Add `static_assert`s for template parameters of `basic_fixed_string`

[unterumarmung]

Example from implementation of basic_string_view by Microsoft: https://github.com/microsoft/STL/blob/main/stl/inc/xstring#L1224-L1230

[DBJDBJ]

static_assert() inside a template controls only the instatiation of the type definition of the template. Not the mechanism of template type definition.

template <typename T>
struct not_a_vector {
  static_assert(std::is_trivially_copyable_v<T>,
                "\n\nnot_a_vector<>T -- can handle only trivially "
                "copiable types!\n\n");
} ;

// no warning even if -Wall is used
   using should_not_compile_but_it_does
               = not_a_vector<std::string>  ;

Consider instead controling the type definition.

template <typename T,
std::enable_if_t< std::is_trivially_copyable_v<T>, int > = 0
>
class better_not_a_vector {
} ;

using should_not_compile_and_it_does_not =
               better_not_a_vector<std::string>;

using should_compile_and_it_does =
               better_not_a_vector<bool>;

better_not_a_vector<T> can not be promoted to type definition if the author's requirement is not satisfied.

Gobolt here

[unterumarmung]

I don't consider this as an issue. It will be a compile-time error anyway, when the type alias is used. I don't see any harm in postponing the compile-time error.

[DBJDBJ]

if not_a_vector is in the library not owned by you (it might be owned by MS STL for example) that will go unnoticed, and that might be a problem.

For example:

// this is inside dudkinlib.h
#include <not_a_vector> // not owned by dudkin ltd
// no warning even if -Wall is used
namespace dudkin {
   using  vector_of_strings // should_not_compile_but_it_does
               = not_a_vector<std::string>  ;
}

It might happen this will go unnoticed for some months, and then it will be noticed by some important customer.

// DUDKINLIB paying customer code, few months later
#include <dudkinlib.h>

dudkin::vector_of_strings    customer_ ; // does not compile

What is the solution? not_a_vector vendor has this philosophy called "I don't see any harm in postponing the compile-time error", and consequently does not (or can not) change not_a_vector. Customer has paid for dudkinlib and demands a solution from dudkin ltd.

[DBJDBJ]

Another example from

https://github.com/unterumarmung/hyper_log_log/blob/master/hll/hyper_log_log.hxx

template<typename T, std::size_t k>
class hyper_log_log
{
public:
    static_assert(k >= 4 && k <= 30, "k must be in a range [4; 30]");
} ; 

using timebomb_type = hyper_log_log<bool, 1> ;

// few months (or years) after:
// timebomb_type boom ;

Just a forward declared type constrained template is enough to stop wrong type definitions.

// forward declaration of a type constrained template
template <typename T, std::size_t k,
          std::enable_if_t<k >= 4 && k <= 30, int> = 0>
class better_hyper_log_log ;

User (or you) can not introduce types that will not compile few months or years down the line

// does not compile
// using impossible_type = better_hyper_log_log<bool, 1> ;

using ok_type = better_hyper_log_log<bool, 13> ;

This peculiarity is what one might describe as "C++ foot gun with a timer" ...

[DBJDBJ]

Actually, this is the correct solution

template< typename T, std::size_t k>
#if __cplusplus >= 202002L
    // C++20 (and later) code
requires // requires-clause (ad-hoc constraint)
(k >= 4 && k <= 30) 
#endif
class constrained_hyper_log_log ;

using constrained_type = constrained_hyper_log_log<bool, 13> ;

Constrained types are not constrained instantiation.