947 lines
40 KiB
Rust
947 lines
40 KiB
Rust
//! [`KeychainTxOutIndex`] controls how script pubkeys are revealed for multiple keychains and
|
|
//! indexes [`TxOut`]s with them.
|
|
|
|
use crate::{
|
|
collections::*,
|
|
miniscript::{Descriptor, DescriptorPublicKey},
|
|
spk_iter::BIP32_MAX_INDEX,
|
|
DescriptorExt, DescriptorId, Indexed, Indexer, KeychainIndexed, SpkIterator, SpkTxOutIndex,
|
|
};
|
|
use alloc::{borrow::ToOwned, vec::Vec};
|
|
use bitcoin::{Amount, OutPoint, Script, ScriptBuf, SignedAmount, Transaction, TxOut, Txid};
|
|
use core::{
|
|
fmt::Debug,
|
|
ops::{Bound, RangeBounds},
|
|
};
|
|
|
|
use crate::Append;
|
|
|
|
/// The default lookahead for a [`KeychainTxOutIndex`]
|
|
pub const DEFAULT_LOOKAHEAD: u32 = 25;
|
|
|
|
/// [`KeychainTxOutIndex`] controls how script pubkeys are revealed for multiple keychains, and
|
|
/// indexes [`TxOut`]s with them.
|
|
///
|
|
/// A single keychain is a chain of script pubkeys derived from a single [`Descriptor`]. Keychains
|
|
/// are identified using the `K` generic. Script pubkeys are identified by the keychain that they
|
|
/// are derived from `K`, as well as the derivation index `u32`.
|
|
///
|
|
/// There is a strict 1-to-1 relationship between descriptors and keychains. Each keychain has one
|
|
/// and only one descriptor and each descriptor has one and only one keychain. The
|
|
/// [`insert_descriptor`] method will return an error if you try and violate this invariant. This
|
|
/// rule is a proxy for a stronger rule: no two descriptors should produce the same script pubkey.
|
|
/// Having two descriptors produce the same script pubkey should cause whichever keychain derives
|
|
/// the script pubkey first to be the effective owner of it but you should not rely on this
|
|
/// behaviour. ⚠ It is up you, the developer, not to violate this invariant.
|
|
///
|
|
/// # Revealed script pubkeys
|
|
///
|
|
/// Tracking how script pubkeys are revealed is useful for collecting chain data. For example, if
|
|
/// the user has requested 5 script pubkeys (to receive money with), we only need to use those
|
|
/// script pubkeys to scan for chain data.
|
|
///
|
|
/// Call [`reveal_to_target`] or [`reveal_next_spk`] to reveal more script pubkeys.
|
|
/// Call [`revealed_keychain_spks`] or [`revealed_spks`] to iterate through revealed script pubkeys.
|
|
///
|
|
/// # Lookahead script pubkeys
|
|
///
|
|
/// When an user first recovers a wallet (i.e. from a recovery phrase and/or descriptor), we will
|
|
/// NOT have knowledge of which script pubkeys are revealed. So when we index a transaction or
|
|
/// txout (using [`index_tx`]/[`index_txout`]) we scan the txouts against script pubkeys derived
|
|
/// above the last revealed index. These additionally-derived script pubkeys are called the
|
|
/// lookahead.
|
|
///
|
|
/// The [`KeychainTxOutIndex`] is constructed with the `lookahead` and cannot be altered. See
|
|
/// [`DEFAULT_LOOKAHEAD`] for the value used in the `Default` implementation. Use [`new`] to set a
|
|
/// custom `lookahead`.
|
|
///
|
|
/// # Unbounded script pubkey iterator
|
|
///
|
|
/// For script-pubkey-based chain sources (such as Electrum/Esplora), an initial scan is best done
|
|
/// by iterating though derived script pubkeys one by one and requesting transaction histories for
|
|
/// each script pubkey. We will stop after x-number of script pubkeys have empty histories. An
|
|
/// unbounded script pubkey iterator is useful to pass to such a chain source because it doesn't
|
|
/// require holding a reference to the index.
|
|
///
|
|
/// Call [`unbounded_spk_iter`] to get an unbounded script pubkey iterator for a given keychain.
|
|
/// Call [`all_unbounded_spk_iters`] to get unbounded script pubkey iterators for all keychains.
|
|
///
|
|
/// # Change sets
|
|
///
|
|
/// Methods that can update the last revealed index or add keychains will return [`ChangeSet`] to report
|
|
/// these changes. This should be persisted for future recovery.
