I want to implement a custom iterator class that has some differences with typical iterators. The container in this example is a vector and the idea is to walk throw it but using a specific syntax, like this:
int main()
{
VectorElement E;
for (size_t i = 0; i < 5; i++)
{
Element e;
e.id = i;
E.push_back(e);
}
for (ElementIterator e(E); e.end(); ++e)
cout << "Element " << e.id << endl;
}
I tried with the following definitions, but it gives me a Segmentation Fault at the static_cast line after two iterations. Does someone know how to improve this? Thanks
class Element
{
public:
Element() {};
size_t id;
};
typedef vector<Element> VectorElement;
class ElementIterator: public Element
{
typedef vector<Element>::iterator iter;
public:
ElementIterator(const Element& e) : Element(e) { }
ElementIterator(VectorElement& ve_)
{
ve = &ve_;
it = ve->begin();
}
ElementIterator operator++()
{
++it;
*this = static_cast<ElementIterator>(*it);
return *this;
}
bool end()
{
if (it != ve->end() )
return true;
return false;
}
vector<Element>* ve;
iter it;
};
Here is alternate implementation that does not rely on global variables or deriving from 'Element':
class Element
{
public:
Element() {};
size_t id;
};
class ElementIterator
{
std::vector<Element>& ve;
std::vector<Element>::iterator it;
public:
ElementIterator(std::vector<Element>& ve_) : ve(ve_)
{
it = ve.begin();
}
ElementIterator(const ElementIterator &rhs) : ve(rhs.ve)
{
it = rhs.it;
}
ElementIterator operator++()
{
++it;
return *this;
}
const Element *operator ->() const
{
assert(!this->end());
return &(*it);
}
bool end() const
{
if (it == ve.end())
return true;
return false;
}
};
Main:
int main()
{
std::vector<Element> E;
for (size_t i = 0; i < 5; i++)
{
Element e;
e.id = i;
E.push_back(e);
}
for (ElementIterator e(E); !e.end(); ++e)
std::cout << "Element " << e->id << std::endl;
}
NOTE: I changed the behaviour of 'ElementIterator::end' so that it returns 'true' when it is at the end which seemed more logical.
Try it here:
http://coliru.stacked-crooked.com/a/8bf8b1025d87a882
I suspect that the segfault occurs because the 'ElementIterator::operator++()' increments the internal 'it' attribute and then deferences it without verifying that its valid first. It could be implemented as follows:
ElementIterator::operator ++()
{
++it;
if (!this->end())
*this = static_cast<ElementIterator>(*it);
return *this;
}
I finally use this workaround, that it is not elegant but works. The main problem was that the iterator "it" was overwritten so I put it in the global scope.
class Element
{
public:
Element() {};
size_t id;
};
typedef vector<Element> VectorElement;
typedef vector<Element>::iterator iter;
iter it;
class ElementIterator: public Element
{
public:
ElementIterator(const Element& e) : Element(e) { }
ElementIterator(VectorElement& ve_)
{
ve = &ve_;
it = ve->begin();
}
ElementIterator operator++()
{
++it;
*this = static_cast<ElementIterator>(*it);
return *this;
}
bool end()
{
if (it != ve->end() )
return true;
return false;
}
vector<Element>* ve;
};
Related
I have a custom bitset class implementation in C++. I often iterate over the indexes of bits that are set in the bitset (i.e. for bitset '10011' I want to iterate over numbers 0, 3, 4.) This iteration can be implemented as follows:
struct Bitset {
uint64_t* data_;
size_t chunks_;
std::vector<int> Elements() const {
std::vector<int> ret;
for (size_t i=0;i<chunks_;i++){
uint64_t td = data_[i];
while (td) {
ret.push_back(i*BITS + __builtin_ctzll(td));
td &= ~-td;
}
}
return ret;
}
};
void Iterate(Bitset bitset) {
for (int b : bitset.Elements()) {
std::cout << "bit: " << b << std::endl;
}
}
The above implementation provides clean code for the iteration, but it involves an unnecessary heap allocation with the vector. The following version which essentially inlines the Elements() function is often faster:
void Iterate(Bitset bitset) {
int chunks = bitset.chunks_;
for (int i = 0; i < chunks; i++) {
uint64_t td = bitset.data_[i];
while (td) {
std::cout << "bit: " << i*BITS + __builtin_ctzll(td) << std::endl;
td &= ~-td;
}
}
}
What would be a good way to implement an abstraction for the iteration so that it would be as clean as the above version, but also with no performance cost.
