refactored chainfilter.rs, split out indexer

This commit is contained in:
debris 2016-02-15 15:22:13 +01:00
parent 8b28f627ca
commit 0fab166fba
5 changed files with 351 additions and 260 deletions

View File

@ -0,0 +1,40 @@
// Copyright 2015, 2016 Ethcore (UK) Ltd.
// This file is part of Parity.
// Parity is free software: you can redistribute it and/or modify
// it under the terms of the GNU General Public License as published by
// the Free Software Foundation, either version 3 of the License, or
// (at your option) any later version.
// Parity is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
// GNU General Public License for more details.
// You should have received a copy of the GNU General Public License
// along with Parity. If not, see <http://www.gnu.org/licenses/>.
//! Represents bloom index in cache
/// Represents bloom index in cache
///
/// On cache level 0, every block bloom is represented by different index.
/// On higher cache levels, multiple block blooms are represented by one
/// index. Their `BloomIndex` can be created from block number and given level.
#[derive(Eq, PartialEq, Hash, Clone, Debug)]
pub struct BloomIndex {
/// Bloom level
pub level: u8,
/// Filter Index
pub index: usize,
}
impl BloomIndex {
/// Default constructor for `BloomIndex`
pub fn new(level: u8, index: usize) -> BloomIndex {
BloomIndex {
level: level,
index: index,
}
}
}