|
|
///
|
|
/// ## Synopsis
|
|
///
|
|
/// ```
|
|
/// use bdk_chain::indexer::keychain_txout::KeychainTxOutIndex;
|
|
/// # use bdk_chain::{ miniscript::{Descriptor, DescriptorPublicKey} };
|
|
/// # use core::str::FromStr;
|
|
///
|
|
/// // imagine our service has internal and external addresses but also addresses for users
|
|
/// #[derive(Clone, Debug, PartialEq, Eq, Ord, PartialOrd)]
|
|
/// enum MyKeychain {
|
|
/// External,
|
|
/// Internal,
|
|
/// MyAppUser {
|
|
/// user_id: u32
|
|
/// }
|
|
/// }
|
|
///
|
|
/// let mut txout_index = KeychainTxOutIndex::<MyKeychain>::default();
|
|
///
|
|
/// # let secp = bdk_chain::bitcoin::secp256k1::Secp256k1::signing_only();
|
|
/// # let (external_descriptor,_) = Descriptor::<DescriptorPublicKey>::parse_descriptor(&secp, "tr([73c5da0a/86'/0'/0']xprv9xgqHN7yz9MwCkxsBPN5qetuNdQSUttZNKw1dcYTV4mkaAFiBVGQziHs3NRSWMkCzvgjEe3n9xV8oYywvM8at9yRqyaZVz6TYYhX98VjsUk/0/*)").unwrap();
|
|
/// # let (internal_descriptor,_) = Descriptor::<DescriptorPublicKey>::parse_descriptor(&secp, "tr([73c5da0a/86'/0'/0']xprv9xgqHN7yz9MwCkxsBPN5qetuNdQSUttZNKw1dcYTV4mkaAFiBVGQziHs3NRSWMkCzvgjEe3n9xV8oYywvM8at9yRqyaZVz6TYYhX98VjsUk/1/*)").unwrap();
|
|
/// # let (descriptor_42, _) = Descriptor::<DescriptorPublicKey>::parse_descriptor(&secp, "tr([73c5da0a/86'/0'/0']xprv9xgqHN7yz9MwCkxsBPN5qetuNdQSUttZNKw1dcYTV4mkaAFiBVGQziHs3NRSWMkCzvgjEe3n9xV8oYywvM8at9yRqyaZVz6TYYhX98VjsUk/2/*)").unwrap();
|
|
/// let _ = txout_index.insert_descriptor(MyKeychain::External, external_descriptor)?;
|
|
/// let _ = txout_index.insert_descriptor(MyKeychain::Internal, internal_descriptor)?;
|
|
/// let _ = txout_index.insert_descriptor(MyKeychain::MyAppUser { user_id: 42 }, descriptor_42)?;
|
|
///
|
|
/// let new_spk_for_user = txout_index.reveal_next_spk(&MyKeychain::MyAppUser{ user_id: 42 });
|
|
/// # Ok::<_, bdk_chain::indexer::keychain_txout::InsertDescriptorError<_>>(())
|
|
/// ```
|
|
///
|
|
/// [`Ord`]: core::cmp::Ord
|
|
/// [`SpkTxOutIndex`]: crate::spk_txout_index::SpkTxOutIndex
|
|
/// [`Descriptor`]: crate::miniscript::Descriptor
|
|
/// [`reveal_to_target`]: Self::reveal_to_target
|
|
/// [`reveal_next_spk`]: Self::reveal_next_spk
|
|
/// [`revealed_keychain_spks`]: Self::revealed_keychain_spks
|
|
/// [`revealed_spks`]: Self::revealed_spks
|
|
/// [`index_tx`]: Self::index_tx
|
|
/// [`index_txout`]: Self::index_txout
|
|
/// [`new`]: Self::new
|
|
/// [`unbounded_spk_iter`]: Self::unbounded_spk_iter
|
|
/// [`all_unbounded_spk_iters`]: Self::all_unbounded_spk_iters
|
|
/// [`outpoints`]: Self::outpoints
|
|
/// [`txouts`]: Self::txouts
|
|
/// [`unused_spks`]: Self::unused_spks
|
|
/// [`insert_descriptor`]: Self::insert_descriptor
|
|
#[derive(Clone, Debug)]
|
|
pub struct KeychainTxOutIndex<K> {
|
|
inner: SpkTxOutIndex<(K, u32)>,
|
|
keychain_to_descriptor_id: BTreeMap<K, DescriptorId>,
|
|
descriptor_id_to_keychain: HashMap<DescriptorId, K>,
|
|
descriptors: HashMap<DescriptorId, Descriptor<DescriptorPublicKey>>,
|
|
last_revealed: HashMap<DescriptorId, u32>,
|
|
lookahead: u32,
|
|
}
|
|
|
|
impl<K> Default for KeychainTxOutIndex<K> {
|
|
fn default() -> Self {
|
|
Self::new(DEFAULT_LOOKAHEAD)
|
|
}
|
|
}
|
|
|
|
impl<K: Clone + Ord + Debug> Indexer for KeychainTxOutIndex<K> {
|
|
type ChangeSet = ChangeSet<K>;
|
|
|
|
fn index_txout(&mut self, outpoint: OutPoint, txout: &TxOut) -> Self::ChangeSet {
|
|
let mut changeset = ChangeSet::default();
|
|
if let Some((keychain, index)) = self.inner.scan_txout(outpoint, txout).cloned() {
|
|
let did = self
|
|
.keychain_to_descriptor_id
|
|
.get(&keychain)
|
|
.expect("invariant");
|
|
if self.last_revealed.get(did) < Some(&index) {
|
|
self.last_revealed.insert(*did, index);
|
|
changeset.last_revealed.insert(*did, index);
|
|
self.replenish_inner_index(*did, &keychain, self.lookahead);
|
|
}
|
|
}
|
|
changeset
|
|
}
|
|
|
|
fn index_tx(&mut self, tx: &bitcoin::Transaction) -> Self::ChangeSet {
|
|
let mut changeset = ChangeSet::<K>::default();
|
|
let txid = tx.compute_txid();
|
|
for (op, txout) in tx.output.iter().enumerate() {
|
|
changeset.append(self.index_txout(OutPoint::new(txid, op as u32), txout));
|
|
}
|
|
changeset
|
|
}
|
|
|
|
fn initial_changeset(&self) -> Self::ChangeSet {
|
|
ChangeSet {
|
|
keychains_added: self
|
|
.keychains()
|
|
.map(|(k, v)| (k.clone(), v.clone()))
|
|
.collect(),
|
|
last_revealed: self.last_revealed.clone().into_iter().collect(),
|
|
}
|
|
}
|
|
|
|
fn apply_changeset(&mut self, changeset: Self::ChangeSet) {
|
|
self.apply_changeset(changeset)
|
|
}
|
|
|
|
fn is_tx_relevant(&self, tx: &bitcoin::Transaction) -> bool {
|
|
self.inner.is_relevant(tx)
|
|
}
|
|
}
|
|
|
|
impl<K> KeychainTxOutIndex<K> {
|
|
/// Construct a [`KeychainTxOutIndex`] with the given `lookahead`.
|
|
///
|
|
/// The `lookahead` is the number of script pubkeys to derive and cache from the internal
|
|
/// descriptors over and above the last revealed script index. Without a lookahead the index
|
|
/// will miss outputs you own when processing transactions whose output script pubkeys lie
|
|
/// beyond the last revealed index. In certain situations, such as when performing an initial
|
|
/// scan of the blockchain during wallet import, it may be uncertain or unknown what the index
|
|
/// of the last revealed script pubkey actually is.
