Can a concept require a constexpr value or function?

Question

So, I'd like to do some advanced type-level hackery for which I would really like to be able to write a concept that requires a type to have a constexpr int value associated with it, which I can use later in the same concept as the integer std::array template parameter.

It's possible to write

template<typename T>
concept bool HasCount = requires {
    typename T::count;
};

but this is not what I want; I would like T::count to be a static constexpr int. However, the code (not even including the needed constexpr)

template<typename T>
concept bool HasCount = requires {
    int T::count;
};

does not compile with "error: expected primary-expression before 'int'" on GCC 7.3.0.

Another failed attempt: it's possible to write this, which would require static int T::count():

template<typename T>
concept bool HasCount = requires {
    {T::count()} -> int;
};

but not this, which is what I want:

template<typename T>
concept bool HasCount = requires {
    {T::count()} -> constexpr int;
    {T::count() constexpr} -> int; // or this
    {constexpr T::count()} -> int; // or this (please forgive me for fuzzing GCC instead of reading the manual, unlike perl C++ is not an empirical science)
};

So, I'd like to know if it's in any way possible to require a concept expression to be constexpr-qualified, or if not if there's a reason why it can't be possible, or if it's just not included in the specification.

Don't know if it is possible to write it directly with concept, but traits can be done [detecting-constexpr-with-sfinae](https://stackoverflow.com/questions/15232758/detecting-constexpr-with-sfinae), and then use it in concept. — Jarod42, May 16 '18 at 07:54

score 6 · Accepted Answer · answered May 16 '18 at 13:51

6

In theory, this is possible by requiring T::count to be a valid expression, and requiring that it's valid to use T::count in a context that requires a constant expression. For example:

#include <type_traits>
#include <utility>

template<int> using helper = void;

template<typename T>
concept bool HasCount = requires {
    // T::count must be a valid expression
    T::count;
    // T::count must have type int const
    requires std::is_same_v<int const, decltype(T::count)>;
    // T::count must be usable in a context that requires a constant expression
    typename ::helper<T::count>;
};

struct S1 {
    static constexpr int count = 42;
};
static_assert(HasCount<S1>);

struct S2 {
    int count = 42;
};
static_assert(!HasCount<S2>);

struct S3 {
    static constexpr double count = 3.14;
};
static_assert(!HasCount<S3>);

but in practice, the implementation of concepts in GCC rejects this program:

<source>:20:16: error: invalid use of non-static data member 'S2::count'
 static_assert(!HasCount<S2>);
                ^~~~~~~~~~~~
<source>:18:17: note: declared here
     int count = 42;
                 ^~
<source>:20:16: error: invalid use of non-static data member 'S2::count'
 static_assert(!HasCount<S2>);
                ^~~~~~~~~~~~
<source>:18:17: note: declared here
     int count = 42;
                 ^~

(which fact I believe to be a bug.)

answered May 16 '18 at 13:51

Casey

41,449
7
95
125

1

It's not every day you stumble across [two compiler bugs](https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85808) writing a dozen lines of code for a Stackoverflow answer. – Casey May 16 '18 at 14:02
The compile error looks like it's because you [forgot `static` inside of `S2`](https://godbolt.org/g/3XiRZ5). One thing I really don't get is why the `::` preceding `helper` is needed. If I make `helper` (without `::`) the first line of `HasCount`, I don't get any compile error, but if I make it the second line (moving it after `T::count;`) I get the error `T::helper is not a type`, and if I make it the last line (simply removing the `::` from your code) I get `std::helper has not been declared`. Is namespace somehow carrying over from one line to the next?... Weird. – jcarpenter2 May 17 '18 at 01:39
Also, thanks but I feel inclined to mention that since writing that post I have arrived at a different and slightly lower-tech method of solving my actual problem: [recursive variadic templates which are specialized not to recurse at the last template argument](http://coliru.stacked-crooked.com/a/24e83cfd27bec09b) in order to easily define type-safe structs representing the result sets of various database queries. – jcarpenter2 May 17 '18 at 02:22
@jcarpenter2 I didn't forget `static` in `S2`, it's intentionally omitted to test that the concept correctly does not admit such a type. And yes, the name lookup bug depends on the order of requirements. – Casey May 17 '18 at 23:11

Can a concept require a constexpr value or function?

1 Answers1