Is there a nice way to iterate over at most N elements in a container using a range-based for loop and/or algorithms from the standard library (that's the whole point, I know I can just use the "old" for loop with a condition).
Basically, I'm looking for something that corresponds to this Python code:
for i in arr[:N]:
print(i)
Since C++20 you can add the range adaptor std::views::take from the Ranges library to your range-based for loop. This way you can implement a similar solution to the one in PiotrNycz's answer, but without using Boost:
int main() {
std::vector<int> v {1, 2, 3, 4, 5, 6, 7, 8, 9};
const int N = 4;
for (int i : v | std::views::take(N))
std::cout << i << std::endl;
return 0;
}
The nice thing about this solution is that N may be larger than the size of the vector. This means, for the example above, it is safe to use N = 13; the complete vector will then be printed.
As I personally would use either this or this answer (+1 for both), just for increasing your knowledge - there are boost adapters you can use. For your case - the sliced seems the most appropriate:
#include <boost/range/adaptor/sliced.hpp>
#include <vector>
#include <iostream>
int main(int argc, const char* argv[])
{
std::vector<int> input={1,2,3,4,5,6,7,8,9};
const int N = 4;
using boost::adaptors::sliced;
for (auto&& e: input | sliced(0, N))
std::cout << e << std::endl;
}
One important note: N is required by sliced to be not greater than distance(range) - so safer(and slower) version is as follows:
for (auto&& e: input | sliced(0, std::min(N, input.size())))
So - once again - I would use simpler, old C/C++ approach (this you wanted to avoid in your question ;)
boost::adaptors::filtered. But for "index view" - probably not (I am not sure)... –
Meggy Here is the cheapest save solution that works for all forward iterators I could come up with:
auto begin = std::begin(range);
auto end = std::end(range);
if (std::distance(begin, end) > N)
end = std::next(begin,N);
This might run through the range almost twice, but I see no other way to get the length of the range.
std::advance(begin, N) instead of std::next. The former may take advantage of RandomAccessInterator if it is available, the latter will not. –
Faunus RandomAccessInterator? You could even easily implement std::next using std::advance. –
Candis § 24.4.4.6 for std::next() "Effects: Equivalent to advance(x, n); return x;" I'm not sure that it is a requirement to take advantage of RandomAccessIterator, but it would be a shame if they didn't. –
Faunus std::next because I want the n-th successor of a given iterator, which is exactly what std::next is there for. –
Candis std::cin). –
Furtado You can use the good old break to manually break a loop when needed. It works even with range based loop.
#include <vector>
#include <iostream>
int main() {
std::vector<int> a{2, 3, 4, 5, 6};
int cnt = 0;
int n = 3;
for (int x: a) {
if (cnt++ >= n) break;
std::cout << x << std::endl;
}
}
std::for_each. That'll probably involve futzing with iterators. –
Mariannmarianna C++ is great since you can code your own hideous solutions and hide them under an abstraction layer
#include <vector>
#include <iostream>
//~-~-~-~-~-~-~- abstraction begins here ~-~-~-~-~-//
struct range {
range(std::vector<int>& cnt) : m_container(cnt),
m_end(cnt.end()) {}
range& till(int N) {
if (N >= m_container.size())
m_end = m_container.end();
else
m_end = m_container.begin() + N;
return *this;
}
std::vector<int>& m_container;
std::vector<int>::iterator m_end;
std::vector<int>::iterator begin() {
return m_container.begin();
}
std::vector<int>::iterator end() {
return m_end;
}
};
//~-~-~-~-~-~-~- abstraction ends here ~-~-~-~-~-//
int main() {
std::vector<int> a{11, 22, 33, 44, 55};
int n = 4;
range subRange(a);
for ( int i : subRange.till(n) ) {
std::cout << i << std::endl; // prints 11, then 22, then 33, then 44
}
}
The above code obviously lacks some error checking and other adjustments, but I wanted to just express the idea clearly.