|
|
///
|
|
/// Refer to [struct-level docs](KeychainTxOutIndex) for more about `lookahead`.
|
|
pub fn new(lookahead: u32) -> Self {
|
|
Self {
|
|
inner: SpkTxOutIndex::default(),
|
|
keychain_to_descriptor_id: Default::default(),
|
|
descriptors: Default::default(),
|
|
descriptor_id_to_keychain: Default::default(),
|
|
last_revealed: Default::default(),
|
|
lookahead,
|
|
}
|
|
}
|
|
}
|
|
|
|
/// Methods that are *re-exposed* from the internal [`SpkTxOutIndex`].
|
|
impl<K: Clone + Ord + Debug> KeychainTxOutIndex<K> {
|
|
/// Return a reference to the internal [`SpkTxOutIndex`].
|
|
///
|
|
/// **WARNING**: The internal index will contain lookahead spks. Refer to
|
|
/// [struct-level docs](KeychainTxOutIndex) for more about `lookahead`.
|
|
pub fn inner(&self) -> &SpkTxOutIndex<(K, u32)> {
|
|
&self.inner
|
|
}
|
|
|
|
/// Get the set of indexed outpoints, corresponding to tracked keychains.
|
|
pub fn outpoints(&self) -> &BTreeSet<KeychainIndexed<K, OutPoint>> {
|
|
self.inner.outpoints()
|
|
}
|
|
|
|
/// Iterate over known txouts that spend to tracked script pubkeys.
|
|
pub fn txouts(
|
|
&self,
|
|
) -> impl DoubleEndedIterator<Item = KeychainIndexed<K, (OutPoint, &TxOut)>> + ExactSizeIterator
|
|
{
|
|
self.inner
|
|
.txouts()
|
|
.map(|(index, op, txout)| (index.clone(), (op, txout)))
|
|
}
|
|
|
|
/// Finds all txouts on a transaction that has previously been scanned and indexed.
|
|
pub fn txouts_in_tx(
|
|
&self,
|
|
txid: Txid,
|
|
) -> impl DoubleEndedIterator<Item = KeychainIndexed<K, (OutPoint, &TxOut)>> {
|
|
self.inner
|
|
.txouts_in_tx(txid)
|
|
.map(|(index, op, txout)| (index.clone(), (op, txout)))
|
|
}
|
|
|
|
/// Return the [`TxOut`] of `outpoint` if it has been indexed, and if it corresponds to a
|
|
/// tracked keychain.
|
|
///
|
|
/// The associated keychain and keychain index of the txout's spk is also returned.
|
|
///
|
|
/// This calls [`SpkTxOutIndex::txout`] internally.
|
|
pub fn txout(&self, outpoint: OutPoint) -> Option<KeychainIndexed<K, &TxOut>> {
|
|
self.inner
|
|
.txout(outpoint)
|
|
.map(|(index, txout)| (index.clone(), txout))
|
|
}
|
|
|
|
/// Return the script that exists under the given `keychain`'s `index`.
|
|
///
|
|
/// This calls [`SpkTxOutIndex::spk_at_index`] internally.
|
|
pub fn spk_at_index(&self, keychain: K, index: u32) -> Option<&Script> {
|
|
self.inner.spk_at_index(&(keychain.clone(), index))
|
|
}
|
|
|
|
/// Returns the keychain and keychain index associated with the spk.
|
|
///
|
|
/// This calls [`SpkTxOutIndex::index_of_spk`] internally.
|
|
pub fn index_of_spk(&self, script: &Script) -> Option<&(K, u32)> {
|
|
self.inner.index_of_spk(script)
|
|
}
|
|
|
|
/// Returns whether the spk under the `keychain`'s `index` has been used.
|
|
///
|
|
/// Here, "unused" means that after the script pubkey was stored in the index, the index has
|
|
/// never scanned a transaction output with it.
|
|
///
|
|
/// This calls [`SpkTxOutIndex::is_used`] internally.
|
|
pub fn is_used(&self, keychain: K, index: u32) -> bool {
|
|
self.inner.is_used(&(keychain, index))
|
|
}
|
|
|
|
/// Marks the script pubkey at `index` as used even though the tracker hasn't seen an output
|
|
/// with it.
|
|
///
|
|
/// This only has an effect when the `index` had been added to `self` already and was unused.
|
|
///
|
|
/// Returns whether the spk under the given `keychain` and `index` is successfully
|
|
/// marked as used. Returns false either when there is no descriptor under the given
|
|
/// keychain, or when the spk is already marked as used.
|
|
///
|
|
/// This is useful when you want to reserve a script pubkey for something but don't want to add
|
|
/// the transaction output using it to the index yet. Other callers will consider `index` on
|
|
/// `keychain` used until you call [`unmark_used`].
|
|
///
|
|
/// This calls [`SpkTxOutIndex::mark_used`] internally.
|
|
///
|
|
/// [`unmark_used`]: Self::unmark_used
|
|
pub fn mark_used(&mut self, keychain: K, index: u32) -> bool {
|
|
self.inner.mark_used(&(keychain, index))
|
|
}
|
|
|
|
/// Undoes the effect of [`mark_used`]. Returns whether the `index` is inserted back into
|
|
/// `unused`.
|
|
///
|
|
/// Note that if `self` has scanned an output with this script pubkey, then this will have no
|
|
/// effect.
|
|
///
|
|
/// This calls [`SpkTxOutIndex::unmark_used`] internally.
|
|
///
|
|
/// [`mark_used`]: Self::mark_used
|
|
pub fn unmark_used(&mut self, keychain: K, index: u32) -> bool {
|
|
self.inner.unmark_used(&(keychain, index))
|
|
}
|
|
|
|
/// Computes the total value transfer effect `tx` has on the script pubkeys belonging to the
|
|
/// keychains in `range`. Value is *sent* when a script pubkey in the `range` is on an input and
|
|
/// *received* when it is on an output. For `sent` to be computed correctly, the output being
|
|
/// spent must have already been scanned by the index. Calculating received just uses the
|
|
/// [`Transaction`] outputs directly, so it will be correct even if it has not been scanned.
