zecfoundation
/
rust-bitcoin
mirror of https://github.com/ZcashFoundation/rust-bitcoin.git
1
0
Fork 0
Code Issues Projects Releases Wiki Activity

Moved blockchain and patricia_tree to rust-memblocks

This commit is contained in:
Tamas Blummer 2018-03-09 17:22:31 +01:00
parent 6d13d68f51
commit 77ce6f18d0
4 changed files with 0 additions and 1338 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

621
src/blockdata/blockchain.rs
View File

@ -1,621 +0,0 @@
// Rust Bitcoin Library
// Written in 2014 by
// Andrew Poelstra <apoelstra@wpsoftware.net>
//
// To the extent possible under law, the author(s) have dedicated all
// copyright and related and neighboring rights to this software to
// the public domain worldwide. This software is distributed without
// any warranty.
//
// You should have received a copy of the CC0 Public Domain Dedication
// along with this software.
// If not, see <http://creativecommons.org/publicdomain/zero/1.0/>.
//
//! # Bitcoin Blockchain
//!
//! This module provides the structures and functions to maintain the
//! blockchain.
//!
use std::{marker, ptr};
use blockdata::block::{Block, BlockHeader};
use blockdata::transaction::Transaction;
use blockdata::constants::{DIFFCHANGE_INTERVAL, DIFFCHANGE_TIMESPAN,
TARGET_BLOCK_SPACING, max_target, genesis_block};
use network::constants::Network;
use network::encodable::{ConsensusDecodable, ConsensusEncodable};
use network::serialize::{BitcoinHash, SimpleDecoder, SimpleEncoder};
use util::BitArray;
use util;
use util::Error::{BlockNotFound, DuplicateHash, PrevHashNotFound};
use util::uint::Uint256;
use util::hash::Sha256dHash;
use util::patricia_tree::PatriciaTree;
type BlockTree = PatriciaTree<Uint256, Box<BlockchainNode>>;
type NodePtr = *const BlockchainNode;
/// A link in the blockchain
pub struct BlockchainNode {
/// The actual block
pub block: Block,
/// Total work from genesis to this point
pub total_work: Uint256,
/// Expected value of `block.header.bits` for this block; only changes every
/// `blockdata::constants::DIFFCHANGE_INTERVAL;` blocks
pub required_difficulty: Uint256,
/// Height above genesis
pub height: u32,
/// Whether the transaction data is stored
pub has_txdata: bool,
/// Pointer to block's parent
prev: NodePtr,
/// Pointer to block's child
next: NodePtr
}
impl BlockchainNode {
/// Is the node on the main chain?
pub fn is_on_main_chain(&self, chain: &Blockchain) -> bool {
if self.block.header == unsafe { (*chain.best_tip).block.header } {
true
} else {
unsafe {
let mut scan = self.next;
while !scan.is_null() {
if (*scan).block.header == (*chain.best_tip).block.header {
return true;
}
scan = (*scan).next;
}
}
false
}
}
}
impl<S: SimpleEncoder> ConsensusEncodable<S> for BlockchainNode {
#[inline]
fn consensus_encode(&self, s: &mut S) -> Result<(), S::Error> {
try!(self.block.consensus_encode(s));
try!(self.total_work.consensus_encode(s));
try!(self.required_difficulty.consensus_encode(s));
try!(self.height.consensus_encode(s));
try!(self.has_txdata.consensus_encode(s));
// Don't serialize the prev or next pointers
Ok(())
}
}
impl<D: SimpleDecoder> ConsensusDecodable<D> for BlockchainNode {
#[inline]
fn consensus_decode(d: &mut D) -> Result<BlockchainNode, D::Error> {
Ok(BlockchainNode {
block: try!(ConsensusDecodable::consensus_decode(d)),
total_work: try!(ConsensusDecodable::consensus_decode(d)),
required_difficulty: try!(ConsensusDecodable::consensus_decode(d)),
height: try!(ConsensusDecodable::consensus_decode(d)),
has_txdata: try!(ConsensusDecodable::consensus_decode(d)),
prev: ptr::null(),
next: ptr::null()
})
}
}
impl BitcoinHash for BlockchainNode {
fn bitcoin_hash(&self) -> Sha256dHash {
self.block.header.bitcoin_hash()
}
}
/// The blockchain
pub struct Blockchain {
network: Network,
tree: BlockTree,
best_tip: NodePtr,
best_hash: Sha256dHash,
genesis_hash: Sha256dHash
}
impl<S: SimpleEncoder> ConsensusEncodable<S> for Blockchain {
#[inline]
fn consensus_encode(&self, s: &mut S) -> Result<(), S::Error> {
try!(self.network.consensus_encode(s));
try!(self.tree.consensus_encode(s));
try!(self.best_hash.consensus_encode(s));
try!(self.genesis_hash.consensus_encode(s));
Ok(())
}
}
impl<D: SimpleDecoder> ConsensusDecodable<D> for Blockchain {
fn consensus_decode(d: &mut D) -> Result<Blockchain, D::Error> {
let network: Network = try!(ConsensusDecodable::consensus_decode(d));
let mut tree: BlockTree = try!(ConsensusDecodable::consensus_decode(d));
let best_hash: Sha256dHash = try!(ConsensusDecodable::consensus_decode(d));
let genesis_hash: Sha256dHash = try!(ConsensusDecodable::consensus_decode(d));
// Lookup best tip
let best = match tree.lookup(&best_hash.into_le(), 256) {
Some(node) => &**node as NodePtr,
None => {
return Err(d.error(format!("best tip {:x} not in tree", best_hash)));
}
};
// Lookup genesis
if tree.lookup(&genesis_hash.into_le(), 256).is_none() {
return Err(d.error(format!("genesis {:x} not in tree", genesis_hash)));
}
// Reconnect all prev pointers
let raw_tree = &tree as *const BlockTree;
for node in tree.mut_iter() {
let hash = node.block.header.prev_blockhash.into_le();
let prevptr =
match unsafe { (*raw_tree).lookup(&hash, 256) } {
Some(node) => &**node as NodePtr,
None => ptr::null()
};
node.prev = prevptr;
}
// Reconnect next pointers on the main chain
unsafe {
let mut scan = best;
while !(*scan).prev.is_null() {
let prev = (*scan).prev as *mut BlockchainNode;
(*prev).next = scan;
scan = prev as NodePtr;
}
// Check that "genesis" is the genesis
if (*scan).bitcoin_hash() != genesis_hash {
return Err(d.error(format!("no path from tip {:x} to genesis {:x}",
best_hash, genesis_hash)));
}
}
// Return the chain
Ok(Blockchain {
network: network,
tree: tree,
best_tip: best,
best_hash: best_hash,
genesis_hash: genesis_hash
})
}
}
// TODO: this should maybe be public, in which case it needs to be tagged
// with a PhantomData marker tying it to the tree's lifetime.