View File

@ -58,70 +58,15 @@
use std::collections::{HashMap}; use std::collections::{HashMap};
use util::hash::*; use util::hash::*;
use util::sha3::*; use util::sha3::*;
use chainfilter::{BloomIndex, FilterDataSource};
/// Represents bloom index in cache use chainfilter::indexer::Indexer;
///
/// On cache level 0, every block bloom is represented by different index.
/// On higher cache levels, multiple block blooms are represented by one
/// index. Their `BloomIndex` can be created from block number and given level.
#[derive(Eq, PartialEq, Hash, Clone, Debug)]
pub struct BloomIndex {
/// Bloom level
pub level: u8,
/// Filter Index
pub index: usize,
}
impl BloomIndex {
/// Default constructor for `BloomIndex`
pub fn new(level: u8, index: usize) -> BloomIndex {
BloomIndex {
level: level,
index: index,
}
}
}
/// Types implementing this trait should provide read access for bloom filters database.
pub trait FilterDataSource {
/// returns reference to log at given position if it exists
fn bloom_at_index(&self, index: &BloomIndex) -> Option<H2048>;
}
/// In memory cache for blooms.
///
/// Stores all blooms in HashMap, which indexes them by `BloomIndex`.
pub struct MemoryCache {
blooms: HashMap<BloomIndex, H2048>,
}
impl MemoryCache {
/// Default constructor for MemoryCache
pub fn new() -> MemoryCache {
MemoryCache { blooms: HashMap::new() }
}
/// inserts all blooms into cache
///
/// if bloom at given index already exists, overwrites it
pub fn insert_blooms(&mut self, blooms: HashMap<BloomIndex, H2048>) {
self.blooms.extend(blooms);
}
}
impl FilterDataSource for MemoryCache {
fn bloom_at_index(&self, index: &BloomIndex) -> Option<H2048> {
self.blooms.get(index).cloned()
}
}
/// Should be used for search operations on blockchain. /// Should be used for search operations on blockchain.
pub struct ChainFilter<'a, D> pub struct ChainFilter<'a, D>
where D: FilterDataSource + 'a where D: FilterDataSource + 'a
{ {
data_source: &'a D, data_source: &'a D,
index_size: usize, indexer: Indexer
level_sizes: Vec<usize>,
} }
impl<'a, D> ChainFilter<'a, D> where D: FilterDataSource impl<'a, D> ChainFilter<'a, D> where D: FilterDataSource
@ -130,73 +75,15 @@ impl<'a, D> ChainFilter<'a, D> where D: FilterDataSource
/// ///
/// Borrows `FilterDataSource` for reading. /// Borrows `FilterDataSource` for reading.
pub fn new(data_source: &'a D, index_size: usize, levels: u8) -> Self { pub fn new(data_source: &'a D, index_size: usize, levels: u8) -> Self {
if levels == 0 { ChainFilter {
panic!("ChainFilter requires at least 1 level");
}
let mut filter = ChainFilter {
data_source: data_source, data_source: data_source,
index_size: index_size, indexer: Indexer::new(index_size, levels)
// 0 level has always a size of 1
level_sizes: vec![1]
};
// cache level sizes, so we do not have to calculate them all the time
// eg. if levels == 3, index_size = 16
// level_sizes = [1, 16, 256]
let additional: Vec<usize> = (1..).into_iter()
.scan(1, |acc, _| {
*acc = *acc * index_size;
Some(*acc)
})
.take(levels as usize - 1)
.collect();
filter.level_sizes.extend(additional);
filter
}
/// unsafely get level size
fn level_size(&self, level: u8) -> usize {
self.level_sizes[level as usize]
}
/// converts block number and level to `BloomIndex`
fn bloom_index(&self, block_number: usize, level: u8) -> BloomIndex {
BloomIndex {
level: level,
index: block_number / self.level_size(level),
} }
} }
/// return bloom which are dependencies for given index
///
/// bloom indexes are ordered from lowest to highest
fn lower_level_bloom_indexes(&self, index: &BloomIndex) -> Vec<BloomIndex> {
// this is the lowest level
if index.level == 0 {
return vec![];
}
let new_level = index.level - 1;
let offset = self.index_size * index.index;
(0..self.index_size).map(|i| BloomIndex::new(new_level, offset + i)).collect()
}
/// return number of levels
fn levels(&self) -> u8 {
self.level_sizes.len() as u8
}
/// returns max filter level
fn max_level(&self) -> u8 {
self.level_sizes.len() as u8 - 1
}
/// internal function which does bloom search recursively /// internal function which does bloom search recursively
fn blocks(&self, bloom: &H2048, from_block: usize, to_block: usize, level: u8, offset: usize) -> Option<Vec<usize>> { fn blocks(&self, bloom: &H2048, from_block: usize, to_block: usize, level: u8, offset: usize) -> Option<Vec<usize>> {
let index = self.bloom_index(offset, level); let index = self.indexer.bloom_index(offset, level);
match self.data_source.bloom_at_index(&index) { match self.data_source.bloom_at_index(&index) {
None => return None, None => return None,