|
|
pub fn sent_and_received(
|
|
&self,
|
|
tx: &Transaction,
|
|
range: impl RangeBounds<K>,
|
|
) -> (Amount, Amount) {
|
|
self.inner
|
|
.sent_and_received(tx, self.map_to_inner_bounds(range))
|
|
}
|
|
|
|
/// Computes the net value that this transaction gives to the script pubkeys in the index and
|
|
/// *takes* from the transaction outputs in the index. Shorthand for calling
|
|
/// [`sent_and_received`] and subtracting sent from received.
|
|
///
|
|
/// This calls [`SpkTxOutIndex::net_value`] internally.
|
|
///
|
|
/// [`sent_and_received`]: Self::sent_and_received
|
|
pub fn net_value(&self, tx: &Transaction, range: impl RangeBounds<K>) -> SignedAmount {
|
|
self.inner.net_value(tx, self.map_to_inner_bounds(range))
|
|
}
|
|
}
|
|
|
|
impl<K: Clone + Ord + Debug> KeychainTxOutIndex<K> {
|
|
/// Return all keychains and their corresponding descriptors.
|
|
pub fn keychains(
|
|
&self,
|
|
) -> impl DoubleEndedIterator<Item = (&K, &Descriptor<DescriptorPublicKey>)> + ExactSizeIterator + '_
|
|
{
|
|
self.keychain_to_descriptor_id
|
|
.iter()
|
|
.map(|(k, did)| (k, self.descriptors.get(did).expect("invariant")))
|
|
}
|
|
|
|
/// Insert a descriptor with a keychain associated to it.
|
|
///
|
|
/// Adding a descriptor means you will be able to derive new script pubkeys under it and the
|
|
/// txout index will discover transaction outputs with those script pubkeys (once they've been
|
|
/// derived and added to the index).
|
|
///
|
|
/// keychain <-> descriptor is a one-to-one mapping that cannot be changed. Attempting to do so
|
|
/// will return a [`InsertDescriptorError<K>`].
|
|
///
|
|
/// `[KeychainTxOutIndex]` will prevent you from inserting two descriptors which derive the same
|
|
/// script pubkey at index 0, but it's up to you to ensure that descriptors don't collide at
|
|
/// other indices. If they do nothing catastrophic happens at the `KeychainTxOutIndex` level
|
|
/// (one keychain just becomes the defacto owner of that spk arbitrarily) but this may have
|
|
/// subtle implications up the application stack like one UTXO being missing from one keychain
|
|
/// because it has been assigned to another which produces the same script pubkey.
|
|
pub fn insert_descriptor(
|
|
&mut self,
|
|
keychain: K,
|
|
descriptor: Descriptor<DescriptorPublicKey>,
|
|
) -> Result<ChangeSet<K>, InsertDescriptorError<K>> {
|
|
let mut changeset = ChangeSet::<K>::default();
|
|
let did = descriptor.descriptor_id();
|
|
if !self.keychain_to_descriptor_id.contains_key(&keychain)
|
|
&& !self.descriptor_id_to_keychain.contains_key(&did)
|
|
{
|
|
self.descriptors.insert(did, descriptor.clone());
|
|
self.keychain_to_descriptor_id.insert(keychain.clone(), did);
|
|
self.descriptor_id_to_keychain.insert(did, keychain.clone());
|
|
self.replenish_inner_index(did, &keychain, self.lookahead);
|
|
changeset
|
|
.keychains_added
|
|
.insert(keychain.clone(), descriptor);
|
|
} else {
|
|
if let Some(existing_desc_id) = self.keychain_to_descriptor_id.get(&keychain) {
|
|
let descriptor = self.descriptors.get(existing_desc_id).expect("invariant");
|
|
if *existing_desc_id != did {
|
|
return Err(InsertDescriptorError::KeychainAlreadyAssigned {
|
|
existing_assignment: descriptor.clone(),
|
|
keychain,
|
|
});
|
|
}
|
|
}
|
|
|
|
if let Some(existing_keychain) = self.descriptor_id_to_keychain.get(&did) {
|
|
let descriptor = self.descriptors.get(&did).expect("invariant").clone();
|
|
|
|
if *existing_keychain != keychain {
|
|
return Err(InsertDescriptorError::DescriptorAlreadyAssigned {
|
|
existing_assignment: existing_keychain.clone(),
|
|
descriptor,
|
|
});
|
|
}
|
|
}
|
|
}
|
|
|
|
Ok(changeset)
|
|
}
|
|
|
|
/// Gets the descriptor associated with the keychain. Returns `None` if the keychain doesn't
|
|
/// have a descriptor associated with it.
|
|
pub fn get_descriptor(&self, keychain: &K) -> Option<&Descriptor<DescriptorPublicKey>> {
|
|
let did = self.keychain_to_descriptor_id.get(keychain)?;
|
|
self.descriptors.get(did)
|
|
}
|
|
|
|
/// Get the lookahead setting.
|
|
///
|
|
/// Refer to [`new`] for more information on the `lookahead`.
|
|
///
|
|
/// [`new`]: Self::new
|
|
pub fn lookahead(&self) -> u32 {
|
|
self.lookahead
|
|
}
|
|
|
|
/// Store lookahead scripts until `target_index` (inclusive).
|
|
///
|
|
/// This does not change the global `lookahead` setting.