struct LocatorHashIter {
index: NodePtr,
count: usize,
skip: usize
}
impl LocatorHashIter {
fn new(init: NodePtr) -> LocatorHashIter {
LocatorHashIter { index: init, count: 0, skip: 1 }
}
}
impl Iterator for LocatorHashIter {
type Item = Sha256dHash;
fn next(&mut self) -> Option<Sha256dHash> {
if self.index.is_null() {
return None;
}
let ret = Some(unsafe { (*self.index).bitcoin_hash() });
// Rewind once (if we are at the genesis, this will set self.index to None)
self.index = unsafe { (*self.index).prev };
// If we are not at the genesis, rewind `self.skip` times, or until we are.
if !self.index.is_null() {
for _ in 1..self.skip {
unsafe {
if (*self.index).prev.is_null() {
break;
}
self.index = (*self.index).prev;
}
}
}
self.count += 1;
if self.count > 10 {
self.skip *= 2;
}
ret
}
}
/// An iterator over blocks in blockheight order
pub struct BlockIter<'tree> {
index: NodePtr,
// Note: we don't actually touch the blockchain. But we need
// to keep it borrowed to prevent it being mutated, since some
// mutable blockchain methods call .mut_borrow() on the block
// links, which would blow up if the iterator did a regular
// borrow at the same time.
marker: marker::PhantomData<&'tree Blockchain>
}
/// An iterator over blocks in reverse blockheight order. Note that this
/// is essentially the same as if we'd implemented `DoubleEndedIterator`
/// on `BlockIter` --- but we can't do that since if `BlockIter` is started
/// off the main chain, it will not reach the best tip, so the iterator
/// and its `.rev()` would be iterators over different chains! To avoid
/// this suprising behaviour we simply use separate iterators.
pub struct RevBlockIter<'tree> {
index: NodePtr,
// See comment in BlockIter for why we need this
marker: marker::PhantomData<&'tree Blockchain>
}
/// An iterator over blocks in reverse blockheight order, which yielding only
/// stale blocks (ending at the point where it would've returned a block on
/// the main chain). It does this by checking if the `next` pointer of the
/// next-to-by-yielded block matches the currently-yielded block. If not, scan
/// forward from next-to-be-yielded block. If we hit the best tip, set the
/// next-to-by-yielded block to None instead.
///
/// So to handle reorgs, you create a `RevStaleBlockIter` starting from the last
/// known block, and play it until it runs out, rewinding every block except for
/// the last one. Since the `UtxoSet` `rewind` function sets its `last_hash()` to
/// the prevblockhash of the rewinded block (which will be on the main chain at
/// the end of the iteration), you can then sync it up same as if you were doing
/// a plain old fast-forward.
pub struct RevStaleBlockIter<'tree> {
index: NodePtr,
chain: &'tree Blockchain
}
impl<'tree> Iterator for BlockIter<'tree> {
type Item = &'tree BlockchainNode;
fn next(&mut self) -> Option<&'tree BlockchainNode> {
if self.index.is_null() {
return None;
}
unsafe {
let ret = Some(&*self.index);
self.index = (*self.index).next;
ret
}
}
}
impl<'tree> Iterator for RevBlockIter<'tree> {
type Item = &'tree BlockchainNode;
fn next(&mut self) -> Option<&'tree BlockchainNode> {
if self.index.is_null() {
return None;
}
unsafe {
let ret = Some(&*self.index);
self.index = (*self.index).prev;
ret
}
}
}
impl<'tree> Iterator for RevStaleBlockIter<'tree> {
type Item = &'tree Block;
fn next(&mut self) -> Option<&'tree Block> {
if self.index.is_null() {
return None;
}
unsafe {
let ret = Some(&(*self.index).block);
let next_index = (*self.index).prev;
// Check if the next block is going to be on the main chain
if !next_index.is_null() &&
(*next_index).next != self.index &&
(&*next_index).is_on_main_chain(self.chain) {
self.index = ptr::null();
} else {
self.index = next_index;
}
ret
}
}
}
/// This function emulates the `GetCompact(SetCompact(n))` in the satoshi code,
/// which drops the precision to something that can be encoded precisely in
/// the nBits block header field. Savour the perversity. This is in Bitcoin
/// consensus code. What. Gaah!
fn satoshi_the_precision(n: Uint256) -> Uint256 {
// Shift by B bits right then left to turn the low bits to zero
let bits = 8 * ((n.bits() + 7) / 8 - 3);
let mut ret = n >> bits;
// Oh, did I say B was that fucked up formula? I meant sometimes also + 8.