@ -213,10 +100,10 @@ impl<'a, D> ChainFilter<'a, D> where D: FilterDataSource
} }
}; };
let level_size = self.level_size(level - 1); let level_size = self.indexer.level_size(level - 1);
let from_index = self.bloom_index(from_block, level - 1); let from_index = self.indexer.bloom_index(from_block, level - 1);
let to_index = self.bloom_index(to_block, level - 1); let to_index = self.indexer.bloom_index(to_block, level - 1);
let res: Vec<usize> = self.lower_level_bloom_indexes(&index).into_iter() let res: Vec<usize> = self.indexer.lower_level_bloom_indexes(&index).into_iter()
// chose only blooms in range // chose only blooms in range
.filter(|li| li.index >= from_index.index && li.index <= to_index.index) .filter(|li| li.index >= from_index.index && li.index <= to_index.index)
// map them to offsets // map them to offsets
@ -235,8 +122,8 @@ impl<'a, D> ChainFilter<'a, D> where D: FilterDataSource
pub fn add_bloom(&self, bloom: &H2048, block_number: usize) -> HashMap<BloomIndex, H2048> { pub fn add_bloom(&self, bloom: &H2048, block_number: usize) -> HashMap<BloomIndex, H2048> {
let mut result: HashMap<BloomIndex, H2048> = HashMap::new(); let mut result: HashMap<BloomIndex, H2048> = HashMap::new();
for level in 0..self.levels() { for level in 0..self.indexer.levels() {
let bloom_index = self.bloom_index(block_number, level); let bloom_index = self.indexer.bloom_index(block_number, level);
let new_bloom = match self.data_source.bloom_at_index(&bloom_index) { let new_bloom = match self.data_source.bloom_at_index(&bloom_index) {
Some(old_bloom) => old_bloom | bloom.clone(), Some(old_bloom) => old_bloom | bloom.clone(),
None => bloom.clone(), None => bloom.clone(),
@ -252,9 +139,9 @@ impl<'a, D> ChainFilter<'a, D> where D: FilterDataSource
pub fn add_blooms(&self, blooms: &[H2048], block_number: usize) -> HashMap<BloomIndex, H2048> { pub fn add_blooms(&self, blooms: &[H2048], block_number: usize) -> HashMap<BloomIndex, H2048> {
let mut result: HashMap<BloomIndex, H2048> = HashMap::new(); let mut result: HashMap<BloomIndex, H2048> = HashMap::new();
for level in 0..self.levels() { for level in 0..self.indexer.levels() {
for i in 0..blooms.len() { for i in 0..blooms.len() {
let bloom_index = self.bloom_index(block_number + i, level); let bloom_index = self.indexer.bloom_index(block_number + i, level);
let is_new_bloom = match result.get_mut(&bloom_index) { let is_new_bloom = match result.get_mut(&bloom_index) {
// it was already modified // it was already modified
@ -283,13 +170,13 @@ impl<'a, D> ChainFilter<'a, D> where D: FilterDataSource
pub fn reset_bloom(&self, bloom: &H2048, block_number: usize) -> HashMap<BloomIndex, H2048> { pub fn reset_bloom(&self, bloom: &H2048, block_number: usize) -> HashMap<BloomIndex, H2048> {
let mut result: HashMap<BloomIndex, H2048> = HashMap::new(); let mut result: HashMap<BloomIndex, H2048> = HashMap::new();
let mut reset_index = self.bloom_index(block_number, 0); let mut reset_index = self.indexer.bloom_index(block_number, 0);
result.insert(reset_index.clone(), bloom.clone()); result.insert(reset_index.clone(), bloom.clone());
for level in 1..self.levels() { for level in 1..self.indexer.levels() {
let index = self.bloom_index(block_number, level); let index = self.indexer.bloom_index(block_number, level);
// get all bloom indexes that were used to construct this bloom // get all bloom indexes that were used to construct this bloom
let lower_indexes = self.lower_level_bloom_indexes(&index); let lower_indexes = self.indexer.lower_level_bloom_indexes(&index);
let new_bloom = lower_indexes.into_iter() let new_bloom = lower_indexes.into_iter()
// skip reseted one // skip reseted one
.filter(|li| li != &reset_index) .filter(|li| li != &reset_index)
@ -313,23 +200,23 @@ impl<'a, D> ChainFilter<'a, D> where D: FilterDataSource
// insert all new blooms at level 0 // insert all new blooms at level 0
for (i, bloom) in blooms.iter().enumerate() { for (i, bloom) in blooms.iter().enumerate() {
result.insert(self.bloom_index(block_number + i, 0), bloom.clone()); result.insert(self.indexer.bloom_index(block_number + i, 0), bloom.clone());
} }
// reset the rest of blooms // reset the rest of blooms
for reset_number in block_number + blooms.len()..old_highest_block { for reset_number in block_number + blooms.len()..old_highest_block {
result.insert(self.bloom_index(reset_number, 0), H2048::new()); result.insert(self.indexer.bloom_index(reset_number, 0), H2048::new());
} }
for level in 1..self.levels() { for level in 1..self.indexer.levels() {
for i in 0..blooms.len() { for i in 0..blooms.len() {