|
|
pub fn lookahead_to_target(&mut self, keychain: &K, target_index: u32) {
|
|
if let Some((next_index, _)) = self.next_index(keychain) {
|
|
let temp_lookahead = (target_index + 1)
|
|
.checked_sub(next_index)
|
|
.filter(|&index| index > 0);
|
|
|
|
if let Some(temp_lookahead) = temp_lookahead {
|
|
self.replenish_inner_index_keychain(keychain, temp_lookahead);
|
|
}
|
|
}
|
|
}
|
|
|
|
fn replenish_inner_index_did(&mut self, did: DescriptorId, lookahead: u32) {
|
|
if let Some(keychain) = self.descriptor_id_to_keychain.get(&did).cloned() {
|
|
self.replenish_inner_index(did, &keychain, lookahead);
|
|
}
|
|
}
|
|
|
|
fn replenish_inner_index_keychain(&mut self, keychain: &K, lookahead: u32) {
|
|
if let Some(did) = self.keychain_to_descriptor_id.get(keychain) {
|
|
self.replenish_inner_index(*did, keychain, lookahead);
|
|
}
|
|
}
|
|
|
|
/// Syncs the state of the inner spk index after changes to a keychain
|
|
fn replenish_inner_index(&mut self, did: DescriptorId, keychain: &K, lookahead: u32) {
|
|
let descriptor = self.descriptors.get(&did).expect("invariant");
|
|
let next_store_index = self
|
|
.inner
|
|
.all_spks()
|
|
.range(&(keychain.clone(), u32::MIN)..=&(keychain.clone(), u32::MAX))
|
|
.last()
|
|
.map_or(0, |((_, index), _)| *index + 1);
|
|
let next_reveal_index = self.last_revealed.get(&did).map_or(0, |v| *v + 1);
|
|
for (new_index, new_spk) in
|
|
SpkIterator::new_with_range(descriptor, next_store_index..next_reveal_index + lookahead)
|
|
{
|
|
let _inserted = self
|
|
.inner
|
|
.insert_spk((keychain.clone(), new_index), new_spk);
|
|
debug_assert!(_inserted, "replenish lookahead: must not have existing spk: keychain={:?}, lookahead={}, next_store_index={}, next_reveal_index={}", keychain, lookahead, next_store_index, next_reveal_index);
|
|
}
|
|
}
|
|
|
|
/// Get an unbounded spk iterator over a given `keychain`. Returns `None` if the provided
|
|
/// keychain doesn't exist
|
|
pub fn unbounded_spk_iter(
|
|
&self,
|
|
keychain: &K,
|
|
) -> Option<SpkIterator<Descriptor<DescriptorPublicKey>>> {
|
|
let descriptor = self.get_descriptor(keychain)?.clone();
|
|
Some(SpkIterator::new(descriptor))
|
|
}
|
|
|
|
/// Get unbounded spk iterators for all keychains.
|
|
pub fn all_unbounded_spk_iters(
|
|
&self,
|
|
) -> BTreeMap<K, SpkIterator<Descriptor<DescriptorPublicKey>>> {
|
|
self.keychain_to_descriptor_id
|
|
.iter()
|
|
.map(|(k, did)| {
|
|
(
|
|
k.clone(),
|
|
SpkIterator::new(self.descriptors.get(did).expect("invariant").clone()),
|
|
)
|
|
})
|
|
.collect()
|
|
}
|
|
|
|
/// Iterate over revealed spks of keychains in `range`
|
|
pub fn revealed_spks(
|
|
&self,
|
|
range: impl RangeBounds<K>,
|
|
) -> impl Iterator<Item = KeychainIndexed<K, &Script>> {
|
|
let start = range.start_bound();
|
|
let end = range.end_bound();
|
|
let mut iter_last_revealed = self
|
|
.keychain_to_descriptor_id
|
|
.range((start, end))
|
|
.map(|(k, did)| (k, self.last_revealed.get(did).cloned()));
|
|
let mut iter_spks = self
|
|
.inner
|
|
.all_spks()
|
|
.range(self.map_to_inner_bounds((start, end)));
|
|
let mut current_keychain = iter_last_revealed.next();
|
|
// The reason we need a tricky algorithm is because of the "lookahead" feature which means
|
|
// that some of the spks in the SpkTxoutIndex will not have been revealed yet. So we need to
|
|
// filter out those spks that are above the last_revealed for that keychain. To do this we
|
|
// iterate through the last_revealed for each keychain and the spks for each keychain in
|
|
// tandem. This minimizes BTreeMap queries.
|
|
core::iter::from_fn(move || loop {
|
|
let ((keychain, index), spk) = iter_spks.next()?;
|
|
// We need to find the last revealed that matches the current spk we are considering so
|
|
// we skip ahead.
|
|
while current_keychain?.0 < keychain {
|
|
current_keychain = iter_last_revealed.next();
|
|
}
|
|
let (current_keychain, last_revealed) = current_keychain?;
|
|
|
|
if current_keychain == keychain && Some(*index) <= last_revealed {
|
|
break Some(((keychain.clone(), *index), spk.as_script()));
|
|
}
|
|
})
|
|
}
|
|
|
|
/// Iterate over revealed spks of the given `keychain` with ascending indices.
|
|
///
|
|
/// This is a double ended iterator so you can easily reverse it to get an iterator where
|
|
/// the script pubkeys that were most recently revealed are first.
|
|
pub fn revealed_keychain_spks<'a>(
|
|
&'a self,
|
|
keychain: &'a K,
|
|
) -> impl DoubleEndedIterator<Item = Indexed<&Script>> + 'a {
|
|
let end = self
|
|
.last_revealed_index(keychain)
|
|
.map(|v| v + 1)
|
|
.unwrap_or(0);
|
|
self.inner
|
|
.all_spks()
|
|
.range((keychain.clone(), 0)..(keychain.clone(), end))
|
|
.map(|((_, index), spk)| (*index, spk.as_script()))
|
|
}
|
|
|
|
/// Iterate over revealed, but unused, spks of all keychains.
|
|
pub fn unused_spks(
|
|
&self,
|
|
) -> impl DoubleEndedIterator<Item = KeychainIndexed<K, &Script>> + Clone {
|
|
self.keychain_to_descriptor_id.keys().flat_map(|keychain| {
|
|
self.unused_keychain_spks(keychain)
|
|
.map(|(i, spk)| ((keychain.clone(), i), spk))
|
|
})
|
|
}
|
|
|
|
/// Iterate over revealed, but unused, spks of the given `keychain`.
|
|
/// Returns an empty iterator if the provided keychain doesn't exist.