if ret.bit(23) {
ret = (ret >> 8) << 8;
}
ret << bits
}
impl Blockchain {
/// Constructs a new blockchain
pub fn new(network: Network) -> Blockchain {
let genesis = genesis_block(network);
let genhash = genesis.header.bitcoin_hash();
let new_node = Box::new(BlockchainNode {
total_work: Default::default(),
required_difficulty: genesis.header.target(),
block: genesis,
height: 0,
has_txdata: true,
prev: ptr::null(),
next: ptr::null()
});
let raw_ptr = &*new_node as NodePtr;
Blockchain {
network: network,
tree: {
let mut pat = PatriciaTree::new();
pat.insert(&genhash.into_le(), 256, new_node);
pat
},
best_hash: genhash,
genesis_hash: genhash,
best_tip: raw_ptr
}
}
fn replace_txdata(&mut self, hash: &Uint256, txdata: Vec<Transaction>, has_txdata: bool) -> Result<(), util::Error> {
match self.tree.lookup_mut(hash, 256) {
Some(existing_block) => {
existing_block.block.txdata.clone_from(&txdata);
existing_block.has_txdata = has_txdata;
Ok(())
},
None => Err(BlockNotFound)
}
}
/// Looks up a block in the chain and returns the BlockchainNode containing it
pub fn get_block(&self, hash: Sha256dHash) -> Option<&BlockchainNode> {
self.tree.lookup(&hash.into_le(), 256).map(|node| &**node)
}
/// Locates a block in the chain and overwrites its txdata
pub fn add_txdata(&mut self, block: Block) -> Result<(), util::Error> {
self.replace_txdata(&block.header.bitcoin_hash().into_le(), block.txdata, true)
}
/// Locates a block in the chain and removes its txdata
pub fn remove_txdata(&mut self, hash: Sha256dHash) -> Result<(), util::Error> {
self.replace_txdata(&hash.into_le(), vec![], false)
}
/// Adds a block header to the chain
pub fn add_header(&mut self, header: BlockHeader) -> Result<(), util::Error> {
self.real_add_block(Block { header: header, txdata: vec![] }, false)
}
/// Adds a block to the chain
pub fn add_block(&mut self, block: Block) -> Result<(), util::Error> {
self.real_add_block(block, true)
}
fn real_add_block(&mut self, block: Block, has_txdata: bool) -> Result<(), util::Error> {
// get_prev optimizes the common case where we are extending the best tip
#[inline]
fn get_prev(chain: &Blockchain, hash: Sha256dHash) -> Option<NodePtr> {
if hash == chain.best_hash {
Some(chain.best_tip)
} else {
chain.tree.lookup(&hash.into_le(), 256).map(|boxptr| &**boxptr as NodePtr)
}
}
// Check for multiple inserts (bitcoind from c9a09183 to 3c85d2ec doesn't
// handle locator hashes properly and may return blocks multiple times,
// and this may also happen in case of a reorg.
if self.tree.lookup(&block.header.bitcoin_hash().into_le(), 256).is_some() {
return Err(DuplicateHash);
}
// Construct node, if possible
let new_block = match get_prev(self, block.header.prev_blockhash) {
Some(prev) => {
let difficulty =
// Compute required difficulty if this is a diffchange block
if (unsafe { (*prev).height } + 1) % DIFFCHANGE_INTERVAL == 0 {
let timespan = unsafe {
// Scan back DIFFCHANGE_INTERVAL blocks
let mut scan = prev;
for _ in 0..(DIFFCHANGE_INTERVAL - 1) {
scan = (*scan).prev;
}
// Get clamped timespan between first and last blocks
match (*prev).block.header.time - (*scan).block.header.time {
n if n < DIFFCHANGE_TIMESPAN / 4 => DIFFCHANGE_TIMESPAN / 4,
n if n > DIFFCHANGE_TIMESPAN * 4 => DIFFCHANGE_TIMESPAN * 4,
n => n
}
};
// Compute new target
let mut target = unsafe { (*prev).block.header.target() };
target = target.mul_u32(timespan);
target = target / Uint256::from_u64(DIFFCHANGE_TIMESPAN as u64).unwrap();
// Clamp below MAX_TARGET (difficulty 1)
let max = max_target(self.network);
if target > max { target = max };
// Compactify (make expressible in the 8+24 nBits float format
satoshi_the_precision(target)
// On non-diffchange blocks, Testnet has a rule that any 20-minute-long
// block intervals result the difficulty
} else if self.network == Network::Testnet &&
block.header.time > unsafe { (*prev).block.header.time } + 2*TARGET_BLOCK_SPACING {
max_target(self.network)
// On the other hand, if we are in Testnet and the block interval is less
// than 20 minutes, we need to scan backward to find a block for which the
// previous rule did not apply, to find the "real" difficulty.
