Skip to main content

C++ templates are turing complete

Here you will find a short C++ program that takes more than 24 hrs to compile on a dedicated dual processor, 1GB memory machine!! Here we go...

template<int Depth, int A, typename B>
struct K17 {
static const int x =
K17 <Depth+1, 0, K17<Depth,A,B> >::x
+ K17 <Depth+1, 1, K17<Depth,A,B> >::x
+ K17 <Depth+1, 2, K17<Depth,A,B> >::x
+ K17 <Depth+1, 3, K17<Depth,A,B> >::x
+ K17 <Depth+1, 4, K17<Depth,A,B> >::x;
};
template <int A, typename B>
struct K17 <16,A,B> { static const int x = 1;
};
static const int z = K17 <0,0,int>::x;
int main(void) { }

Source: C++ Templates are Turing Complete by Todd L. Veldhuizen

This program is taken from the above paper which takes unreasonably long to compile. I belive, a simple dynamic programming solution will reduce the exponential time required by this program to compile to polynomial time. I also believe, it might be quite difficult to apply dynamic programming solution, in general, to all C++ programs of this nature. You need to have a good understanding of template meta-programming to make sense of this program. One good article by the same author is here:

This post is motivated by a anonymous comment I received on an earlier post. I am quoting him here:

"It is ___provably___ impossible to write a correct C++ parser which will complete compilation with either success or failure because the C++ template system is Turing complete. This means that code generation is based on a turing complete program. Code generation in C++ isn't based on a program description, but an actual turing complete program. As such, it is subject to the halting problem. Therefore, it is unknowable whether a compilation will complete, and unknowable if you are looking at a valid C++ program."

Comments

Crunch said…
there is an error in the code :

K17 >:x

should be :

K17 >::x

otherwise, it won't compile
cialis generic said…
Thanks mate... just dropped by. Will look for BIKE STN when we get to Seattle. Still in Buenos Airies.
David Stone said…
The only reason it takes so long to compile is because it adds in unused template arguments A and B. Once those are deleted, you give addition of the same thing 5 times, so you just replace that with multiplication.

There is also a compile-time error caused by integer overflow in a constant expression. The type of x and z need to be able to hold a 38-bit integer. Something like std::int64_t should be good enough.

Once you make the changes to get rid of useless code and allow it to compile, it finishes in about .2 seconds on my machine.

Popular posts from this blog

Multi-dimensional arrays in C++11

What new can be said about multi-dimensional arrays in C++? As it turns out, quite a bit! With the advent of C++11, we get new standard library class std::array. We also get new language features, such as template aliases and variadic templates. So I'll talk about interesting ways in which they come together.

It all started with a simple question of how to define a multi-dimensional std::array. It is a great example of deceptively simple things. Are the following the two arrays identical except that one is native and the other one is std::array?

int native[3][4];
std::array<std::array<int, 3>, 4> arr;

No! They are not. In fact, arr is more like an int[4][3]. Note the difference in the array subscripts. The native array is an array of 3 elements where every element is itself an array of 4 integers. 3 rows and 4 columns. If you want a std::array with the same layout, what you really need is:

std::array<std::array<int, 4>, 3> arr;

That's quite annoying for two r…

Understanding Fold Expressions

C++17 has an interesting new feature called fold expressions. Fold expressions offer a compact syntax to apply a binary operation to the elements of a parameter pack. Here’s an example. template <typename... Args> auto addall(Args... args) { return (... + args); } addall(1,2,3,4,5); // returns 15. This particular example is a unary left fold. It's equivalent to ((((1+2)+3)+4)+5). It reduces/folds the parameter pack of integers into a single integer by applying the binary operator successively. It's unary because it does not explicitly specify an init (a.k.a. identity) argument. So, let add it. template <typename... Args> auto addall(Args... args) { return (0 + ... + args); } addall(1,2,3,4,5); // returns 15. This version of addall is a binary left fold. The init argument is 0 and it's redundant (in this case). That's because this fold expression is equivalent to (((((0+1)+2)+3)+4)+5). Explicit identity elements will come in handy a little la…

Folding Monadic Functions

In the previous two blog posts (Understanding Fold Expressions and Folding Functions) we looked at the basic usage of C++17 fold expressions and how simple functions can be folded to create a composite one. We’ll continue our stride and see how "embellished" functions may be composed in fold expressions.

First, let me define what I mean by embellished functions. Instead of just returning a simple value, these functions are going to return a generic container of the desired value. The choice of container is very broad but not arbitrary. There are some constraints on the container and once you select a generic container, all functions must return values of the same container. Let's begin with std::vector.
// Hide the allocator template argument of std::vector. // It causes problems and is irrelevant here. template <class T> struct Vector : std::vector<T> {}; struct Continent { }; struct Country { }; struct State { }; struct City { }; auto get_countries…