This works since range-based for loops produce code similar to the following
{
auto && __range = range_expression ;
for (auto __begin = begin_expr,
__end = end_expr;
__begin != __end; ++__begin) {
range_declaration = *__begin;
loop_statement
}
}
cfr. begin_expr and end_expr
range(a) is a temporary, till() returns a reference to it and that reference is bound in the range-based for loop (auto && __range = range_expression). The intermediate temporaries in the expression are then deleted before the loop is executed - you end up with a dangling reference. –
Hampshire If your container doesn't have (or might not have) RandomAccessIterator, there is still a way to skin this cat:
int cnt = 0;
for(auto it=container.begin(); it != container.end() && cnt < N ; ++it,++cnt) {
//
}
At least for me, it is very readable :-). And it has O(N) complexity regardless of container type.
std::for_each. That'll probably involve futzing with iterators. –
Mariannmarianna This is an index iterator. Mostly boilerplate, leaving it out, because I'm lazy.
template<class T>
struct indexT
//: std::iterator< /* ... */ > // or do your own typedefs, or don't bother
{
T t = {};
indexT()=default;
indexT(T tin):t(tin){}
indexT& operator++(){ ++t; return *this; }
indexT operator++(int){ auto tmp = *this; ++t; return tmp; }
T operator*()const{return t;}
bool operator==( indexT const& o )const{ return t==o.t; }
bool operator!=( indexT const& o )const{ return t!=o.t; }
// etc if you want full functionality.
// The above is enough for a `for(:)` range-loop
};
it wraps a scalar type T, and on * returns a copy. It also works on iterators, amusingly, which is useful here, as it lets us inherit effectively from a pointer:
template<class ItA, class ItB>
struct indexing_iterator:indexT<ItA> {
ItB b;
// TODO: add the typedefs required for an iterator here
// that are going to be different than indexT<ItA>, like value_type
// and reference etc. (for simple use, not needed)
indexing_iterator(ItA a, ItB bin):ItA(a), b(bin) {}
indexT<ItA>& a() { return *this; }
indexT<ItA> const& a() const { return *this; }
decltype(auto) operator*() {
return b[**a()];
}
decltype(auto) operator->() {
return std::addressof(b[**a()]);
}
};
The indexing iterator wraps two iterators, the second of which must be random-access. It uses the first iterator to get an index, which it uses to look up a value from the second.
Next, we have is a range type. A SFINAE-improved one can be found many places. It makes iterating over a range of iterators in a for(:) loop easy:
template<class Iterator>
struct range {
Iterator b = {};
Iterator e = {};
Iterator begin() { return b; }
Iterator end() { return e; }
range(Iterator s, Iterator f):b(s),e(f) {}
range(Iterator s, size_t n):b(s), e(s+n) {}
range()=default;
decltype(auto) operator[](size_t N) { return b[N]; }
decltype(auto) operator[] (size_t N) const { return b[N]; }\
decltype(auto) front() { return *b; }
decltype(auto) back() { return *std::prev(e); }
bool empty() const { return begin()==end(); }
size_t size() const { return end()-begin(); }
};
Here are helpers to make working with ranges of indexT easy:
template<class T>
using indexT_range = range<indexT<T>>;
using index = indexT<size_t>;
using index_range = range<index>;
template<class C>
size_t size(C&&c){return c.size();}
template<class T, std::size_t N>
size_t size(T(&)[N]){return N;}
index_range indexes( size_t start, size_t finish ) {
return {index{start},index{finish}};
}
template<class C>
index_range indexes( C&& c ) {
return make_indexes( 0, size(c) );
}
index_range intersect( index_range lhs, index_range rhs ) {
if (lhs.b.t > rhs.e.t || rhs.b.t > lhs.b.t) return {};
return {index{(std::max)(lhs.b.t, rhs.b.t)}, index{(std::min)(lhs.e.t, rhs.e.t)}};
}
ok, almost there.
index_filter_it takes a range of indexes and a random access iterator, and makes a range of indexed iterators into that random access iterator's data:
template<class R, class It>
auto index_filter_it( R&& r, It it ) {
using std::begin; using std::end;
using ItA = decltype( begin(r) );
using R = range<indexing_iterator<ItA, It>>;
return R{{begin(r),it}, {end(r),it}};
}
index_filter takes an index_range and a random access container, intersects their indexes, then calls index_filter_it:
template<class C>
auto index_filter( index_range r, C& c ) {
r = intersect( r, indexes(c) );
using std::begin;
return index_filter_it( r, begin(c) );
}
and now we have:
for (auto&& i : index_filter( indexes(0,6), arr )) {
}
and viola, we have a large musical instrument.
Fancier filters are possible.
size_t filter[] = {1,3,0,18,22,2,4};
using std::begin;
for (auto&& i : index_filter_it( filter, begin(arr) ) )
will visit 1, 3, 0, 18, 22, 2, 4 in arr. It does not, however, bounds-check, unless arr.begin()[] bounds-checks.