} else if self.network == Network::Testnet {
// Scan back DIFFCHANGE_INTERVAL blocks
unsafe {
let mut scan = prev;
while (*scan).height % DIFFCHANGE_INTERVAL != 0 &&
(*scan).required_difficulty == max_target(self.network) {
scan = (*scan).prev;
}
(*scan).required_difficulty
}
// Otherwise just use the last block's difficulty
} else {
unsafe { (*prev).required_difficulty }
};
// Create node
let ret = Box::new(BlockchainNode {
total_work: block.header.work() + unsafe { (*prev).total_work },
block: block,
required_difficulty: difficulty,
height: unsafe { (*prev).height + 1 },
has_txdata: has_txdata,
prev: prev,
next: ptr::null()
});
unsafe {
let prev = prev as *mut BlockchainNode;
(*prev).next = &*ret as NodePtr;
}
ret
},
None => {
return Err(PrevHashNotFound);
}
};
// spv validate the block
try!(new_block.block.header.spv_validate(&new_block.required_difficulty));
// Insert the new block
let raw_ptr = &*new_block as NodePtr;
self.tree.insert(&new_block.block.header.bitcoin_hash().into_le(), 256, new_block);
// Replace the best tip if necessary
if unsafe { (*raw_ptr).total_work > (*self.best_tip).total_work } {
self.set_best_tip(raw_ptr);
}
Ok(())
}
/// Sets the best tip (not public)
fn set_best_tip(&mut self, tip: NodePtr) {
// Fix next links
unsafe {
let mut scan = self.best_tip;
// Scan backward
while !(*scan).prev.is_null() {
// If we hit the old best, there is no need to reorg.
if scan == self.best_tip { break; }
// Otherwise set the next-ptr and carry on
let prev = (*scan).prev as *mut BlockchainNode;
(*prev).next = scan;
scan = (*scan).prev;
}
}
// Set best
self.best_hash = unsafe { (*tip).bitcoin_hash() };
self.best_tip = tip;
}
/// Returns the genesis block's blockhash
pub fn genesis_hash(&self) -> Sha256dHash {
self.genesis_hash
}
/// Returns the best tip
pub fn best_tip(&self) -> &Block {
unsafe { &(*self.best_tip).block }
}
/// Returns the best tip height
pub fn best_tip_height(&self) -> u32 {
unsafe { (*self.best_tip).height }
}
/// Returns the best tip's blockhash
pub fn best_tip_hash(&self) -> Sha256dHash {
self.best_hash
}
/// Returns an array of locator hashes used in `getheaders` messages
pub fn locator_hashes(&self) -> Vec<Sha256dHash> {
LocatorHashIter::new(self.best_tip).collect()
}
/// An iterator over all blocks in the chain starting from `start_hash`
pub fn iter(&self, start_hash: Sha256dHash) -> BlockIter {
let start = match self.tree.lookup(&start_hash.into_le(), 256) {
Some(boxptr) => &**boxptr as NodePtr,
None => ptr::null()
};
BlockIter {
index: start,
marker: marker::PhantomData
}
}
/// An iterator over all blocks in reverse order to the genesis, starting with `start_hash`
pub fn rev_iter(&self, start_hash: Sha256dHash) -> RevBlockIter {
let start = match self.tree.lookup(&start_hash.into_le(), 256) {
Some(boxptr) => &**boxptr as NodePtr,
None => ptr::null()
};
RevBlockIter {
index: start,
marker: marker::PhantomData
}
}
/// An iterator over all blocks -not- in the best chain, in reverse order, starting from `start_hash`
pub fn rev_stale_iter(&self, start_hash: Sha256dHash) -> RevStaleBlockIter {
let start = match self.tree.lookup(&start_hash.into_le(), 256) {
Some(boxptr) => {
// If we are already on the main chain, we have a dead iterator
if boxptr.is_on_main_chain(self) {
ptr::null()
} else {
&**boxptr as NodePtr
}
}
None => ptr::null()
};
RevStaleBlockIter {
index: start,
chain: self
}
}
}
#[cfg(test)]
mod tests {
use blockdata::blockchain::Blockchain;
use blockdata::constants::genesis_block;
use network::constants::Network::Bitcoin;
use network::serialize::{BitcoinHash, deserialize, serialize};
#[test]
fn blockchain_serialize_test() {
let empty_chain = Blockchain::new(Bitcoin);
assert_eq!(empty_chain.best_tip().header.bitcoin_hash(),
genesis_block(Bitcoin).header.bitcoin_hash());
let serial = serialize(&empty_chain);
let deserial: Result<Blockchain, _> = deserialize(&serial.unwrap());
assert!(deserial.is_ok());
let read_chain = deserial.unwrap();
assert_eq!(read_chain.best_tip().header.bitcoin_hash(),
genesis_block(Bitcoin).header.bitcoin_hash());
}
}

1
src/blockdata/mod.rs
View File

@ -23,7 +23,6 @@ pub mod opcodes;
#[cfg(not(feature="broken_consensus_code"))] pub mod script;
pub mod transaction;
pub mod block;
pub mod blockchain;
#[cfg(feature="broken_consensus_code")]
/// # Script -- including consensus code

1
src/util/mod.rs
View File

@ -25,7 +25,6 @@ pub mod decimal;
pub mod hash;
pub mod iter;
pub mod misc;
pub mod patricia_tree;
pub mod uint;
use std::{error, fmt, io};

715
src/util/patricia_tree.rs
View File

@ -1,715 +0,0 @@
// Rust Bitcoin Library
// Written in 2014 by
// Andrew Poelstra <apoelstra@wpsoftware.net>
//
// To the extent possible under law, the author(s) have dedicated all
// copyright and related and neighboring rights to this software to
// the public domain worldwide. This software is distributed without
// any warranty.
//
// You should have received a copy of the CC0 Public Domain Dedication
// along with this software.
// If not, see <http://creativecommons.org/publicdomain/zero/1.0/>.
//
//! # Patricia/Radix Trie
//!
//! A Patricia trie is a trie in which nodes with only one child are
//! merged with the child, giving huge space savings for sparse tries.
//! A radix tree is more general, working with keys that are arbitrary
//! strings; a Patricia tree uses bitstrings.