Just iterate over your class. Provide your own implementation of an iterator class for your Bitset and provide begin() and end() methods. A simplest (untested!) implementation could look something like this:
#include <vector>
#include <cstdint>
#include <iostream>
struct Bitset {
uint64_t* data_;
size_t chunks_;
struct iterator {
uint64_t *pnt;
uint_fast8_t pos;
iterator(uint64_t *pnt, size_t pos) :
pnt(pnt), pos(pos) {}
bool operator !=(const iterator& o) {
return o.pnt != pnt || o.pos != pos;
}
void operator ++() {
pos++;
if (pos == 64) {
pnt++;
pos = 0;
}
}
bool operator *() {
return *pnt & (1 << pos);
}
};
iterator begin() { return iterator(data_, 0); }
iterator end() { return iterator(data_ + chunks_, 64); }
};
void Iterate(Bitset bitset) {
for (auto&& b : bitset) {
std::cout << "bit: " << b << std::endl;
}
}
I believe for your strange while (td) { ... i*BITS + __builtin_ctzll(td) ... loop that I don't understand that could be something along (untested!):
constexpr int BITS = 100000;
struct Bitset {
uint64_t* data_;
size_t chunks_;
struct iterator {
uint64_t *data_;
int i = 0;
uint64_t td = 0;
iterator(uint64_t *data_, int i, uint64_t td) :
data_(data_), i(i), td(td) {}
bool operator !=(const iterator& o) {
return o.data_ != data_ || o.i != i || o.td != td;
}
void operator ++() {
if (td == 0) {
td = *data_;
data_++;
} else {
td &= ~-td;
}
}
bool operator *() {
return i * BITS + __builtin_ctzll(td);
}
};
iterator begin() { return iterator(data_, 0, *data_); }
iterator end() { return iterator(data_ + chunks_, 0, 0); }
};
As KamilCuk suggested, I used an iterator to solve this problem. Now the implementation looks like:
struct Bitset {
uint64_t* data_;
size_t chunks_;
class BitsetIterator {
private:
const Bitset* const bitset_;
size_t pos_;
uint64_t tb_;
public:
BitsetIterator(const Bitset* const bitset, size_t pos, uint64_t tb) :
bitset_(bitset), pos_(pos), tb_(tb) { }
bool operator!=(const BitsetIterator& other) const {
return pos_ != other.pos_ || tb_ != other.tb_;
}
const BitsetIterator& operator++() {
tb_ &= ~-tb_;
while (tb_ == 0 && pos_ < bitset_->chunks_) {
pos_++;
if (pos_ < bitset_->chunks_) {
tb_ = bitset_->data_[pos_];
}
}
return *this;
}
int operator*() const {
return pos_*BITS + __builtin_ctzll(tb_);
}
};
BitsetIterator begin() const {
size_t pos = 0;
while (pos < chunks_ && data_[pos] == 0) {
pos++;
}
if (pos < chunks_) {
return BitsetIterator(this, pos, data_[pos]);
} else {
return BitsetIterator(this, pos, 0);
}
}
BitsetIterator end() const {
return BitsetIterator(this, chunks_, 0);
}
};
void Iterate(Bitset bitset) {
for (int b : bitset) {
std::cout << "bit: " << b << std::endl;
}
}
This avoids heap allocation and is much faster than the version that uses vector. I'm not sure if this provides exactly same performance as the version without any abstractions, but it should be very close.