|
|
pub fn unused_keychain_spks(
|
|
&self,
|
|
keychain: &K,
|
|
) -> impl DoubleEndedIterator<Item = Indexed<&Script>> + Clone {
|
|
let end = match self.keychain_to_descriptor_id.get(keychain) {
|
|
Some(did) => self.last_revealed.get(did).map(|v| *v + 1).unwrap_or(0),
|
|
None => 0,
|
|
};
|
|
|
|
self.inner
|
|
.unused_spks((keychain.clone(), 0)..(keychain.clone(), end))
|
|
.map(|((_, i), spk)| (*i, spk))
|
|
}
|
|
|
|
/// Get the next derivation index for `keychain`. The next index is the index after the last revealed
|
|
/// derivation index.
|
|
///
|
|
/// The second field in the returned tuple represents whether the next derivation index is new.
|
|
/// There are two scenarios where the next derivation index is reused (not new):
|
|
///
|
|
/// 1. The keychain's descriptor has no wildcard, and a script has already been revealed.
|
|
/// 2. The number of revealed scripts has already reached 2^31 (refer to BIP-32).
|
|
///
|
|
/// Not checking the second field of the tuple may result in address reuse.
|
|
///
|
|
/// Returns None if the provided `keychain` doesn't exist.
|
|
pub fn next_index(&self, keychain: &K) -> Option<(u32, bool)> {
|
|
let did = self.keychain_to_descriptor_id.get(keychain)?;
|
|
let last_index = self.last_revealed.get(did).cloned();
|
|
let descriptor = self.descriptors.get(did).expect("invariant");
|
|
|
|
// we can only get the next index if the wildcard exists.
|
|
let has_wildcard = descriptor.has_wildcard();
|
|
|
|
Some(match last_index {
|
|
// if there is no index, next_index is always 0.
|
|
None => (0, true),
|
|
// descriptors without wildcards can only have one index.
|
|
Some(_) if !has_wildcard => (0, false),
|
|
// derivation index must be < 2^31 (BIP-32).
|
|
Some(index) if index > BIP32_MAX_INDEX => {
|
|
unreachable!("index is out of bounds")
|
|
}
|
|
Some(index) if index == BIP32_MAX_INDEX => (index, false),
|
|
// get the next derivation index.
|
|
Some(index) => (index + 1, true),
|
|
})
|
|
}
|
|
|
|
/// Get the last derivation index that is revealed for each keychain.
|
|
///
|
|
/// Keychains with no revealed indices will not be included in the returned [`BTreeMap`].
|
|
pub fn last_revealed_indices(&self) -> BTreeMap<K, u32> {
|
|
self.last_revealed
|
|
.iter()
|
|
.filter_map(|(desc_id, index)| {
|
|
let keychain = self.descriptor_id_to_keychain.get(desc_id)?;
|
|
Some((keychain.clone(), *index))
|
|
})
|
|
.collect()
|
|
}
|
|
|
|
/// Get the last derivation index revealed for `keychain`. Returns None if the keychain doesn't
|
|
/// exist, or if the keychain doesn't have any revealed scripts.
|
|
pub fn last_revealed_index(&self, keychain: &K) -> Option<u32> {
|
|
let descriptor_id = self.keychain_to_descriptor_id.get(keychain)?;
|
|
self.last_revealed.get(descriptor_id).cloned()
|
|
}
|
|
|
|
/// Convenience method to call [`Self::reveal_to_target`] on multiple keychains.
|
|
pub fn reveal_to_target_multi(&mut self, keychains: &BTreeMap<K, u32>) -> ChangeSet<K> {
|
|
let mut changeset = ChangeSet::default();
|
|
|
|
for (keychain, &index) in keychains {
|
|
if let Some((_, new_changeset)) = self.reveal_to_target(keychain, index) {
|
|
changeset.append(new_changeset);
|
|
}
|
|
}
|
|
|
|
changeset
|
|
}
|
|
|
|
/// Reveals script pubkeys of the `keychain`'s descriptor **up to and including** the
|
|
/// `target_index`.
|
|
///
|
|
/// If the `target_index` cannot be reached (due to the descriptor having no wildcard and/or
|
|
/// the `target_index` is in the hardened index range), this method will make a best-effort and
|
|
/// reveal up to the last possible index.
|
|
///
|
|
/// This returns list of newly revealed indices (alongside their scripts) and a
|
|
/// [`ChangeSet`], which reports updates to the latest revealed index. If no new script
|
|
/// pubkeys are revealed, then both of these will be empty.
|
|
///
|
|
/// Returns None if the provided `keychain` doesn't exist.
|
|
#[must_use]
|
|
pub fn reveal_to_target(
|
|
&mut self,
|
|
keychain: &K,
|
|
target_index: u32,
|
|
) -> Option<(Vec<Indexed<ScriptBuf>>, ChangeSet<K>)> {
|
|
let mut changeset = ChangeSet::default();
|
|
let mut spks: Vec<Indexed<ScriptBuf>> = vec![];
|
|
while let Some((i, new)) = self.next_index(keychain) {
|
|
if !new || i > target_index {
|
|
break;
|
|
}
|
|
match self.reveal_next_spk(keychain) {
|
|
Some(((i, spk), change)) => {
|
|
spks.push((i, spk));
|
|
changeset.append(change);
|
|
}
|
|
None => break,
|
|
}
|
|
}
|
|
|
|
Some((spks, changeset))
|
|
}
|
|
|
|
/// Attempts to reveal the next script pubkey for `keychain`.
|
|
///
|
|
/// Returns the derivation index of the revealed script pubkey, the revealed script pubkey and a
|
|
/// [`ChangeSet`] which represents changes in the last revealed index (if any).
|
|
/// Returns None if the provided keychain doesn't exist.
|
|
///
|
|
/// When a new script cannot be revealed, we return the last revealed script and an empty
|
|
/// [`ChangeSet`]. There are two scenarios when a new script pubkey cannot be derived:
|
|
///
|
|
/// 1. The descriptor has no wildcard and already has one script revealed.
|
|
/// 2. The descriptor has already revealed scripts up to the numeric bound.