//!
use std::fmt::Debug;
use std::marker;
use std::{cmp, fmt, ops, ptr};
use network::encodable::{ConsensusDecodable, ConsensusEncodable};
use network::serialize::{SimpleDecoder, SimpleEncoder};
use util::BitArray;
/// Patricia troo
pub struct PatriciaTree<K: Copy, V> {
data: Option<V>,
child_l: Option<Box<PatriciaTree<K, V>>>,
child_r: Option<Box<PatriciaTree<K, V>>>,
skip_prefix: K,
skip_len: u8
}
impl<K, V> PatriciaTree<K, V>
where K: Copy + BitArray + cmp::Eq +
ops::BitXor<K, Output=K> +
ops::Add<K, Output=K> +
ops::Shr<usize, Output=K> +
ops::Shl<usize, Output=K>
{
/// Constructs a new Patricia tree
pub fn new() -> PatriciaTree<K, V> {
PatriciaTree {
data: None,
child_l: None,
child_r: None,
skip_prefix: BitArray::zero(),
skip_len: 0
}
}
/// Lookup a value by exactly matching `key` and return a referenc
pub fn lookup_mut(&mut self, key: &K, key_len: usize) -> Option<&mut V> {
// Caution: `lookup_mut` never modifies its self parameter (in fact its
// internal recursion uses a non-mutable self, so we are OK to just
// transmute our self pointer into a mutable self before passing it in.
use std::mem::transmute;
unsafe { transmute(self.lookup(key, key_len)) }
}
/// Lookup a value by exactly matching `key` and return a mutable reference
pub fn lookup(&self, key: &K, key_len: usize) -> Option<&V> {
let mut node = self;
let mut key_idx = 0;
loop {
// If the search key is shorter than the node prefix, there is no
// way we can match, so fail.
if key_len - key_idx < node.skip_len as usize {
return None;
}
// Key fails to match prefix --- no match
if node.skip_prefix != key.bit_slice(key_idx, key_idx + node.skip_len as usize) {
return None;
}
// Key matches prefix: if they are an exact match, return the data
if node.skip_len as usize == key_len - key_idx {
return node.data.as_ref();
} else {
// Key matches prefix: search key longer than node key, recurse
key_idx += 1 + node.skip_len as usize;
let subtree = if key.bit(key_idx - 1) { &node.child_r } else { &node.child_l };
match *subtree {
Some(ref bx) => {
node = &**bx; // bx is a &Box<U> here, so &**bx gets &U
}
None => { return None; }
}
}
} // end loop
}
/// Inserts a value with key `key`, returning true on success. If a value is already
/// stored against `key`, do nothing and return false.
#[inline]
pub fn insert(&mut self, key: &K, key_len: usize, value: V) -> bool {
self.real_insert(key, key_len, value, false)
}
/// Inserts a value with key `key`, returning true on success. If a value is already
/// stored against `key`, overwrite it and return false.
#[inline]
pub fn insert_or_update(&mut self, key: &K, key_len: usize, value: V) -> bool {
self.real_insert(key, key_len, value, true)
}
fn real_insert(&mut self, key: &K, key_len: usize, value: V, overwrite: bool) -> bool {
let mut node = self;
let mut idx = 0;
loop {
// Mask in case search key is shorter than node key
let slice_len = cmp::min(node.skip_len as usize, key_len - idx);
let masked_prefix = node.skip_prefix.mask(slice_len);
let key_slice = key.bit_slice(idx, idx + slice_len);
// Prefixes do not match: split key
if masked_prefix != key_slice {
let diff = (masked_prefix ^ key_slice).trailing_zeros();
// Remove the old node's children
let child_l = node.child_l.take();
let child_r = node.child_r.take();
let value_neighbor = node.data.take();
let tmp = node; // borrowck hack
let (insert, neighbor) = if key_slice.bit(diff)
{ (&mut tmp.child_r, &mut tmp.child_l) }
else { (&mut tmp.child_l, &mut tmp.child_r) };
*insert = Some(Box::new(PatriciaTree {
data: None,
child_l: None,
child_r: None,
skip_prefix: key.bit_slice(idx + diff + 1, key_len),
skip_len: (key_len - idx - diff - 1) as u8
}));
*neighbor = Some(Box::new(PatriciaTree {
data: value_neighbor,
child_l: child_l,
child_r: child_r,
skip_prefix: tmp.skip_prefix >> (diff + 1),
skip_len: tmp.skip_len - diff as u8 - 1
}));
// Chop the prefix down
tmp.skip_len = diff as u8;
tmp.skip_prefix = tmp.skip_prefix.mask(diff);
// Recurse
idx += 1 + diff;
node = &mut **insert.as_mut().unwrap();
}
// Prefixes match
else {
let slice_len = key_len - idx;
// Search key is shorter than skip prefix: truncate the prefix and attach
// the old data as a child
if node.skip_len as usize > slice_len {
// Remove the old node's children
let child_l = node.child_l.take();
let child_r = node.child_r.take();
let value_neighbor = node.data.take();
// Put the old data in a new child, with the remainder of the prefix
let new_child = if node.skip_prefix.bit(slice_len)
{ &mut node.child_r } else { &mut node.child_l };
*new_child = Some(Box::new(PatriciaTree {
data: value_neighbor,
child_l: child_l,
child_r: child_r,
skip_prefix: node.skip_prefix >> (slice_len + 1),
skip_len: node.skip_len - slice_len as u8 - 1
}));
// Chop the prefix down and put the new data in place
node.skip_len = slice_len as u8;
node.skip_prefix = key_slice;
node.data = Some(value);
return true;
}
// If we have an exact match, great, insert it
else if node.skip_len as usize == slice_len {
if node.data.is_none() {
node.data = Some(value);
return true;
}
if overwrite {
node.data = Some(value);
}
return false;
}
// Search key longer than node key, recurse
else {
let tmp = node; // hack to appease borrowck
idx += tmp.skip_len as usize + 1;
let subtree = if key.bit(idx - 1)
{ &mut tmp.child_r } else { &mut tmp.child_l };
// Recurse, adding a new node if necessary
if subtree.is_none() {
*subtree = Some(Box::new(PatriciaTree {
data: None,
child_l: None,
child_r: None,
skip_prefix: key.bit_slice(idx, key_len),
skip_len: (key_len - idx) as u8
}));
}
// subtree.get_mut_ref is a &mut Box<U> here, so &mut ** gets a &mut U
node = &mut **subtree.as_mut().unwrap();
} // end search_len vs prefix len
} // end if prefixes match
} // end loop
}
/// Deletes a value with key `key`, returning it on success. If no value with
/// the given key is found, return None
pub fn delete(&mut self, key: &K, key_len: usize) -> Option<V> {
/// Return value is (deletable, actual return value), where `deletable` is true
/// is true when the entire node can be deleted (i.e. it has no children)
fn recurse<K, V>(tree: &mut PatriciaTree<K, V>, key: &K, key_len: usize) -> (bool, Option<V>)
where K: Copy + BitArray + cmp::Eq +
ops::Add<K, Output=K> +
ops::Shr<usize, Output=K> +
ops::Shl<usize, Output=K>
{
// If the search key is shorter than the node prefix, there is no
// way we can match, so fail.
