hi,

I've got a vector of class pointers I need to sort according to the value of a member of the class.

I thought of using C stdlib.h qsort but I'm not sure it's meant to be used on members of a C++ vector.

What should I do ?

I prefer this method:

The value of doing it this way is you'll be able to sort inside other containers, not just vector.

If 'someValue' wont change, than you should use multiset which sorts on insertion for great speed boosts.

int main() {
  multiset<YourWhateverPointer> items;
  
  items.push_back(YourWhateverPointer(new YourWhateverClass));
  items.push_back(YourWhateverPointer(new YourWhateverClass));
  items.push_back(YourWhateverPointer(new YourWhateverClass));

  // items is already sorted.
}

In my example you would make a LevelFeaturePointer class and implement the operator < on it. The answer is in my last post if you read it carefully.

There's no casting.

v[0].ptr->member

Pro-tip: Try to write code that never needs to leverage type casting.

What are you talking about? Why would you want to wrap a pointer in a special class and provide it with a comparison operator that only makes any sense in the context of sorting and only allows you to sort by one specific property? If you then wanted to sort by another property, would you need to make another pointer wrapper just for that? What does it have to do with being able to sort inside other containers? You're using an overload of the same sort function used in Arthur's example. In short, horrible, horrible idea. OP should totally go with Arthur's example. Why would you even want anything else?

Works exactly the same if you change the vector to a list or anything else.

features_for_level.push_back( LevelFeaturePointer( (LevelFeature *) rp ));

No! Stop casting things. This code will require no casting.

Because you're cluttering your program space with functions everywhere on the global namespace and the definition of sort is not connected to the class. The more you put into the class the better.

I would actually update the code to something like this:

And then you can make YourWhateverPointer a generic class with a template parameter. But I'll leave that as an exercise for the reader.

http://www.themusingsofalostprogrammer.com/2012/05/beauty-of-c-is-in-its-sorting.html

edit: missed that StasB has already addressed the point about global namespace.

static bool Test:comparebyheight(Test* t1, Test* t2) { return (t1->height<t2->height); }

Just a side note, the standard libraries will sometimes do random things if the comparator function doesn't address equal values. Handling equal values is called making the sort "stable". So I'd write axilmar's code as:

bool compare_foos(Foo *a, Foo *b) {
   return a->bar <= b->bar;
}

Just the simple addition of the = sign in the comparison.

Yeah, exactly, and that's exactly what the '=' does. Think about it.

EDIT: Also, in many cases is DOES matter, that's why it exists. Try implementing A* with a vector without a stable sort and nobody to tell you the difference. You'll likely go nuts before figuring it out.

Yes, entirely. Or you're just mad because I proved you wrong and now you're being a whiney little jerk. All I did was point out something that could be important for the OP to know. You thought you saw an opportunity to show us how smort you are and I rebuffed you. Now go get your mom to make you some cookies and milk and stop being an asshat.

AFAICT he's just started it in the last couple of months, even though he's been here 4 years. Maybe something else happened.

All you did was to say something completely nonsensical. All I did was to correct you. You rebuffed me? I'm sorry, I totally missed it since you did not actually address anything I said.

Your advice is factually incorrect and misleading. Your suggestion may actually lead to incorrect results. The comparison callback has nothing to do with the stability of the sorting. You just need to provide the appropriate algorithm with the approproate callback function which may or may not need an additional '='. I'm not mad at you or anyone. I'm sorry if you need three spoons of sugar to go with the criticism of your nonsense. I couldn't care less what people here think about my intelligence. I will keep calling you on your bullsh*t and you feel free to call me on mine. Unlike you, I don't hesitate to admit that I'm wrong, apologize and thank whoever corrects me.

[EDIT]

Oh. Excuse me. Did I say "may"? Your suggestion in fact DOES give incorrect results. Run this and see for yourself: