Iterators: Signs of Weakness in Object-Oriented Languages | Hacker News Reader

Iterators: Signs of Weakness in Object-Oriented Languages

home.pipeline.com

29 pointspcof10 comments

While C++0x will have proper lamdbas and closures (I've said this how many times in the past few weeks?), I think it's worth noting the "proper" way to emulate a closure in C++ is with a function object. Example:

  struct Summation {
    int& sum;
    Summation(int& s): sum(s) {}
    void operator()(int n)
    {
      sum += n;
    }
  };

  vector<int> numbers;
  int sum = 0;
  // give numbers some interesting values
  for_each(numbers.begin(), numbers.end(), Summation(sum));
  cout << "sum of all numbers: " << sum << endl;

C++0x lambdas will generate code very similar to this. The need to explicitly define a named struct/class and function object is an obvious weakness of this technique.

Although I do agree with his overall point, which is that the need to use iterators for all sequences is a code smell. It forces the programmer to work at a lower level when they often just want to say "Perform this operation on my data."

Interesting. Your comment about C++0x closures inspired me to look it up on Wikipedia.

The verbosity of the fragment above is greatly reduced. You had to write 'sum' or 's' lots of extra times ('Summation(sum)', 'sum(s)', 'int& sum', and 'int& s'), and you won't have to do that anymore -- according to wikipedia your function will change in C++0x to:

    for_each( numbers.begin(), numbers.end(), [&sum](int n) { sum += n; } );

Which is semantically much less to think about. It also still has the feature / annoyance (?) of having to specifically identify which variables to capture from the environment ('sum', as a reference; it seems you can capture variables by value too).

Fortunately, it seems the C++ people even thought of that annoyance, and allow you to write:

    for_each( numbers.begin(), numbers.end(), [&](int n) { sum += n; } );

capturing references to all variables in the enclosing scope. Neat!

Of course, C++ already has std::accumulate, which is basically the same thing as a general fold or a summation, depending on which version you use. So, there isn't really much of a reason to write sum manually at all. :p

http://www.sgi.com/tech/stl/accumulate.html

Indeed. I actually use this technique in my own code, and I feel like I'm laying down pipe all over the place to make sure the captured variable gets routed to the function object correctly. Once you do it a few times, it is obviously boiler-plate code and obviously a hindrance to momentum.

Conveniently ignored Smalltalk, which at the time, would have been seeing a lot more use than it does now.

I missed the date of the article at first, it wasn't until I looked at the C code and had to double take it, because it wasn't ANSI C, and my initial reaction was "hang on wait a minute" then checked the date on the article.... yeah.

It's worth reading the Abstract Heresies' entry that turned my attention to this paper: http://funcall.blogspot.com/2010/05/c-vs-lisp.html (After giving it a thought I posted this entry to HN too, as I probably should have done in the first place... :))

The date is interesting, 1992 - it would still take years for Java to implement it...wrong.

It seems odd that he complains about lack of local (modifiable) state while at the same time talking up the advantages of pure functional programming. It's painful to have to explicitly capture state instead of depending on a closure to do it, but you can do it, and somewhat nicely using objects in C++.

Also, is it not right that iterators were introduced in C++ largely to allow generic functions across arrays and other data structures?

Finally, do iterators necessarily imply statefulness? Doesn't seem like it. Taking + 1 as "successor": double sum(const double* start, const double* end) { if (start == end) { return 0.0; } else { return *start + sum(start + 1, end); } }

PDF version - http://citeseerx.ist.psu.edu/viewdoc/download;jsessionid=D07...

Something about the writing style in the Introduction section reminds me of this classic Dilbert strip: http://tomayko.com/writings/that-dilbert-cartoon

It seems these are lessons we must perpetually relearn.

Reply on news.ycombinator.com