if key_len < tree.skip_len as usize {
return (false, None);
}
// Key fails to match prefix --- no match
if tree.skip_prefix != key.mask(tree.skip_len as usize) {
return (false, None);
}
// If we are here, the key matches the prefix
if tree.skip_len as usize == key_len {
// Exact match -- delete and return
let ret = tree.data.take();
let bit = tree.child_r.is_some();
// First try to consolidate if there is only one child
if tree.child_l.is_some() && tree.child_r.is_some() {
// Two children means we cannot consolidate or delete
return (false, ret);
}
match (tree.child_l.take(), tree.child_r.take()) {
(Some(_), Some(_)) => unreachable!(),
(Some(child), None) | (None, Some(child)) => {
let child = *child; /* workaround for rustc #28536 */
let PatriciaTree { data, child_l, child_r, skip_len, skip_prefix } = child;
tree.data = data;
tree.child_l = child_l;
tree.child_r = child_r;
let new_bit = if bit { let ret: K = BitArray::one();
ret << (tree.skip_len as usize) }
else { BitArray::zero() };
tree.skip_prefix = tree.skip_prefix +
new_bit +
(skip_prefix << (1 + tree.skip_len as usize));
tree.skip_len += 1 + skip_len;
return (false, ret);
}
// No children means this node is deletable
(None, None) => { return (true, ret); }
}
}
// Otherwise, the key is longer than the prefix and we need to recurse
let next_bit = key.bit(tree.skip_len as usize);
// Recursively get the return value. This awkward scope is required
// to shorten the time we mutably borrow the node's children -- we
// might want to borrow the sibling later, so the borrow needs to end.
let ret = {
let target = if next_bit { &mut tree.child_r } else { &mut tree.child_l };
// If we can't recurse, fail
if target.is_none() {
return (false, None);
}
// Otherwise, do it
let (delete_child, ret) = recurse(&mut **target.as_mut().unwrap(),
&(*key >> (tree.skip_len as usize + 1)),
key_len - tree.skip_len as usize - 1);
if delete_child {
target.take();
}
ret
};
// The above block may have deleted the target. If we now have only one
// child, merge it into the parent. (If we have no children, mark this
// node for deletion.)
if tree.data.is_some() {
// First though, if this is a data node, we can neither delete nor
// consolidate it.
return (false, ret);
}
match (tree.child_r.is_some(), tree.child_l.take(), tree.child_r.take()) {
// Two children? Can't do anything, just sheepishly put them back
(_, Some(child_l), Some(child_r)) => {
tree.child_l = Some(child_l);
tree.child_r = Some(child_r);
(false, ret)
}
// One child? Consolidate
(bit, Some(child), None) | (bit, None, Some(child)) => {
let child = *child; /* workaround for rustc #28536 */
let PatriciaTree { data, child_l, child_r, skip_len, skip_prefix } = child;
tree.data = data;
tree.child_l = child_l;
tree.child_r = child_r;
let new_bit = if bit { let ret: K = BitArray::one();
ret << (tree.skip_len as usize) }
else { BitArray::zero() };
tree.skip_prefix = tree.skip_prefix +
new_bit +
(skip_prefix << (1 + tree.skip_len as usize));
tree.skip_len += 1 + skip_len;
(false, ret)
}
// No children? Delete
(_, None, None) => {
(true, ret)
}
}
}
let (_, ret) = recurse(self, key, key_len);
ret
}
/// Count all the nodes
pub fn node_count(&self) -> usize {
fn recurse<K: Copy, V>(node: &Option<Box<PatriciaTree<K, V>>>) -> usize {
match *node {
Some(ref node) => { 1 + recurse(&node.child_l) + recurse(&node.child_r) }
None => 0
}
}
1 + recurse(&self.child_l) + recurse(&self.child_r)
}
/// Returns an iterator over all elements in the tree
pub fn iter(&self) -> Items<K, V> {
Items {
node: Some(self),
parents: vec![],
started: false
}
}
/// Returns a mutable iterator over all elements in the tree
pub fn mut_iter(&mut self) -> MutItems<K, V> {
MutItems {
node: self as *mut _,
parents: vec![],
started: false,
marker: marker::PhantomData
}
}
}
impl<K: Copy + BitArray, V: Debug> Debug for PatriciaTree<K, V> {
/// Print the entire tree
fn fmt(&self, f: &mut fmt::Formatter) -> Result<(), fmt::Error> {
fn recurse<K, V>(tree: &PatriciaTree<K, V>, f: &mut fmt::Formatter, depth: usize) -> Result<(), fmt::Error>
where K: Copy + BitArray, V: Debug
{
for i in 0..tree.skip_len as usize {
