Something about reading portions of data from file using a C++ istream_iterator

Question

Target: There is text file (on HDD) containing integers divided with some kind of delimiter.

Example:

5245
234224
6534
1234

I need to read them into STL container.

int main(int argc, char * argv[]) {
  using namespace std;

  // 1. prepare the file stream
  string fileName;
  if (argc > 1)
    fileName = argv[1];
  else {
    cout << "Provide the filename to read from: ";
    cin >> fileName;
  }
  unique_ptr<ifstream, ifstream_deleter<ifstream>> ptrToStream(new ifstream(fileName, ios::out));
  if (!ptrToStream->good()) {
    cerr << "Error opening file " << fileName << endl;
    return -1;
  }

  // 2. value by value reading will be too slow on large data so buffer data
  typedef unsigned int values_type;
  const int BUFFER_SIZE(4); // 4 is for testing purposes. 16MB or larger in real life
  vector<values_type> numbersBuffer(BUFFER_SIZE);
  numbersBuffer.insert(numbersBuffer.begin(), istream_iterator<values_type>(*ptrToStream), istream_iterator<values_type>());
  // ...

The main drawback of this code is how can I handle the issue when file size is extremely large, so I cannot store all of it's contents in memory ? I also do not want to use push_back as it is non efficient in comparison to interval insert.

---

So, the question is: how can I read not more than BUFFER_SIZE elements from the file effectively using STL?

Answer 1

The approach to limit reading from input iterators is to create a wrapper which counts the number of elements processed so far and whose end iterator compares to this number. Doing this generically isn't quite trivial, doing it specifically for std::istream_iterator<T> shouldn't be too hard. That said, I think the easiest way to do it is this:

std::vector<T> buffer;
buffer.reserve(size);
std::istreambuf_iterator<T> it(in), end;
for (std::vector<T>::size_type count(0), capacity(size);
     it != end && count != capacity; ++it, ++count) {
    buffer.push_back(*it);
}

I realize that you don't want to push_back() because it is allegedly slow. However, compared to the I/O operation I doubt that you'll be able to measure the small overhead, especially with typical implementation of the I/O library.

Just to round things off with an example of a wrapped iterator: below is an example how a counting wrapper for std::istream_iterator<T> could look like. There are many different ways this could be done, this is just one of them.

#include <iostream>
#include <iterator>
#include <vector>
#include <sstream>

template <typename T>
class counted_istream_iterator:
    public std::iterator<std::input_iterator_tag, T, std::ptrdiff_t>
{
public:
    explicit counted_istream_iterator(std::istream& in): count_(), it_(in) {}
    explicit counted_istream_iterator(size_t count): count_(count), it_() {}

    T const& operator*() { return *this->it_; }
    T const* operator->() { return it_->it_.operator->(); }
    counted_istream_iterator& operator++() {
        ++this->count_; ++this->it_; return *this;
    }
    counted_istream_iterator operator++(int) {
        counted_istream_iterator rc(*this); ++*this; return rc;
    }

    bool operator== (counted_istream_iterator const& other) const {
        return this->count_ == other.count_ || this->it_ == other.it_;
    }
    bool operator!= (counted_istream_iterator const& other) const {
        return !(*this == other);
    }
private:
    std::ptrdiff_t           count_;
    std::istream_iterator<T> it_;
};

void read(int count)
{
    std::istringstream in("0 1 2 3 4 5 6 7 8 9");
    std::vector<int>   vec;
    vec.insert(vec.end(), counted_istream_iterator<int>(in),
               counted_istream_iterator<int>(count));
    std::cout << "size=" << vec.size() << "\n";
}

int main()
{
    read(4);
    read(100);
}

Answer 2

There is possible way to solve my problem:

// 2. value by value reading will be too slow on large data so buffer data
typedef unsigned int values_type;
const int BUFFER_SIZE(4);
vector<values_type> numbersBuffer;
numbersBuffer.reserve(BUFFER_SIZE);
istream_iterator<values_type> begin(*ptrToStream), end;
while (begin != end) {
  copy_n(begin, BUFFER_SIZE, numbersBuffer.begin());
  for_each(numbersBuffer.begin(), numbersBuffer.end(), [](values_type const &val){ std::cout << val << std::endl; });
  ++begin;
}

But it has one drawback. If input file contains the following:

8785
245245454545
7767

then 8785 will be read, but 245245454545 and 7767 will not, because 245245454545 cannot be converted to unsigned int. Error will be silent. :(