I implemented my first custom iterator, which is supposed to return an std::pair at each iteration. The only problem, is it loops infinitely and I do not know how to provide stop condition. The code looks like so:
#include <iostream>
#include <vector>
#include <utility>
class Simple
{
private:
std::vector<std::size_t> indices;
std::vector<int> values;
public:
void insert(std::size_t index, int value)
{
indices.push_back(index);
values.push_back(value);
}
int at(std::size_t index)
{
return values[indices[index]];
}
class Iterator
{
private:
const std::vector<std::size_t>* indices;
const std::vector<int>* values;
std::size_t pos = 0;
public:
Iterator(const std::vector<std::size_t>* indices_, const std::vector<int>* values_, const std::size_t &pos_ = 0):
values(values_), indices(indices_), pos(pos_){ }
bool operator==(const Iterator& other){
return this == &other;
}
bool operator!=(const Iterator& other){
return !operator==(other);
}
Iterator operator++() {
pos++;
Iterator i = *this;
return i;
}
std::pair<std::size_t, int> operator*()
{
if (pos < (*values).size())
{
return std::make_pair((*indices)[pos], (*values)[pos]);
}
std::cout << "EMPTY PAIR" << std::endl; //loops infinitely and prints this message
return std::pair<std::size_t,int>{};
}
std::pair<std::size_t, int>* operator->()
{
if (pos < (*values).size())
{
std::pair<std::size_t,int> *p;
*p = std::make_pair((*indices)[pos], (*values)[pos]);
return p;
}
return nullptr;
}
};
Iterator begin() const
{
return Iterator(&indices, &values, 0);
}
Iterator end() const
{
return Iterator(&indices, &values, values.size());
}
};
int main() {
Simple s;
s.insert(10, 100);
std::cout << s.at(10) << std::endl;
int i = 0;
for (const std::pair<std::size_t, int> &p : s)
{
std::cout << p.first << " " << p.second << std::endl;
if (i > 3) break; // otherwise it will loop infinitely
i++;
}
return 0;
}
I would like to stress, that this example is simplified and for demonstration purposes, so you may not ask me, why I'm not using map or unorded_map (just because what I'm implementing looks like a map). The question is solely about iterator and I need understanding what may be wrong with iterator itself and how to provide stop condition.
How to implement an iterator of just on values of a map/unordered_map using boost::iterator_adaptor? I've tried following code but it does not work because of the line with comment.
Is there a solution to avoid the problem?
The question here is slightly different from map_values adapter example shown in boost code as here the value field in map is another container like list or vector and the requirement here is to iterate over all elements of those lists for every key of the map.
The deref of iterator is of type of value_type of those list/vector.The end of iterator is the end of list of last key
#include <vector>
#include <boost/unordered_map.hpp>
#include <cassert>
#include <iostream>
#include <boost/iterator/iterator_adaptor.hpp>
class DS {
public:
DS() : _map() {}
~DS() {
for (Map::iterator it = _map.begin(); it != _map.end(); ++it) {
delete (it->second);
}
}
void add(int key_, const std::vector< int > &value_)
{
IntList *ptr = new IntList(value_);
assert(ptr);
_map.insert(Map::value_type(key_, ptr));
}
private:
typedef std::vector< int > IntList;
typedef boost::unordered_map< int, IntList* > Map;
Map _map;
public:
class KeyIter : public boost::iterator_adaptor< KeyIter,
Map::const_iterator,
int,
boost::forward_traversal_tag,
int>
{
public:
KeyIter() : KeyIter::iterator_adaptor_() {}
private:
friend class DS;
friend class boost::iterator_core_access;
explicit KeyIter(Map::const_iterator it) : KeyIter::iterator_adaptor_(it) {}
explicit KeyIter(Map::iterator it) : KeyIter::iterator_adaptor_(it) {}
int dereference() const { return this->base()->first; }
};
class ValueIter : public boost::iterator_adaptor< ValueIter,
Map::const_iterator,
int,
boost::forward_traversal_tag,
int>
{
public:
ValueIter()
: ValueIter::iterator_adaptor_()
, _lIt()
{}
private:
friend class DS;
friend class boost::iterator_core_access;
explicit ValueIter(Map::const_iterator it)
: ValueIter::iterator_adaptor_(it)
, _lIt()
, _mIt(it)
{
IntList *pt = it->second; // <<-- issue here is I can't find if I've already reached the end of the map
if (pt) {
_lIt = it->second->begin();
}
}
int dereference() const { return *_lIt; }
void increment()
{
if (_lIt == _mIt->second->end()) {
++_mIt;
_lIt = _mIt->second->begin();
} else {
++_lIt;
}
}
IntList::iterator _lIt;
Map::const_iterator _mIt;
};
KeyIter beginKey() const { return KeyIter(_map.begin()); }
KeyIter endKey() const { return KeyIter(_map.end()); }
ValueIter beginValue() const { return ValueIter(_map.begin()); }
ValueIter endValue() const { return ValueIter(_map.end()); }
};
int main(int argc, char** argv)
{
DS ds;
std::vector< int > v1;
v1.push_back(10);
v1.push_back(30);
v1.push_back(50);
ds.add(90, v1);
std::vector< int > v2;
v2.push_back(20);
v2.push_back(40);
v2.push_back(60);
ds.add(120, v2);
std::cout << "------------ keys ---------------" << std::endl;
for (DS::KeyIter it = ds.beginKey(); it != ds.endKey(); ++it) {
std::cout << (*it) << std::endl;
}
std::cout << "------------ values ---------------" << std::endl;
// std::cout << (*(ds.beginValue())) << std::endl;
for (DS::ValueIter it = ds.beginValue(); it != ds.endValue(); ++it) {
std::cout << (*it) << std::endl;
}
return 0;
}
Implemented in c++11. You should be able to do the conversion to boost/c++03 fairly simply.
