mirror of
https://github.com/ClickHouse/ClickHouse.git
synced 2024-11-10 09:32:06 +00:00
460 lines
13 KiB
C++
460 lines
13 KiB
C++
//
|
|
// LinearHashTable.h
|
|
//
|
|
// Library: Foundation
|
|
// Package: Hashing
|
|
// Module: LinearHashTable
|
|
//
|
|
// Definition of the LinearHashTable class.
|
|
//
|
|
// Copyright (c) 2006, Applied Informatics Software Engineering GmbH.
|
|
// and Contributors.
|
|
//
|
|
// SPDX-License-Identifier: BSL-1.0
|
|
//
|
|
|
|
|
|
#ifndef Foundation_LinearHashTable_INCLUDED
|
|
#define Foundation_LinearHashTable_INCLUDED
|
|
|
|
|
|
#include <algorithm>
|
|
#include <cstddef>
|
|
#include <functional>
|
|
#include <utility>
|
|
#include <vector>
|
|
#include "Poco/Foundation.h"
|
|
#include "Poco/Hash.h"
|
|
|
|
|
|
namespace Poco
|
|
{
|
|
|
|
|
|
template <class Value, class HashFunc = Hash<Value>>
|
|
class LinearHashTable
|
|
/// This class implements a linear hash table.
|
|
///
|
|
/// In a linear hash table, the available address space
|
|
/// grows or shrinks dynamically. A linear hash table thus
|
|
/// supports any number of insertions or deletions without
|
|
/// lookup or insertion performance deterioration.
|
|
///
|
|
/// Linear hashing was discovered by Witold Litwin in 1980
|
|
/// and described in the paper LINEAR HASHING: A NEW TOOL FOR FILE AND TABLE ADDRESSING.
|
|
///
|
|
/// For more information on linear hashing, see <http://en.wikipedia.org/wiki/Linear_hash>.
|
|
///
|
|
/// The LinearHashTable is not thread safe.
|
|
///
|
|
/// Value must support comparison for equality.
|
|
///
|
|
/// Find, insert and delete operations are basically O(1) with regard
|
|
/// to the total number of elements in the table, and O(N) with regard
|
|
/// to the number of elements in the bucket where the element is stored.
|
|
/// On average, every bucket stores one element; the exact number depends
|
|
/// on the quality of the hash function. In most cases, the maximum number of
|
|
/// elements in a bucket should not exceed 3.
|
|
{
|
|
public:
|
|
typedef Value ValueType;
|
|
typedef Value & Reference;
|
|
typedef const Value & ConstReference;
|
|
typedef Value * Pointer;
|
|
typedef const Value * ConstPointer;
|
|
typedef HashFunc Hash;
|
|
typedef std::vector<Value> Bucket;
|
|
typedef std::vector<Bucket> BucketVec;
|
|
typedef typename Bucket::iterator BucketIterator;
|
|
typedef typename BucketVec::iterator BucketVecIterator;
|
|
|
|
class ConstIterator : public std::iterator<std::forward_iterator_tag, Value>
|
|
{
|
|
public:
|
|
ConstIterator() : _initialized(false) { }
|
|
|
|
ConstIterator(const BucketVecIterator & vecIt, const BucketVecIterator & endIt, const BucketIterator & buckIt)
|
|
: _vecIt(vecIt), _endIt(endIt), _buckIt(buckIt), _initialized(true)
|
|
{
|
|
}
|
|
|
|
ConstIterator(const ConstIterator & it) : _vecIt(it._vecIt), _endIt(it._endIt), _buckIt(it._buckIt), _initialized(it._initialized)
|
|
|
|
{
|
|
}
|
|
|
|
ConstIterator & operator=(const ConstIterator & it)
|
|
{
|
|
ConstIterator tmp(it);
|
|
swap(tmp);
|
|
return *this;
|
|
}
|