|
|
/// 3. There is no descriptor associated with the given keychain.
|
|
pub fn reveal_next_spk(&mut self, keychain: &K) -> Option<(Indexed<ScriptBuf>, ChangeSet<K>)> {
|
|
let (next_index, new) = self.next_index(keychain)?;
|
|
let mut changeset = ChangeSet::default();
|
|
|
|
if new {
|
|
let did = self.keychain_to_descriptor_id.get(keychain)?;
|
|
self.last_revealed.insert(*did, next_index);
|
|
changeset.last_revealed.insert(*did, next_index);
|
|
self.replenish_inner_index(*did, keychain, self.lookahead);
|
|
}
|
|
let script = self
|
|
.inner
|
|
.spk_at_index(&(keychain.clone(), next_index))
|
|
.expect("we just inserted it");
|
|
Some(((next_index, script.into()), changeset))
|
|
}
|
|
|
|
/// Gets the next unused script pubkey in the keychain. I.e., the script pubkey with the lowest
|
|
/// index that has not been used yet.
|
|
///
|
|
/// This will derive and reveal a new script pubkey if no more unused script pubkeys exist.
|
|
///
|
|
/// If the descriptor has no wildcard and already has a used script pubkey or if a descriptor
|
|
/// has used all scripts up to the derivation bounds, then the last derived script pubkey will be
|
|
/// returned.
|
|
///
|
|
/// Returns `None` if there are no script pubkeys that have been used and no new script pubkey
|
|
/// could be revealed (see [`reveal_next_spk`] for when this happens).
|
|
///
|
|
/// [`reveal_next_spk`]: Self::reveal_next_spk
|
|
pub fn next_unused_spk(&mut self, keychain: &K) -> Option<(Indexed<ScriptBuf>, ChangeSet<K>)> {
|
|
let next_unused = self
|
|
.unused_keychain_spks(keychain)
|
|
.next()
|
|
.map(|(i, spk)| ((i, spk.to_owned()), ChangeSet::default()));
|
|
|
|
next_unused.or_else(|| self.reveal_next_spk(keychain))
|
|
}
|
|
|
|
/// Iterate over all [`OutPoint`]s that have `TxOut`s with script pubkeys derived from
|
|
/// `keychain`.
|
|
pub fn keychain_outpoints<'a>(
|
|
&'a self,
|
|
keychain: &'a K,
|
|
) -> impl DoubleEndedIterator<Item = Indexed<OutPoint>> + 'a {
|
|
self.keychain_outpoints_in_range(keychain..=keychain)
|
|
.map(|((_, i), op)| (i, op))
|
|
}
|
|
|
|
/// Iterate over [`OutPoint`]s that have script pubkeys derived from keychains in `range`.
|
|
pub fn keychain_outpoints_in_range<'a>(
|
|
&'a self,
|
|
range: impl RangeBounds<K> + 'a,
|
|
) -> impl DoubleEndedIterator<Item = KeychainIndexed<K, OutPoint>> + 'a {
|
|
self.inner
|
|
.outputs_in_range(self.map_to_inner_bounds(range))
|
|
.map(|((k, i), op)| ((k.clone(), *i), op))
|
|
}
|
|
|
|
fn map_to_inner_bounds(&self, bound: impl RangeBounds<K>) -> impl RangeBounds<(K, u32)> {
|
|
let start = match bound.start_bound() {
|
|
Bound::Included(keychain) => Bound::Included((keychain.clone(), u32::MIN)),
|
|
Bound::Excluded(keychain) => Bound::Excluded((keychain.clone(), u32::MAX)),
|
|
Bound::Unbounded => Bound::Unbounded,
|
|
};
|
|
let end = match bound.end_bound() {
|
|
Bound::Included(keychain) => Bound::Included((keychain.clone(), u32::MAX)),
|
|
Bound::Excluded(keychain) => Bound::Excluded((keychain.clone(), u32::MIN)),
|
|
Bound::Unbounded => Bound::Unbounded,
|
|
};
|
|
|
|
(start, end)
|
|
}
|
|
|
|
/// Returns the highest derivation index of the `keychain` where [`KeychainTxOutIndex`] has
|
|
/// found a [`TxOut`] with it's script pubkey.
|
|
pub fn last_used_index(&self, keychain: &K) -> Option<u32> {
|
|
self.keychain_outpoints(keychain).last().map(|(i, _)| i)
|
|
}
|
|
|
|
/// Returns the highest derivation index of each keychain that [`KeychainTxOutIndex`] has found
|
|
/// a [`TxOut`] with it's script pubkey.
|
|
pub fn last_used_indices(&self) -> BTreeMap<K, u32> {
|
|
self.keychain_to_descriptor_id
|
|
.iter()
|
|
.filter_map(|(keychain, _)| {
|
|
self.last_used_index(keychain)
|
|
.map(|index| (keychain.clone(), index))
|
|
})
|
|
.collect()
|
|
}
|
|
|
|
/// Applies the `ChangeSet<K>` to the [`KeychainTxOutIndex<K>`]
|
|
///
|
|
/// Keychains added by the `keychains_added` field of `ChangeSet<K>` respect the one-to-one
|
|
/// keychain <-> descriptor invariant by silently ignoring attempts to violate it (but will
|
|
/// panic if `debug_assertions` are enabled).