This iterator is FORWARD ONLY and it's quite fragile (see the comparison operator).
user discretion advised.
#include <iostream>
#include <vector>
#include <unordered_map>
typedef std::vector< int > IntList;
typedef std::unordered_map< int, IntList* > Map;
struct whole_map_const_iterator
{
using C1 = IntList;
using C2 = Map;
using I1 = C1::const_iterator;
using I2 = C2::const_iterator;
using value_type = I1::value_type;
using reference = I1::reference;
whole_map_const_iterator(I2 i2) : _i2(i2) {}
bool operator==(const whole_map_const_iterator& r) const {
if (_i2 != r._i2)
return false;
if (deferred_i1 && r.deferred_i1)
return true;
if (deferred_i1 != r.deferred_i1)
return false;
return _i1 == r._i1;
}
bool operator!=(const whole_map_const_iterator& r) const { return !(*this == r); }
reference operator*() const {
check_deferred();
return *_i1;
}
void check_deferred() const {
if (deferred_i1) {
_i1 = _i2->second->begin();
_i1limit = _i2->second->end();
deferred_i1 = false;
}
}
void go_next()
{
check_deferred();
if (++_i1 == _i1limit) {
++_i2;
deferred_i1 = true;
}
}
whole_map_const_iterator& operator++() {
go_next();
return *this;
}
whole_map_const_iterator operator++(int) {
auto result = *this;
go_next();
return result;
}
I2 _i2;
mutable I1 _i1 = {}, _i1limit = {};
mutable bool deferred_i1 = true;
};
IntList a { 1, 2, 3, 4, 5 };
IntList b { 6, 7, 8, 9, 10 };
Map m { { 1, &a }, { 2, &b } };
int main()
{
using namespace std;
auto from = whole_map_const_iterator(m.begin());
auto to = whole_map_const_iterator(m.end());
for ( ; from != to ; ++from) {
std::cout << *from << std::endl;
}
return 0;
}
example output:
6
7
8
9
10
1
2
3
4
5
For bonus points, answer this question:
Q: Why all that damn complication over the deferred flag?
I am implementing a container class that ensures uniqueness among the elements and restricts insertion and deletion to only the end only. Somewhat like a stack of unique elements or an ordered set with functions like push and pop. And it must also have a fixed maximum size.