|
|
|
void swap(ConstIterator & it)
|
|
{
|
|
using std::swap;
|
|
// uninitialized iterators crash when swapped
|
|
if (_initialized)
|
|
{
|
|
swap(_vecIt, it._vecIt);
|
|
swap(_endIt, it._endIt);
|
|
swap(_buckIt, it._buckIt);
|
|
swap(_initialized, it._initialized);
|
|
}
|
|
else
|
|
{
|
|
_vecIt = it._vecIt;
|
|
_endIt = it._endIt;
|
|
_buckIt = it._buckIt;
|
|
_initialized = it._initialized;
|
|
}
|
|
}
|
|
|
|
bool operator==(const ConstIterator & it) const { return _vecIt == it._vecIt && (_vecIt == _endIt || _buckIt == it._buckIt); }
|
|
|
|
bool operator!=(const ConstIterator & it) const { return _vecIt != it._vecIt || (_vecIt != _endIt && _buckIt != it._buckIt); }
|
|
|
|
const typename Bucket::value_type & operator*() const { return *_buckIt; }
|
|
|
|
const typename Bucket::value_type * operator->() const { return &*_buckIt; }
|
|
|
|
ConstIterator & operator++() // prefix
|
|
{
|
|
if (_vecIt != _endIt)
|
|
{
|
|
++_buckIt;
|
|
while (_vecIt != _endIt && _buckIt == _vecIt->end())
|
|
{
|
|
++_vecIt;
|
|
if (_vecIt != _endIt)
|
|
_buckIt = _vecIt->begin();
|
|
}
|
|
}
|
|
return *this;
|
|
}
|
|
|
|
ConstIterator operator++(int) // postfix
|
|
{
|
|
ConstIterator tmp(*this);
|
|
++*this;
|
|
return tmp;
|
|
}
|
|
|
|
protected:
|
|
BucketVecIterator _vecIt;
|
|
BucketVecIterator _endIt;
|
|
BucketIterator _buckIt;
|
|
bool _initialized;
|
|
|
|
friend class LinearHashTable;
|
|
};
|
|
|
|
class Iterator : public ConstIterator
|
|
{
|
|
public:
|
|
Iterator() { }
|
|
|
|
Iterator(const BucketVecIterator & vecIt, const BucketVecIterator & endIt, const BucketIterator & buckIt)
|
|
: ConstIterator(vecIt, endIt, buckIt)
|
|
{
|
|
}
|
|
|
|
Iterator(const Iterator & it) : ConstIterator(it) { }
|
|
|
|
Iterator & operator=(const Iterator & it)
|
|
{
|
|
Iterator tmp(it);
|
|
ConstIterator::swap(tmp);
|
|
return *this;
|
|
}
|
|
|
|
void swap(Iterator & it) { ConstIterator::swap(it); }
|
|
|
|
typename Bucket::value_type & operator*() { return *this->_buckIt; }
|
|
|
|
const typename Bucket::value_type & operator*() const { return *this->_buckIt; }
|
|
|
|
typename Bucket::value_type * operator->() { return &*this->_buckIt; }
|
|
|
|
const typename Bucket::value_type * operator->() const { return &*this->_buckIt; }
|
|
|
|
Iterator & operator++() // prefix
|
|
{
|
|
ConstIterator::operator++();
|
|
return *this;
|
|
}
|
|
|
|
Iterator operator++(int) // postfix
|
|
{
|
|
Iterator tmp(*this);
|
|
++*this;
|
|
return tmp;
|
|
}
|
|
|
|
friend class LinearHashTable;
|
|
};
|
|
|
|
LinearHashTable(std::size_t initialReserve = 64) : _split(0), _front(1), _size(0)
|
|
/// Creates the LinearHashTable, using the given initialReserve.
|
|
{
|
|
_buckets.reserve(calcSize(initialReserve));
|
|
_buckets.push_back(Bucket());
|
|
}
|
|
|
|
LinearHashTable(const LinearHashTable & table)
|
|
: _buckets(table._buckets), _split(table._split), _front(table._front), _size(table._size)
|
|
/// Creates the LinearHashTable by copying another one.
|
|
{
|
|
}
|
|
|
|
~LinearHashTable()
|
|
/// Destroys the LinearHashTable.
