, it does not necessarily mean it is confirmed in
that block. The tx can be confirmed in an ancestor block of the anchor
block.
With this new definition, we need a new trait `ConfirmationHeight` that
has one method `confirmation_height`. This trait can be used to extend
`Anchor` for those implementations that can give us the exact
conirmation height of a tx (which is useful in most cases).
Another change is to add another variant to the `ObservedAs` enum;
`ObservedAs::ConfirmedImplicit(A)`. If a tx does not have an anchor, but
another tx that spends it has an anchor that in in the best chain, we
can assume that tx is also in the best chain. The logic of
`TxGraph::try_get_chain_position` is also changed to reflect this.
Some methods from `IndexedTxGraph` have been moved to `TxGraph` as they
do not require the `Indexer`. Some `TxGraph` methods have been renamed
for clarity and consistency.
When constructing the `Condition` struct we recursively call `get_condition` on all the items in a threshold and short-circuit if there's an error somewhere (for example, because the policy-path hasn't been provided for a specific threshold).
This can cause issues when the user doesn't care about a subtree, because we still try to call `get_condition` on all the items and fail if something is missing, even if the specific subtree isn't selected and won't be used later on.
This commit changes the logic so that we first filter only the `selected` items, and then unwrap the error using the question mark. If errors happened somewhere else they will be ignored, as they should.
### Notes to the reviewers
I think it makes sense to backport this to `0.27`: even though it's not a critical issue (and there's a workaround[^1] for the bug) it may be a while before the new `1.0` is released. I wouldn't do a release just for this, but I would just leave it there and maybe in a few weeks if there are other fixes to be backported to pre-1.0 they could all be released.
### Changelog notice
- Fixed a bug in the policy condition calculation
### Checklists
#### All Submissions:
* [x] I've signed all my commits
* [x] I followed the [contribution guidelines](https://github.com/bitcoindevkit/bdk/blob/master/CONTRIBUTING.md)
* [x] I ran `cargo fmt` and `cargo clippy` before committing
#### New Features:
* [ ] I've added tests for the new feature
* [ ] I've added docs for the new feature
#### Bugfixes:
* [ ] This pull request breaks the existing API
* [x] I've added tests to reproduce the issue which are now passing
* [ ] I'm linking the issue being fixed by this PR
[^1]: The workaround is to simply set the items in the policy tree even if they won't be used. For example, if the item causing troubles is a `thresh(1, ...)` just set `[0]` in the policy path for that id.
The problem with the previous `ChainOracle` interface is that it had no
guarantee for consistency. For example, a block deemed to be part of the
"best chain" can be reorged out. So when `ChainOracle` is called
multiple times for an operation (such as getting the UTXO set), the
returned result may be inconsistent.
This PR changes `ChainOracle::is_block_in_chain` to take in another
input `static_block`, ensuring `block` is an ancestor of `static_block`.
Thus, if `static_block` is consistent across the operation, the result
will be consistent also.
`is_block_in_chain` now returns `Option<bool>`. The `None` case means
that the oracle implementation cannot determine whether block is an
ancestor of static block. `IndexedTxGraph::list_chain_txouts` handles
this case by checking child spends that are in chain, and if so, the
parent tx must be in chain too.
[bdk_chain_redesign] Better names, comments and generic bounds
* Instead of implementing `ChainPosition` for `ObservedIn<BlockId>` to
use `FullTxOut` methods (`is_spendable_at` and `is_mature`), we create
alternative versions of those methods that require bounds with `Anchor`.
This removes all `ObservedIn<A>: ChainPosition` bounds for methods of
`IndexedTxGraph`.
Introduce `chain_oracle::Cache` which is a cache for requests to the
chain oracle. `ChainOracle` has also been moved to the `chain_oracle`
module.
Introduce `get_tip_in_best_chain` method to the `ChainOracle` trait.
This allows for guaranteeing that chain state can be consistent across
operations with `IndexedTxGraph`.
Alekos Filini [Fri, 31 Mar 2023 18:08:23 +0000 (20:08 +0200)]
Fix policy condition calculation
When constructing the `Condition` struct we recursively call
`get_condition` on all the items in a threshold and short-circuit if
there's an error somewhere (for example, because the policy-path hasn't
been provided for a specific threshold).
This can cause issues when the user doesn't care about a subtree, because
we still try to call `get_condition` on all the items and fail if
something is missing, even if the specific subtree isn't selected and
won't be used later on.
This commit changes the logic so that we first filter only the `selected`
items, and then unwrap the error using the question mark. If errors
happened somewhere else they will be ignored, as it should.
* Introduce `GraphedTx` struct to access transaction data of graphed
transactions.
* Ability to insert/access anchors and "seen at" values for graphed
transactions.
* `Additions` now records changes to anchors and last_seen_at.
projects / bdk / log