let index = self.bloom_index(block_number + i, level); let index = self.indexer.bloom_index(block_number + i, level);
let new_bloom = { let new_bloom = {
// use new blooms before db blooms where necessary // use new blooms before db blooms where necessary
let bloom_at = | index | { result.get(&index).cloned().or_else(|| self.data_source.bloom_at_index(&index)) }; let bloom_at = | index | { result.get(&index).cloned().or_else(|| self.data_source.bloom_at_index(&index)) };
self.lower_level_bloom_indexes(&index) self.indexer.lower_level_bloom_indexes(&index)
.into_iter() .into_iter()
// get blooms // get blooms
.map(bloom_at) .map(bloom_at)
@ -369,10 +256,10 @@ impl<'a, D> ChainFilter<'a, D> where D: FilterDataSource
pub fn blocks_with_bloom(&self, bloom: &H2048, from_block: usize, to_block: usize) -> Vec<usize> { pub fn blocks_with_bloom(&self, bloom: &H2048, from_block: usize, to_block: usize) -> Vec<usize> {
let mut result = vec![]; let mut result = vec![];
// lets start from highest level // lets start from highest level
let max_level = self.max_level(); let max_level = self.indexer.max_level();
let level_size = self.level_size(max_level); let level_size = self.indexer.level_size(max_level);
let from_index = self.bloom_index(from_block, max_level); let from_index = self.indexer.bloom_index(from_block, max_level);
let to_index = self.bloom_index(to_block, max_level); let to_index = self.indexer.bloom_index(to_block, max_level);
for index in from_index.index..to_index.index + 1 { for index in from_index.index..to_index.index + 1 {
// offset will be used to calculate where we are right now // offset will be used to calculate where we are right now
@ -387,123 +274,3 @@ impl<'a, D> ChainFilter<'a, D> where D: FilterDataSource
result result
} }
} }
#[cfg(test)]
mod tests {
use std::str::FromStr;
use util::hash::*;
use chainfilter::*;
use util::sha3::*;
#[test]
fn test_level_size() {
let cache = MemoryCache::new();
let filter = ChainFilter::new(&cache, 16, 3);
assert_eq!(filter.level_size(0), 1);
assert_eq!(filter.level_size(1), 16);
assert_eq!(filter.level_size(2), 256);
}
#[test]
fn test_bloom_index() {
let cache = MemoryCache::new();
let filter = ChainFilter::new(&cache, 16, 3);
let bi0 = filter.bloom_index(0, 0);
assert_eq!(bi0.level, 0);
assert_eq!(bi0.index, 0);
let bi1 = filter.bloom_index(1, 0);
assert_eq!(bi1.level, 0);
assert_eq!(bi1.index, 1);
let bi2 = filter.bloom_index(2, 0);
assert_eq!(bi2.level, 0);
assert_eq!(bi2.index, 2);
let bi3 = filter.bloom_index(3, 1);
assert_eq!(bi3.level, 1);
assert_eq!(bi3.index, 0);
let bi4 = filter.bloom_index(15, 1);
assert_eq!(bi4.level, 1);
assert_eq!(bi4.index, 0);
let bi5 = filter.bloom_index(16, 1);
assert_eq!(bi5.level, 1);
assert_eq!(bi5.index, 1);
let bi6 = filter.bloom_index(255, 2);
assert_eq!(bi6.level, 2);
assert_eq!(bi6.index, 0);
let bi7 = filter.bloom_index(256, 2);
assert_eq!(bi7.level, 2);
assert_eq!(bi7.index, 1);
}
#[test]
fn test_lower_level_bloom_indexes() {
let cache = MemoryCache::new();
let filter = ChainFilter::new(&cache, 16, 3);
let bi = filter.bloom_index(256, 2);
assert_eq!(bi.level, 2);
assert_eq!(bi.index, 1);
let mut ebis = vec![];
for i in 16..32 {
ebis.push(BloomIndex::new(1, i));
}
let bis = filter.lower_level_bloom_indexes(&bi);
assert_eq!(ebis, bis);
}
#[test]
fn test_topic_basic_search() {
let index_size = 16;
let bloom_levels = 3;
let mut cache = MemoryCache::new();
let topic = H256::from_str("8d936b1bd3fc635710969ccfba471fb17d598d9d1971b538dd712e1e4b4f4dba").unwrap();
let modified_blooms = {
let filter = ChainFilter::new(&cache, index_size, bloom_levels);
let block_number = 23;
let mut bloom = H2048::new();
bloom.shift_bloomed(&topic.sha3());
filter.add_bloom(&bloom, block_number)
};
// number of modified blooms should always be equal number of levels
assert_eq!(modified_blooms.len(), bloom_levels as usize);
cache.insert_blooms(modified_blooms);
{
let filter = ChainFilter::new(&cache, index_size, bloom_levels);
let blocks = filter.blocks_with_topic(&topic, 0, 100);
assert_eq!(blocks.len(), 1);
assert_eq!(blocks[0], 23);
}
{
let filter = ChainFilter::new(&cache, index_size, bloom_levels);
let blocks = filter.blocks_with_topic(&topic, 0, 23);
assert_eq!(blocks.len(), 0);
}
{
let filter = ChainFilter::new(&cache, index_size, bloom_levels);
let blocks = filter.blocks_with_topic(&topic, 23, 24);
assert_eq!(blocks.len(), 1);
assert_eq!(blocks[0], 23);
}
{
let filter = ChainFilter::new(&cache, index_size, bloom_levels);
let blocks = filter.blocks_with_topic(&topic, 24, 100);
assert_eq!(blocks.len(), 0);
}
}
}

