Memory issue when joining ranges in a for-loop

[mathbagu]

We are facing a memory issue (process is killed by the OOMKiller) when we join ranges inside a for loop. The idea here, is to flatten a list of ranges before returning the new range.

#include <vector>
#include <boost/range/any_range.hpp>
#include <boost/range/join.hpp>

struct A {
};

using Range = typename boost::any_range<A, boost::forward_traversal_tag, A&, std::ptrdiff_t>;

struct B {
    std::vector<A> v{10, A()};
};

struct C {
    std::vector<B> v{30, B()};

    Range getAll() const {
        Range all;
        for (const auto& b : v) {
            all = boost::range::join(all, b.v);
        }
        return all;
    }
}

int main(int argc, char** argv) {
    C c;
    for (const A& a : c.getAll()) {
        // Do something with a
    }

    return 0;
}

[mathbagu]

Finally I solve my issue creating a FlattenIterator based on this implementation:

template <typename Iterator>
class FlatteningIterator :
    public boost::iterator_adaptor<
        FlatteningIterator<Iterator>,
        Iterator,
        typename Iterator::value_type::iterator::value_type,
        boost::forward_traversal_tag,
        typename Iterator::value_type::iterator::value_type> {
public:
    using OuterIterator = Iterator;
    using InnerIterator = typename Iterator::value_type::iterator;
    using Value = typename Iterator::value_type::iterator::value_type;

    using base = boost::iterator_adaptor<FlatteningIterator<Iterator>, Iterator, Value, boost::forward_traversal_tag, Value>;

public:
    FlatteningIterator() = default;

    FlatteningIterator(Iterator it) :
        base(it),
        m_outer_end(it),
        m_inner_begin(),
        m_inner_end() {
    }

    FlatteningIterator(Iterator begin, Iterator end) :
        base(begin),
        m_outer_end(end) {

        if (begin != end) {
            m_inner_begin = (*begin).begin(),
            m_inner_end = (*begin).end();
            skipEmptyRanges();
        }
    }

private:
    void increment() {
        if (this->base_reference() == m_outer_end) {
            return;
        }

        ++m_inner_begin;
        skipEmptyRanges();
    }

    Value dereference() const {
        return *m_inner_begin;
    }

    void skipEmptyRanges() {
        while ((this->base_reference() != m_outer_end) && (m_inner_begin == m_inner_end)) {
            ++this->base_reference();
            if (this->base_reference() != m_outer_end) {
                m_inner_begin = (*this->base_reference()).begin();
                m_inner_end = (*this->base_reference()).end();
            }
        }
    }

    friend class boost::iterator_core_access;

private:
    OuterIterator m_outer_end;
    InnerIterator m_inner_begin;
    InnerIterator m_inner_end;
};

struct FlattenForwarder {};

namespace {

const FlattenForwarder flattened;

}  // namespace

template <typename SinglePassRange>
using FlattenRange = boost::iterator_range<FlatteningIterator<typename boost::range_iterator<SinglePassRange>::type>>;

template <typename SinglePassRange>
FlattenRange<SinglePassRange> operator|(SinglePassRange& range, FlattenForwarder) {
    using Iterator = typename boost::range_iterator<SinglePassRange>::type;

    return boost::make_iterator_range(FlatteningIterator<Iterator>(boost::begin(range), boost::end(range)), FlatteningIterator<Iterator>(boost::end(range)));
}

template <typename SinglePassRange>
FlattenRange<SinglePassRange> operator|(const SinglePassRange& range, FlattenForwarder) {
    using Iterator = typename boost::range_iterator<const SinglePassRange>::type;

    return boost::make_iterator_range(FlatteningIterator<Iterator>(boost::begin(range), boost::end(range)), FlatteningIterator<Iterator>(boost::end(range)));
}

It would be nice if you can integrate this feature