try!(write!(f, "{:}", if tree.skip_prefix.bit(i) { 1 } else { 0 }));
}
try!(writeln!(f, ": {:?}", tree.data));
// left gets no indentation
if let Some(ref t) = tree.child_l {
for _ in 0..(depth + tree.skip_len as usize) {
try!(write!(f, "-"));
}
try!(write!(f, "0"));
try!(recurse(&**t, f, depth + tree.skip_len as usize + 1));
}
// right one gets indentation
if let Some(ref t) = tree.child_r {
for _ in 0..(depth + tree.skip_len as usize) {
try!(write!(f, "_"));
}
try!(write!(f, "1"));
try!(recurse(&**t, f, depth + tree.skip_len as usize + 1));
}
Ok(())
}
recurse(self, f, 0)
}
}
impl<S, K, V> ConsensusEncodable<S> for PatriciaTree<K, V>
where S: SimpleEncoder,
K: Copy + ConsensusEncodable<S>,
V: ConsensusEncodable<S>
{
fn consensus_encode(&self, s: &mut S) -> Result<(), S::Error> {
// Depth-first serialization: serialize self, then children
try!(self.skip_prefix.consensus_encode(s));
try!(self.skip_len.consensus_encode(s));
try!(self.data.consensus_encode(s));
try!(self.child_l.consensus_encode(s));
try!(self.child_r.consensus_encode(s));
Ok(())
}
}
impl<D, K, V> ConsensusDecodable<D> for PatriciaTree<K, V>
where D: SimpleDecoder,
K: Copy + ConsensusDecodable<D>,
V: ConsensusDecodable<D>
{
fn consensus_decode(d: &mut D) -> Result<PatriciaTree<K, V>, D::Error> {
Ok(PatriciaTree {
skip_prefix: try!(ConsensusDecodable::consensus_decode(d)),
skip_len: try!(ConsensusDecodable::consensus_decode(d)),
data: try!(ConsensusDecodable::consensus_decode(d)),
child_l: try!(ConsensusDecodable::consensus_decode(d)),
child_r: try!(ConsensusDecodable::consensus_decode(d))
})
}
}
/// Iterator
pub struct Items<'tree, K: Copy + 'tree, V: 'tree> {
started: bool,
node: Option<&'tree PatriciaTree<K, V>>,
parents: Vec<&'tree PatriciaTree<K, V>>
}
/// Mutable iterator
pub struct MutItems<'tree, K: Copy + 'tree, V: 'tree> {
started: bool,
node: *mut PatriciaTree<K, V>,
parents: Vec<*mut PatriciaTree<K, V>>,
marker: marker::PhantomData<&'tree PatriciaTree<K, V>>
}
impl<'a, K: Copy, V> Iterator for Items<'a, K, V> {
type Item = &'a V;
fn next(&mut self) -> Option<&'a V> {
fn borrow_opt<K: Copy, V>(opt_ptr: &Option<Box<PatriciaTree<K, V>>>) -> Option<&PatriciaTree<K, V>> {
opt_ptr.as_ref().map(|b| &**b)
}
// If we haven't started, maybe return the "last" return value,
// which will be the root node.
if !self.started {
if self.node.is_some() && (**self.node.as_ref().unwrap()).data.is_some() {
return self.node.unwrap().data.as_ref();
}
self.started = true;
}
// Find next data-containing node
while self.node.is_some() {
let mut node = self.node.take();
// Try to go left
let child_l = borrow_opt(&node.unwrap().child_l);
if child_l.is_some() {
self.parents.push(node.unwrap());
self.node = child_l;
// Try to go right, going back up the tree if necessary
} else {
while node.is_some() {
let child_r = borrow_opt(&node.unwrap().child_r);
if child_r.is_some() {
self.node = child_r;
break;
}
node = self.parents.pop();
}
}
// Stop if we've found data.
if self.node.is_some() && self.node.unwrap().data.is_some() {
break;
}
} // end loop
// Return data
self.node.and_then(|node| node.data.as_ref())
}
}
impl<'a, K: Copy, V> Iterator for MutItems<'a, K, V> {
type Item = &'a mut V;
fn next(&mut self) -> Option<&'a mut V> {
fn borrow_opt<K: Copy, V>(opt_ptr: &Option<Box<PatriciaTree<K, V>>>) -> *mut PatriciaTree<K, V> {
match *opt_ptr {
Some(ref data) => &**data as *const _ as *mut _,
None => ptr::null_mut()
}
}
// If we haven't started, maybe return the "last" return value,
// which will be the root node.
if !self.started {
unsafe {
if !self.node.is_null() && (*self.node).data.is_some() {
return (*self.node).data.as_mut();
}
}
self.started = true;
}
// Find next data-containing node
while !self.node.is_null() {
// Try to go left
let child_l = unsafe { borrow_opt(&(*self.node).child_l) };
if !child_l.is_null() {
self.parents.push(self.node);
self.node = child_l;
// Try to go right, going back up the tree if necessary
} else {
while !self.node.is_null() {
let child_r = unsafe { borrow_opt(&(*self.node).child_r) };
if !child_r.is_null() {
self.node = child_r;
break;
}
self.node = self.parents.pop().unwrap_or(ptr::null_mut());
}
}
// Stop if we've found data.