View File

@ -0,0 +1,154 @@
// Copyright 2015, 2016 Ethcore (UK) Ltd.
// This file is part of Parity.
// Parity is free software: you can redistribute it and/or modify
// it under the terms of the GNU General Public License as published by
// the Free Software Foundation, either version 3 of the License, or
// (at your option) any later version.
// Parity is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
// GNU General Public License for more details.
// You should have received a copy of the GNU General Public License
// along with Parity. If not, see <http://www.gnu.org/licenses/>.
//! Simplifies working with bloom indexes.
use chainfilter::BloomIndex;
/// Simplifies working with bloom indexes.
pub struct Indexer {
index_size: usize,
level_sizes: Vec<usize>
}
impl Indexer {
/// Creates new indexer.
pub fn new(index_size: usize, levels: u8) -> Self {
if levels == 0 {
panic!("Indexer requires at least 1 level.");
}
let mut level_sizes = vec![1];
level_sizes.extend_from_slice(&(1..).into_iter()
.scan(1, |acc, _| {
*acc = *acc * index_size;
Some(*acc)
})
.take(levels as usize - 1)
.collect::<Vec<usize>>());
Indexer {
index_size: index_size,
level_sizes: level_sizes,
}
}
/// unsafely get level size.
pub fn level_size(&self, level: u8) -> usize {
self.level_sizes[level as usize]
}
/// Converts block number and level to `BloomIndex`.
pub fn bloom_index(&self, block_number: usize, level: u8) -> BloomIndex {
BloomIndex {
level: level,
index: block_number / self.level_size(level),
}
}
/// Return bloom which are dependencies for given index.
///
/// Bloom indexes are ordered from lowest to highest.
pub fn lower_level_bloom_indexes(&self, index: &BloomIndex) -> Vec<BloomIndex> {
// this is the lowest level
if index.level == 0 {
return vec![];
}
let new_level = index.level - 1;
let offset = self.index_size * index.index;
(0..self.index_size).map(|i| BloomIndex::new(new_level, offset + i)).collect()
}
/// Return number of levels.
pub fn levels(&self) -> u8 {
self.level_sizes.len() as u8
}
/// Returns max indexer level.
pub fn max_level(&self) -> u8 {
self.level_sizes.len() as u8 - 1
}
}
#[cfg(test)]
mod tests {
use chainfilter::BloomIndex;
use chainfilter::indexer::Indexer;
#[test]
fn test_level_size() {
let indexer = Indexer::new(16, 3);
assert_eq!(indexer.level_size(0), 1);
assert_eq!(indexer.level_size(1), 16);
assert_eq!(indexer.level_size(2), 256);
}
#[test]
fn test_bloom_index() {
let indexer = Indexer::new(16, 3);
let bi0 = indexer.bloom_index(0, 0);
assert_eq!(bi0.level, 0);
assert_eq!(bi0.index, 0);
let bi1 = indexer.bloom_index(1, 0);
assert_eq!(bi1.level, 0);
assert_eq!(bi1.index, 1);
let bi2 = indexer.bloom_index(2, 0);
assert_eq!(bi2.level, 0);
assert_eq!(bi2.index, 2);
let bi3 = indexer.bloom_index(3, 1);
assert_eq!(bi3.level, 1);
assert_eq!(bi3.index, 0);
let bi4 = indexer.bloom_index(15, 1);
assert_eq!(bi4.level, 1);
assert_eq!(bi4.index, 0);
let bi5 = indexer.bloom_index(16, 1);
assert_eq!(bi5.level, 1);
assert_eq!(bi5.index, 1);
let bi6 = indexer.bloom_index(255, 2);
assert_eq!(bi6.level, 2);
assert_eq!(bi6.index, 0);
let bi7 = indexer.bloom_index(256, 2);
assert_eq!(bi7.level, 2);
assert_eq!(bi7.index, 1);
}
#[test]
fn test_lower_level_bloom_indexes() {
let indexer = Indexer::new(16, 3);
let bi = indexer.bloom_index(256, 2);
assert_eq!(bi.level, 2);
assert_eq!(bi.index, 1);
let mut ebis = vec![];
for i in 16..32 {
ebis.push(BloomIndex::new(1, i));
}
let bis = indexer.lower_level_bloom_indexes(&bi);
assert_eq!(ebis, bis);
}
}