There are probably errors in the above code, and you should probably just use boost.
If you implement - and [] on indexT, you can even daisy chain these ranges.
Since C++20 you can add the range adaptor std::views::take from the Ranges library to your range-based for loop. This way you can implement a similar solution to the one in PiotrNycz's answer, but without using Boost:
int main() {
std::vector<int> v {1, 2, 3, 4, 5, 6, 7, 8, 9};
const int N = 4;
for (int i : v | std::views::take(N))
std::cout << i << std::endl;
return 0;
}
The nice thing about this solution is that N may be larger than the size of the vector. This means, for the example above, it is safe to use N = 13; the complete vector will then be printed.
This solution doesn't go past end(), has O(N) complexity for std::list (doesn't use std::distance) works with std::for_each, and only requires ForwardIterator:
std::vector<int> vect = {1,2,3,4,5,6,7,8};
auto stop_iter = vect.begin();
const size_t stop_count = 5;
if(stop_count <= vect.size())
{
std::advance(stop_iter, n)
}
else
{
stop_iter = vect.end();
}
std::for_each(vect.vegin(), stop_iter, [](auto val){ /* do stuff */ });
The only thing it doesn't do is work with InputIterator such as std::istream_iterator - you'll have to use external counter for that.
InputIterator either, since it requires RandomAccessRange. –
Bateau First we write an iterator which stops at a given index:
template<class I>
class at_most_iterator
: public boost::iterator_facade<at_most_iterator<I>,
typename I::value_type,
boost::forward_traversal_tag>
{
private:
I it_;
int index_;
public:
at_most_iterator(I it, int index) : it_(it), index_(index) {}
at_most_iterator() {}
private:
friend class boost::iterator_core_access;
void increment()
{
++it_;
++index_;
}
bool equal(at_most_iterator const& other) const
{
return this->index_ == other.index_ || this->it_ == other.it_;
}
typename std::iterator_traits<I>::reference dereference() const
{
return *it_;
}
};
We can now write an algorithme for making a rage of this iterator from a given range:
template<class X>
boost::iterator_range<
at_most_iterator<typename X::iterator>>
at_most(int i, X& xs)
{
typedef typename X::iterator iterator;
return std::make_pair(
at_most_iterator<iterator>(xs.begin(), 0),
at_most_iterator<iterator>(xs.end(), i)
);
}
Usage:
int main(int argc, char** argv)
{
std::vector<int> xs = {1, 2, 3, 4, 5, 6, 7, 8, 9};
for(int x : at_most(5, xs))
std::cout << x << "\n";
return 0;
}
equal method is bothering me. I understand why you use a ||, however I can think of issues with cyclic iterators (for example). I would propose to only refer to index_ there, and not bothering with the iterators at all. Also (nit), don't use int for index_, prefer something like size_t as int could be as small as 16 bits for example. –
Furtado || this->it_ == other.it_ seems to be the wrong solution as it breaks a cycling iterator (and yes, the iterator pair concept in C++ makes things harder, a single object would be all too easy). I wonder if sliced in Boost adaptors handles cycling iterators. –
Furtado Since Ranges became part of C++ (in C++20), the way to achieve this is with a view adapter:
container | std::views::take(N)
As the view is a range in its own right, we can use that in a for loop, or pass it to algorithms.
For example, the Python loop in the question is expressed as
for (auto const& i: arr | std::views::take(N)) {
std::cout << i << '\n';
}
Or, without the explicit loop:
using value_type = typename decltype(arr)::value_type;
auto print = std::ostream_iterator<value_type>{std::cout, '\n'};
std::ranges::copy(arr | std::views::take(N), print);
My favourite answer to this (now that we have C++20), is like this:
for (const auto& foo : std::span{bar.begin(), num_iterations}) {
// stuff
}
Using a span like this, we can iterate over any subsequence easily, using a range-based for:
for (const auto& foo : std::span{bar.begin() + subsequence_start, num_iterations}) {
// stuff
}
std::span anyway. The real answer is for (const auto& foo: container | std::views::take(N)) –
Underachieve © 2022 - 2024 — McMap. All rights reserved.
c.size() < N ? c.size() : N– Bellinifor (auto&& e: input | ranges::view::take(N)) { /* stuff */ }. – Amygdalate