template <class T, int max_size>
class FixedSizedUniqueStack
{
std::vector<T> m_vec;
std::unordered_set<T> m_uset;
public:
FixedSizedUniqueStack():m_vec(max_size),m_uset(){}
bool push(T x)
{
bool success = true;
if( m_uset.insert(x).second ) m_vec.push_back(x);
else success = false;
return success;
}
void pop()
{
if(m_vec.size() > 0)
{
m_uset.erase(m_vec.back());
m_vec.pop_back();
}
}
T back()
{
return m_vec.back();
}
};
#include <vector>
#include <unordered_set>
#include <initializer_list>
template <class T, int max_size>
class FixedSizedUniqueStack: public std::initializer_list<T>
{
protected:
std::vector<T> m_vec;
std::unordered_set<T> m_uset;
public:
FixedSizedUniqueStack():m_vec(),m_uset(){}
FixedSizedUniqueStack( const FixedSizedUniqueStack &x)
{
m_vec = x.m_vec;
m_uset = x.m_uset;
}
FixedSizedUniqueStack& operator= ( const FixedSizedUniqueStack &x)
{
if (this != &x)
{
m_vec = x.m_vec;
m_uset = x.m_uset;
}
return *this;
}
auto size() const -> decltype(m_vec.size())
{
return m_vec.size();
}
int push(const std::initializer_list<T>& il)
{
int errors = 0;
for (auto x: il)
{
if( push(x) )
{
errors++;
}
}
return errors;
}
int push(const T& x)
{
int error =0;
if(m_vec.size() < max_size)
{
if( x < start)
{
error = 1;
}
else if( x > stop)
{
error = 2;
}
else
{
if( m_uset.insert(x).second ) m_vec.push_back(x);
else error = 3;
}
}
else
{
error = 4;
}
return error;
}
void pop()
{
if(!m_vec.empty())
{
m_uset.erase(m_vec.back());
m_vec.pop_back();
}
}
T back()
{
return m_vec.back();
}
auto cbegin() const -> decltype(m_vec.cbegin())
{
return m_vec.cbegin();
}
auto cend() const -> decltype(m_vec.cend())
{
return m_vec.cend();
}
auto begin() const -> decltype(m_vec.begin())
{
return m_vec.begin();
}
auto end() const -> decltype(m_vec.end())
{
return m_vec.end();
}
auto empty() ->decltype(m_vec.empty())
{
return m_vec.empty();
}
I have the following code:
Where I stupidly iterate over a container (self-written).
If I compile this function with the cout it works and the program terminates correctly after the iteration! Note: The cout does not interfere with the self-made container!
If I then comment it out, the test starts to take infinitely long and does not terminate!
My eyeballs where almost dropping inside my head! What the hack is going on here???
My first guess was, that the compiler does something wrong? (but why?) or that maybe branch prediction plays me wrong and somehow increments the iterator twice? (which will result in no end() iterator).
I can really not explain what is going on?
int loops = 100;
int n = 0;
RangeTypes<Type>::ContainerType range(a);
INIT_TIMER
START_TIMER
for(int i=1; i<loops; i++) {
for(auto it=range.begin(); it != range.end(); ++it) {
n += *it;
//std::cout << i << std::endl;
}
}
STOP_TIMER("Range: ")
Actually the RangeType Container looks like this, and is pretty simple, it takes a range, and iterates from start to end!
template<typename Type>
struct RangeTypes {
using RangeType = std::pair<Type,Type> ;
class Iterator {
public:
Iterator(): m_rangePtr(nullptr), m_currVal(0) {};
~Iterator() {};
Iterator(RangeType & range, bool atEnd=false): m_rangePtr(&range) {
if(atEnd) {
m_currVal = m_rangePtr->second;
} else {
m_currVal = m_rangePtr->first;
}
};
/** pre-increment ++it
* Allow to iterate over the end of the sequence
*/
Iterator & operator++() {
if(m_rangePtr) {
++m_currVal;
}
return *this;
}
/** post-increment it++ */
Iterator operator++(int) {
Iterator it(*this);
operator++();
return it;
}
bool operator==(const Iterator &rhs) {
if(m_rangePtr) {
if(m_rangePtr == rhs.m_rangePtr ) { // if sequences are the same
return m_currVal == rhs.m_currVal;
}
}
return false;
}
// Return false if the same!
bool operator!=(const Iterator &rhs) {
return !(*this==rhs);
}
// get current value;
Type operator*() {
return m_currVal;
}
private:
RangeType * m_rangePtr;
Type m_currVal;
};
class ContainerType : public RangeType {
public:
typedef Iterator iterator;
ContainerType(RangeType & range ): RangeType(range) {
// if range wrong, set to no range!
if(this->first > this->second) {
this->first = 0;
this->second = 0;
}
}
iterator begin() {
return iterator(*this);
};
iterator end() {
return iterator(*this,true);
};
};
};
HERE IS THE MWE: Compiled Example
Thanks for any help!!