View File

@ -0,0 +1,35 @@
// Copyright 2015, 2016 Ethcore (UK) Ltd.
// This file is part of Parity.
// Parity is free software: you can redistribute it and/or modify
// it under the terms of the GNU General Public License as published by
// the Free Software Foundation, either version 3 of the License, or
// (at your option) any later version.
// Parity is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
// GNU General Public License for more details.
// You should have received a copy of the GNU General Public License
// along with Parity. If not, see <http://www.gnu.org/licenses/>.
//! Multilevel blockchain bloom filter.
mod bloomindex;
mod chainfilter;
mod indexer;
#[cfg(test)]
mod tests;
pub use self::bloomindex::BloomIndex;
pub use self::chainfilter::ChainFilter;
use util::hash::H2048;
/// Types implementing this trait provide read access for bloom filters database.
pub trait FilterDataSource {
/// returns reference to log at given position if it exists
fn bloom_at_index(&self, index: &BloomIndex) -> Option<H2048>;
}

View File

@ -0,0 +1,95 @@
// Copyright 2015, 2016 Ethcore (UK) Ltd.
// This file is part of Parity.
// Parity is free software: you can redistribute it and/or modify
// it under the terms of the GNU General Public License as published by
// the Free Software Foundation, either version 3 of the License, or
// (at your option) any later version.
// Parity is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
// GNU General Public License for more details.
// You should have received a copy of the GNU General Public License
// along with Parity. If not, see <http://www.gnu.org/licenses/>.
use std::collections::HashMap;
use std::str::FromStr;
use util::hash::*;
use util::sha3::*;
use chainfilter::{BloomIndex, FilterDataSource, ChainFilter};
/// In memory cache for blooms.
///
/// Stores all blooms in HashMap, which indexes them by `BloomIndex`.
pub struct MemoryCache {
blooms: HashMap<BloomIndex, H2048>,
}
impl MemoryCache {
/// Default constructor for MemoryCache
pub fn new() -> MemoryCache {
MemoryCache { blooms: HashMap::new() }
}
/// inserts all blooms into cache
///
/// if bloom at given index already exists, overwrites it
pub fn insert_blooms(&mut self, blooms: HashMap<BloomIndex, H2048>) {
self.blooms.extend(blooms);
}
}
impl FilterDataSource for MemoryCache {
fn bloom_at_index(&self, index: &BloomIndex) -> Option<H2048> {
self.blooms.get(index).cloned()
}
}
#[test]
fn test_topic_basic_search() {
let index_size = 16;
let bloom_levels = 3;
let mut cache = MemoryCache::new();
let topic = H256::from_str("8d936b1bd3fc635710969ccfba471fb17d598d9d1971b538dd712e1e4b4f4dba").unwrap();
let modified_blooms = {
let filter = ChainFilter::new(&cache, index_size, bloom_levels);
let block_number = 23;
let mut bloom = H2048::new();
bloom.shift_bloomed(&topic.sha3());
filter.add_bloom(&bloom, block_number)
};
// number of modified blooms should always be equal number of levels
assert_eq!(modified_blooms.len(), bloom_levels as usize);
cache.insert_blooms(modified_blooms);
{
let filter = ChainFilter::new(&cache, index_size, bloom_levels);
let blocks = filter.blocks_with_topic(&topic, 0, 100);
assert_eq!(blocks.len(), 1);
assert_eq!(blocks[0], 23);
}
{
let filter = ChainFilter::new(&cache, index_size, bloom_levels);
let blocks = filter.blocks_with_topic(&topic, 0, 23);
assert_eq!(blocks.len(), 0);
}
{
let filter = ChainFilter::new(&cache, index_size, bloom_levels);
let blocks = filter.blocks_with_topic(&topic, 23, 24);
assert_eq!(blocks.len(), 1);
assert_eq!(blocks[0], 23);
}
{
let filter = ChainFilter::new(&cache, index_size, bloom_levels);
let blocks = filter.blocks_with_topic(&topic, 24, 100);
assert_eq!(blocks.len(), 0);
}
}