|
|
pub fn apply_changeset(&mut self, changeset: ChangeSet<K>) {
|
|
let ChangeSet {
|
|
keychains_added,
|
|
last_revealed,
|
|
} = changeset;
|
|
for (keychain, descriptor) in keychains_added {
|
|
let _ignore_invariant_violation = self.insert_descriptor(keychain, descriptor);
|
|
}
|
|
|
|
for (&desc_id, &index) in &last_revealed {
|
|
let v = self.last_revealed.entry(desc_id).or_default();
|
|
*v = index.max(*v);
|
|
}
|
|
|
|
for did in last_revealed.keys() {
|
|
self.replenish_inner_index_did(*did, self.lookahead);
|
|
}
|
|
}
|
|
}
|
|
|
|
#[derive(Clone, Debug, PartialEq)]
|
|
/// Error returned from [`KeychainTxOutIndex::insert_descriptor`]
|
|
pub enum InsertDescriptorError<K> {
|
|
/// The descriptor has already been assigned to a keychain so you can't assign it to another
|
|
DescriptorAlreadyAssigned {
|
|
/// The descriptor you have attempted to reassign
|
|
descriptor: Descriptor<DescriptorPublicKey>,
|
|
/// The keychain that the descriptor is already assigned to
|
|
existing_assignment: K,
|
|
},
|
|
/// The keychain is already assigned to a descriptor so you can't reassign it
|
|
KeychainAlreadyAssigned {
|
|
/// The keychain that you have attempted to reassign
|
|
keychain: K,
|
|
/// The descriptor that the keychain is already assigned to
|
|
existing_assignment: Descriptor<DescriptorPublicKey>,
|
|
},
|
|
}
|
|
|
|
impl<K: core::fmt::Debug> core::fmt::Display for InsertDescriptorError<K> {
|
|
fn fmt(&self, f: &mut core::fmt::Formatter<'_>) -> core::fmt::Result {
|
|
match self {
|
|
InsertDescriptorError::DescriptorAlreadyAssigned {
|
|
existing_assignment: existing,
|
|
descriptor,
|
|
} => {
|
|
write!(
|
|
f,
|
|
"attempt to re-assign descriptor {descriptor:?} already assigned to {existing:?}"
|
|
)
|
|
}
|
|
InsertDescriptorError::KeychainAlreadyAssigned {
|
|
existing_assignment: existing,
|
|
keychain,
|
|
} => {
|
|
write!(
|
|
f,
|
|
"attempt to re-assign keychain {keychain:?} already assigned to {existing:?}"
|
|
)
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
#[cfg(feature = "std")]
|
|
impl<K: core::fmt::Debug> std::error::Error for InsertDescriptorError<K> {}
|
|
|
|
/// Represents updates to the derivation index of a [`KeychainTxOutIndex`].
|
|
/// It maps each keychain `K` to a descriptor and its last revealed index.
|
|
///
|
|
/// It can be applied to [`KeychainTxOutIndex`] with [`apply_changeset`].
|
|
///
|
|
/// The `last_revealed` field is monotone in that [`append`] will never decrease it.
|
|
/// `keychains_added` is *not* monotone, once it is set any attempt to change it is subject to the
|
|
/// same *one-to-one* keychain <-> descriptor mapping invariant as [`KeychainTxOutIndex`] itself.
|
|
///
|
|
/// [`apply_changeset`]: KeychainTxOutIndex::apply_changeset
|
|
/// [`append`]: Self::append
|
|
#[derive(Clone, Debug, PartialEq)]
|
|
#[cfg_attr(
|
|
feature = "serde",
|
|
derive(serde::Deserialize, serde::Serialize),
|
|
serde(
|
|
crate = "serde_crate",
|
|
bound(
|
|
deserialize = "K: Ord + serde::Deserialize<'de>",
|
|
serialize = "K: Ord + serde::Serialize"
|
|
)
|
|
)
|
|
)]
|
|
#[must_use]
|
|
pub struct ChangeSet<K> {
|
|
/// Contains the keychains that have been added and their respective descriptor
|
|
pub keychains_added: BTreeMap<K, Descriptor<DescriptorPublicKey>>,
|
|
/// Contains for each descriptor_id the last revealed index of derivation
|
|
pub last_revealed: BTreeMap<DescriptorId, u32>,
|
|
}
|
|
|
|
impl<K: Ord> Append for ChangeSet<K> {
|
|
/// Merge another [`ChangeSet<K>`] into self.
|
|
///
|
|
/// For the `keychains_added` field this method respects the invariants of
|
|
/// [`insert_descriptor`]. `last_revealed` always becomes the larger of the two.
|
|
///
|
|
/// [`insert_descriptor`]: KeychainTxOutIndex::insert_descriptor
|
|
fn append(&mut self, other: Self) {
|
|
for (new_keychain, new_descriptor) in other.keychains_added {
|
|
// enforce 1-to-1 invariance
|
|
if !self.keychains_added.contains_key(&new_keychain)
|
|
// FIXME: very inefficient
|
|
&& self
|
|
.keychains_added
|
|
.values()
|
|
.all(|descriptor| descriptor != &new_descriptor)
|
|
{
|
|
self.keychains_added.insert(new_keychain, new_descriptor);
|
|
}
|
|
}
|
|
|
|
// for `last_revealed`, entries of `other` will take precedence ONLY if it is greater than
|
|
// what was originally in `self`.
|
|
for (desc_id, index) in other.last_revealed {
|
|
use crate::collections::btree_map::Entry;
|
|
match self.last_revealed.entry(desc_id) {
|
|
Entry::Vacant(entry) => {
|
|
entry.insert(index);
|
|
}
|
|
Entry::Occupied(mut entry) => {
|
|
if *entry.get() < index {
|
|
entry.insert(index);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
/// Returns whether the changeset are empty.
|
|
fn is_empty(&self) -> bool {
|
|
self.last_revealed.is_empty() && self.keychains_added.is_empty()
|
|
}
|
|
}
|
|
|
|
impl<K> Default for ChangeSet<K> {
|
|
fn default() -> Self {
|
|
Self {
|
|
last_revealed: BTreeMap::default(),
|
|
keychains_added: BTreeMap::default(),
|
|
}
|
|
}
|
|
}
|
|
|
|
#[derive(Clone, Debug, Eq, PartialEq)]
|
|
/// The keychain doesn't exist. Most likley hasn't been inserted with [`KeychainTxOutIndex::insert_descriptor`].
|
|
pub struct NoSuchKeychain<K>(K);
|
|
|
|
impl<K: Debug> core::fmt::Display for NoSuchKeychain<K> {
|
|
fn fmt(&self, f: &mut core::fmt::Formatter<'_>) -> core::fmt::Result {
|
|
write!(f, "no such keychain {:?} exists", &self.0)
|
|
}
|
|
}
|