The Four Polymorphisms in C++ | Hacker News Reader

The Four Polymorphisms in C++

catonmat.net

53 pointspkrumins21 comments

That's a great summary!

I'm always disappointed if people only know one kind of polymorphism and apply that rigorously to all problems. Overviews like this might help people to make a conscious decision on which language construct would be a good fit for their actual design problem.

That's a fair point, but unfortunately in C++ the various types of polymorphism often do not play nicely together.

Look at any on-line forums discussing C++. If you're a bit more cynical, look at a list of tricky C++ interview questions instead. :-)

You'll find discussions about overloading vs. overriding for class member functions. You'll find discussions of which function gets called when you have both overloaded functions and function templates to consider. You'll find discussions of which sequences of implicit conversions are allowed, and via what mechanisms.

A very high proportion of the nasty edge cases for which C++ has become infamous are caused by trying to combine different forms of polymorphism. I'm all for knowing your tools and evaluating all the options, but I can see why in practice a lot of teams do stick almost entirely to (for example) using classes/OOP and avoiding templates.

I get the impression that the answer is no, but is there much use of templates in C++ to make generic classes like in Java/C#?

Intensive use of templates enables you to build great software (especially in terms of performance), but it requires a certain degree of seniority to be properly used.

All of STL.

It may be a good summary, but I completely disagree with author's (implied) assertion that all these should be lumped under one term. All these concepts are different in their mechanics, which is the reason we have different terminology to describe them. Just call overloading overloading. C++ does not need more ambiguity, it needs less.

I sympathize with your point, but I used to think of runtime and parametric polymorphism as two completely separate things. It wasn't until I heard them both described as a type of polymorphism that I realized the relationship between them.

I never implied that they should lumped under one term. On the contrary. I wrote out all the meanings the term can have to disambiguate it.

The blog post is titled "The Four Polymorphisms in C++."

Yes, the four polymorphisms you can find in C++. Am I missing something?

You're missing the fact that "polymorphism" has an established meaning in C++ which excludes two of your examples. You're using a definition of polymorphism that is independent of C++ and applying it to C++, which is certainly a valid exercise, but you haven't bothered to tell us what your definition is or explain why it applies to each example. If I assume that by "polymorphism" you mean when a single name can be used to refer to different things, then consider the following:

  void foo(float f);

  int main()
  {
    foo(1);
    return 0;
  }

What name is polymorphic here? I suppose you can say that foo is polymorphic if you attribute the coercion operation to foo, but that's not the way I think about it. foo is called after the coercion. The coercion is caused by the relation of the type of 1 with the type of the function parameter f.

In your blog post it would have made things much clearer if you had acknowledged the conventional usage of the term in the context of C++ and explained the different definition you were using.

I know it's just a blog post, but the section on "Parametric Polymorphism" says that the same code executes for any type. But in the example given, "a > b" will evaluate to entirely different operations depending on the type, and in fact only works because of the other types of polymorphism. Since this is such a common situation, I might not make it a first class citizen in the taxonomy.

In that example I only wanted to illustrate parametric polymorphism on type T (and ignore the overloading on operator > that happens).

Coercion is not a form of polymorphism. In all the other forms, dispatch depends on the type of a value. When there's coercion going on, be it explicit or implicit, the same code is being executed regardless of the value being coerced. The only code that one might say "changes" is any code that's executed to perform the cast. But that change in casting is basically ad-hoc polymorphism.

This paper [1] says coercion is a kind of ad-hoc polymorphism.

[1] "On Understanding Types, Data Abstraction, and Polymorphism" http://lucacardelli.name/Papers/OnUnderstanding.A4.pdf

Since we are talking about polymorphism, here are several blog posts I wrote on this topic:

http://www.oriontransfer.co.nz/blog/2009-05/what-is-abstract...

http://www.oriontransfer.co.nz/blog/2009-05/run-time-polymor...

http://www.oriontransfer.co.nz/blog/2009-05/compile-time-pol...

If you find it interesting that is great. I enjoyed this blog post.

This was also posted in Reddit, and there are some other comments over there, apparently:

http://www.reddit.com/r/programming/comments/cgckf/the_four_...

Where do macros fit into this taxonomy?

I would say "outside".

You want to offer a working definition of "polymorphism", then?

Macros certainly allow you to write code that's applied to differently typed arguments.

Indeed :

FOO.hxx:

TYPE FOO_TEMPLATE () {

  TYPE foo;
  /*...*/
  return foo;

}

Foo.h:

#DEFINE FOO_TEMPLATE foo_int

#DEFINE TYPE int

#include "Foo.hxx"

#DEFINE FOO_TEMPLATE foo_float

#DEFINE TYPE float

#include "Foo.hxx"

It's a kind of weird and ugly macro trick, bug it can be useful in C when one's need polymorphism and performance (and cannot use C++ at all, which is a far better solution).

[Edit:] I realize now it's not polymorphism like said in the article, because the caller had to tell explicitly which foo_xxx to call. But it did provide "polymorphic" code in Foo.hxx (so only one implementation, no repetition). That said, a cast is also explicit.

Reply on news.ycombinator.com