|
|
{
|
|
}
|
|
|
|
LinearHashTable & operator=(const LinearHashTable & table)
|
|
/// Assigns another LinearHashTable.
|
|
{
|
|
LinearHashTable tmp(table);
|
|
swap(tmp);
|
|
return *this;
|
|
}
|
|
|
|
void swap(LinearHashTable & table)
|
|
/// Swaps the LinearHashTable with another one.
|
|
{
|
|
using std::swap;
|
|
swap(_buckets, table._buckets);
|
|
swap(_split, table._split);
|
|
swap(_front, table._front);
|
|
swap(_size, table._size);
|
|
}
|
|
|
|
ConstIterator begin() const
|
|
/// Returns an iterator pointing to the first entry, if one exists.
|
|
{
|
|
BucketVecIterator it(_buckets.begin());
|
|
BucketVecIterator end(_buckets.end());
|
|
while (it != end && it->empty())
|
|
{
|
|
++it;
|
|
}
|
|
if (it == end)
|
|
return this->end();
|
|
else
|
|
return ConstIterator(it, end, it->begin());
|
|
}
|
|
|
|
ConstIterator end() const
|
|
/// Returns an iterator pointing to the end of the table.
|
|
{
|
|
return ConstIterator(_buckets.end(), _buckets.end(), _buckets.front().end());
|
|
}
|
|
|
|
Iterator begin()
|
|
/// Returns an iterator pointing to the first entry, if one exists.
|
|
{
|
|
BucketVecIterator it(_buckets.begin());
|
|
BucketVecIterator end(_buckets.end());
|
|
while (it != end && it->empty())
|
|
{
|
|
++it;
|
|
}
|
|
if (it == end)
|
|
return this->end();
|
|
else
|
|
return Iterator(it, end, it->begin());
|
|
}
|
|
|
|
Iterator end()
|
|
/// Returns an iterator pointing to the end of the table.
|
|
{
|
|
return Iterator(_buckets.end(), _buckets.end(), _buckets.front().end());
|
|
}
|
|
|
|
ConstIterator find(const Value & value) const
|
|
/// Finds an entry in the table.
|
|
{
|
|
std::size_t addr = bucketAddress(value);
|
|
BucketVecIterator it(_buckets.begin() + addr);
|
|
BucketIterator buckIt(std::find(it->begin(), it->end(), value));
|
|
if (buckIt != it->end())
|
|
return ConstIterator(it, _buckets.end(), buckIt);
|
|
else
|
|
return end();
|
|
}
|
|
|
|
Iterator find(const Value & value)
|
|
/// Finds an entry in the table.
|
|
{
|
|
std::size_t addr = bucketAddress(value);
|
|
BucketVecIterator it(_buckets.begin() + addr);
|
|
BucketIterator buckIt(std::find(it->begin(), it->end(), value));
|
|
if (buckIt != it->end())
|
|
return Iterator(it, _buckets.end(), buckIt);
|
|
else
|
|
return end();
|
|
}
|
|
|
|
std::size_t count(const Value & value) const
|
|
/// Returns the number of elements with the given
|
|
/// value, with is either 1 or 0.
|
|
{
|
|
return find(value) != end() ? 1 : 0;
|
|
}
|
|
|
|
std::pair<Iterator, bool> insert(const Value & value)
|
|
/// Inserts an element into the table.
|
|
///
|
|
/// If the element already exists in the table,
|
|
/// a pair(iterator, false) with iterator pointing to the
|
|
/// existing element is returned.
|
|
/// Otherwise, the element is inserted an a
|
|
/// pair(iterator, true) with iterator
|
|
/// pointing to the new element is returned.
|
|
{
|
|
std::size_t hash = _hash(value);
|
|
std::size_t addr = bucketAddressForHash(hash);
|
|
BucketVecIterator it(_buckets.begin() + addr);
|
|
BucketIterator buckIt(std::find(it->begin(), it->end(), value));
|
|
if (buckIt == it->end())
|
|
{
|
|
split();
|
|
addr = bucketAddressForHash(hash);
|
|
it = _buckets.begin() + addr;
|
|
buckIt = it->insert(it->end(), value);
|
|
++_size;
|
|
return std::make_pair(Iterator(it, _buckets.end(), buckIt), true);
|
|
}
|
|
else
|
|
{
|
|
return std::make_pair(Iterator(it, _buckets.end(), buckIt), false);
|
|
}
|
|
}
|
|
|
|
void erase(Iterator it)
|
|
/// Erases the element pointed to by it.
|
|
{
|
|
if (it != end())
|
|
{
|
|
it._vecIt->erase(it._buckIt);
|
|
--_size;
|
|
merge();
|
|
}
|
|
}
|
|
|
|
void erase(const Value & value)
|
|
/// Erases the element with the given value, if it exists.
|
|
{
|
|
Iterator it = find(value);
|
|
erase(it);
|
|
}
|
|
|
|
void clear()
|
|
/// Erases all elements.
|
|
{
|
|
LinearHashTable empty;
|
|
swap(empty);
|
|
}
|
|
|
|
std::size_t size() const
|
|
/// Returns the number of elements in the table.
|
|
{
|
|
return _size;
|
|
}
|
|
|
|
bool empty() const
|
|
/// Returns true iff the table is empty.
|
|
{
|
|
return _size == 0;
|
|
}
|
|
|
|
std::size_t buckets() const
|
|
/// Returns the number of allocated buckets.
|
|
{
|
|
return _buckets.size();
|
|
}
|
|
|
|
protected:
|
|
std::size_t bucketAddress(const Value & value) const
|
|
{
|
|
std::size_t n = _hash(value);
|
|
if (n % _front >= _split)
|
|
return n % _front;
|
|
else
|
|
return n % (2 * _front);
|
|
}
|
|
|
|
std::size_t bucketAddressForHash(std::size_t hash)
|
|
{
|
|
if (hash % _front >= _split)
|
|
return hash % _front;
|
|
else
|
|
return hash % (2 * _front);
|
|
}
|
|
|
|
void split()
|
|
{
|
|
if (_split == _front)
|
|
{
|
|
_split = 0;
|
|
_front *= 2;
|
|
_buckets.reserve(_front * 2);
|
|
}
|
|
Bucket tmp;
|
|
_buckets.push_back(tmp);
|
|
_buckets[_split].swap(tmp);
|
|
++_split;
|
|
for (BucketIterator it = tmp.begin(); it != tmp.end(); ++it)
|
|
{
|
|
using std::swap;
|
|
std::size_t addr = bucketAddress(*it);
|
|
_buckets[addr].push_back(Value());
|
|
swap(*it, _buckets[addr].back());
|
|
}
|
|
}
|
|
|
|
void merge()
|
|
{
|
|
if (_split == 0)
|
|
{
|
|
_front /= 2;
|
|
_split = _front;
|
|
}
|
|
--_split;
|
|
Bucket tmp;
|
|
tmp.swap(_buckets.back());
|
|
_buckets.pop_back();
|
|
for (BucketIterator it = tmp.begin(); it != tmp.end(); ++it)
|
|
{
|
|
using std::swap;
|
|
std::size_t addr = bucketAddress(*it);
|
|
_buckets[addr].push_back(Value());
|
|
swap(*it, _buckets[addr].back());
|
|
}
|
|
}
|
|
|
|
static std::size_t calcSize(std::size_t initialSize)
|
|
{
|
|
std::size_t size = 32;
|
|
while (size < initialSize)
|
|
size *= 2;
|
|
return size;
|
|
}
|
|
|
|
private:
|
|
// Evil hack: _buckets must be mutable because both ConstIterator and Iterator hold
|
|
// ordinary iterator's (not const_iterator's).
|
|
mutable BucketVec _buckets;
|
|
std::size_t _split;
|
|
std::size_t _front;
|
|
std::size_t _size;
|
|
HashFunc _hash;
|
|
};
|
|
|
|
|
|
} // namespace Poco
|
|
|
|
|
|
#endif // Foundation_LinearHashTable_INCLUDED
|