if !self.node.is_null() && unsafe { (*self.node).data.is_some() } {
break;
}
} // end loop
// Return data
if !self.node.is_null() {
unsafe { (*self.node).data.as_mut() }
} else {
None
}
}
}
#[cfg(test)]
mod tests {
use network::serialize::{deserialize, serialize};
use util::hash::Sha256dHash;
use util::uint::Uint128;
use util::uint::Uint256;
use util::patricia_tree::PatriciaTree;
use util::BitArray;
#[test]
fn patricia_single_insert_lookup_delete_test() {
let mut key = Uint256::from_u64(0xDEADBEEFDEADBEEF).unwrap();
key = key + (key << 64);
let mut tree = PatriciaTree::new();
tree.insert(&key, 100, 100u32);
tree.insert(&key, 120, 100u32);
assert_eq!(tree.lookup(&key, 100), Some(&100u32));
assert_eq!(tree.lookup(&key, 101), None);
assert_eq!(tree.lookup(&key, 99), None);
assert_eq!(tree.delete(&key, 100), Some(100u32));
}
#[test]
fn patricia_insert_lookup_delete_test() {
let mut tree = PatriciaTree::new();
let mut hashes = vec![];
for i in 0u32..5000 {
let hash = Sha256dHash::from_data(&[(i / 0x100) as u8, (i % 0x100) as u8]).into_le().low_128();
tree.insert(&hash, 250, i);
hashes.push(hash);
}
// Check that all inserts are correct
for (n, hash) in hashes.iter().enumerate() {
let ii = n as u32;
let ret = tree.lookup(hash, 250);
assert_eq!(ret, Some(&ii));
}
// Delete all the odd-numbered nodes
for (n, hash) in hashes.iter().enumerate() {
if n % 2 == 1 {
let ii = n as u32;
let ret = tree.delete(hash, 250);
assert_eq!(ret, Some(ii));
}
}
// Confirm all is correct
for (n, hash) in hashes.iter().enumerate() {
let ii = n as u32;
let ret = tree.lookup(hash, 250);
if n % 2 == 0 {
assert_eq!(ret, Some(&ii));
} else {
assert_eq!(ret, None);
}
}
}
#[test]
fn patricia_insert_substring_keys() {
// This test uses a bunch of keys that are substrings of each other
// to make sure insertion and deletion does not lose data
let mut tree = PatriciaTree::new();
let mut hashes = vec![];
// Start by inserting a bunch of chunder
for i in 1u32..500 {
let hash = Sha256dHash::from_data(&[(i / 0x100) as u8, (i % 0x100) as u8]).into_le().low_128();
tree.insert(&hash, 128, i * 1000);
hashes.push(hash);
}
// Do the actual test -- note that we also test insertion and deletion
// at the root here.
for i in 0u32..10 {
tree.insert(&BitArray::zero(), i as usize, i);
}
for i in 0u32..10 {
let m = tree.lookup(&BitArray::zero(), i as usize);
assert_eq!(m, Some(&i));
}
for i in 0u32..10 {
let m = tree.delete(&BitArray::zero(), i as usize);
assert_eq!(m, Some(i));
}
// Check that the chunder was unharmed
for (n, hash) in hashes.iter().enumerate() {
let ii = ((n + 1) * 1000) as u32;
let ret = tree.lookup(hash, 128);
assert_eq!(ret, Some(&ii));
}
}
#[test]
fn patricia_iter_test() {
let n_elems = 5000;
let mut tree = PatriciaTree::new();
let mut data = vec![None; n_elems];
// Start by inserting a bunch of stuff
for i in 0..n_elems {
let hash = Sha256dHash::from_data(&[(i / 0x100) as u8, (i % 0x100) as u8]).into_le().low_128();
tree.insert(&hash, 128, i);
data[i] = Some(());
}
// Iterate over and try to get everything
for n in tree.iter() {
assert!(data[*n].is_some());
data[*n] = None;
}
// Check that we got everything
assert!(data.iter().all(|opt| opt.is_none()));
}
#[test]
fn patricia_mut_iter_test() {
let n_elems = 5000;
let mut tree = PatriciaTree::new();
let mut data = vec![None; n_elems];
// Start by inserting a bunch of stuff
for i in 0..n_elems {
let hash = Sha256dHash::from_data(&[(i / 0x100) as u8, (i % 0x100) as u8]).into_le().low_128();
tree.insert(&hash, 128, i);
data[i] = Some(());
}
// Iterate over and flip all the values
for n in tree.mut_iter() {
*n = n_elems - *n - 1;
}
// Iterate over and try to get everything
for n in tree.mut_iter() {
assert!(data[*n].is_some());
data[*n] = None;
}
// Check that we got everything
assert!(data.iter().all(|opt| opt.is_none()));
}
#[test]
fn patricia_serialize_test() {
// Build a tree
let mut tree = PatriciaTree::new();
let mut hashes = vec![];
for i in 0u32..5000 {
let hash = Sha256dHash::from_data(&[(i / 0x100) as u8, (i % 0x100) as u8]).into_le().low_128();
tree.insert(&hash, 250, i);
hashes.push(hash);
}
// Serialize it
let serialized = serialize(&tree).unwrap();
// Deserialize it
let deserialized: Result<PatriciaTree<Uint128, u32>, _> = deserialize(&serialized);
assert!(deserialized.is_ok());
let new_tree = deserialized.unwrap();
// Check that all inserts are still there
for (n, hash) in hashes.iter().enumerate() {
let ii = n as u32;
let ret = new_tree.lookup(hash, 250);
assert_eq!(ret, Some(&ii));
}
}
}