Function checking if an integer type can fit a value of possibly different (integer) type

Question

Is it possible to create a templated function that checks if a primitive data type can fit a value of potentially different primitive data type? Let's limit the scope to integer types for the moment.

More precisely: Is it possible to create a "one fit all" templated functions yet without getting compiler warnings (boolean expression always true/false, signed/unsigned comparison, unused variable) and without disabling compiler warning checks? The functions should also limit as much as possible checks at runtime (all trivial cases should be excluded at compile time). If possible, I would prefer avoiding using extensions from C++11 and the like (unless a "quick" replacement for "old" C++ exists).

Note: "value" is not known at compile time, only its type.

Example of expected behaviour:

int main(int argc, char** argv) {
    for (int i = 1; i < argc; i++) {
        const int value = atoi(argv[i]);
        std::cout << value << ": ";
        std::cout << CanTypeFitValue<int8_t>(value) << " ";
        std::cout << CanTypeFitValue<uint8_t>(value) << " ";
        std::cout << CanTypeFitValue<int16_t>(value) << " ";
        std::cout << CanTypeFitValue<uint16_t>(value) << " ";
        std::cout << CanTypeFitValue<int32_t>(value) << " ";
        std::cout << CanTypeFitValue<uint32_t>(value) << " ";
        std::cout << CanTypeFitValue<int64_t>(value) << " ";
        std::cout << CanTypeFitValue<uint64_t>(value) << std::endl;
        }
    
}

Output:

./a.out 6 1203032847 2394857 -13423 9324 -192992929

6: 1 1 1 1 1 1 1 1

1203032847: 0 0 0 0 1 1 1 1

2394857: 0 0 0 0 1 1 1 1

-13423: 0 0 1 0 1 0 1 0

9324: 0 0 1 1 1 1 1 1

-192992929: 0 0 0 0 1 0 1 0

Test your code here or here.

Check the assembly generated here.

This question was inspired by this post

Answer 1

Using numeric_limits and types defined in stdint.h

More compact that my first solution, same efficiency.

Drawback: one additional header to be included.

#include <limits>
#include <stdint.h>

using std::numeric_limits;

template <typename T, typename U>
    bool CanTypeFitValue(const U value) {
        const intmax_t botT = intmax_t(numeric_limits<T>::min() );
        const intmax_t botU = intmax_t(numeric_limits<U>::min() );
        const uintmax_t topT = uintmax_t(numeric_limits<T>::max() );
        const uintmax_t topU = uintmax_t(numeric_limits<U>::max() );
        return !( (botT > botU && value < static_cast<U> (botT)) || (topT < topU && value > static_cast<U> (topT)) );        
    }

Assembly code generated (you can change T and U types)

Correctness test

---

Note: a constexpr version was written, but apparently it has some problems. See here and here.

Answer 2

Using the features of C++14 (leave out constexpr for C++11 compatibility) and use of templates, this is what I came up with:

https://ideone.com/OSc9CI (updated version: now also accepts unsigned to signed, short and beautiful)

This basically uses std::enable_if extensively with type_traits std::is_unsigned and std::is_integral. Best to read from bottom up (as the decision tree builds up from there).

Obviously this is nearly all done compile time, so assembly should be fairly small.

This solution can handle integral and floating point target types as well as integral and floating point original types.

If the check isn't trivial (i.e. bounds of data type have to be checked), the actual_type value n is casted to typename std::common_type<target, actual_type>::type statically.

Every decision is_integral and is_unsigned and is_same is done at compile time, so no overhead from this at runtime. The check boils down to some lower_bound(target) <= value and / or value <= upper_bound(target) after the types are casted to a common type (to avoid warnings and prevent overflows).

#include <cmath> // necessary to check for floats too
#include <cstdint> // for testing only
#include <iomanip> // for testing only
#include <iostream> // for testing only
#include <limits> // necessary to check ranges
#include <type_traits> // necessary to check type properties (very efficient, compile time!)

// the upper bound must always be checked
template <typename target_type, typename actual_type>
constexpr bool test_upper_bound(const actual_type n)
{
   typedef typename std::common_type<target_type, actual_type>::type common_type;
   const auto c_n = static_cast<common_type>(n);
   const auto t_max = static_cast<common_type>(std::numeric_limits<target_type>::max());
   return ( c_n <= t_max );
}

// the lower bound is only needed to be checked explicitely in non-trivial cases, see the next three functions
template <typename target_type, typename actual_type>
constexpr typename std::enable_if<!(std::is_unsigned<target_type>::value) && !(std::is_unsigned<actual_type>::value), bool>::type
test_lower_bound(const actual_type n)
{
   typedef typename std::common_type<target_type, actual_type>::type common_type;
   const auto c_n = static_cast<common_type>(n);
   const auto t_min_as_t = std::numeric_limits<target_type>::lowest();
   const auto t_min = static_cast<common_type>(t_min_as_t);
   return (c_n >= t_min);
}

// for signed target types where the actual type is unsigned, the lower bound is trivially satisfied.
template <typename target_type, typename actual_type>
constexpr typename std::enable_if<!(std::is_unsigned<target_type>::value) &&(std::is_unsigned<actual_type>::value), bool>::type
test_lower_bound(const actual_type n)
{
    return true;
}
    
// for unsigned target types, the sign of n musn't be negative
// but that's not an issue with unsigned actual_type
template <typename target_type, typename actual_type>
constexpr typename std::enable_if<std::is_integral<target_type>::value &&
                        std::is_unsigned<target_type>::value &&
                        std::is_integral<actual_type>::value &&
                        std::is_unsigned<actual_type>::value, bool>::type
test_lower_bound(const actual_type)
{
   return true;
}

// for unsigned target types, the sign of n musn't be negative
template <typename target_type, typename actual_type>
constexpr typename std::enable_if<std::is_integral<target_type>::value &&
                        std::is_unsigned<target_type>::value &&
                        (!std::is_integral<actual_type>::value ||
                         !std::is_unsigned<actual_type>::value), bool>::type
test_lower_bound(const actual_type n)
{
   return ( n >= 0 );
}

// value may be integral if the target type is non-integral
template <typename target_type, typename actual_type>
constexpr typename std::enable_if<!std::is_integral<target_type>::value, bool>::type
test_integrality(const actual_type)
{
   return true;
}

// value must be integral if the target type is integral
template <typename target_type, typename actual_type>
constexpr typename std::enable_if<std::is_integral<target_type>::value, bool>::type
test_integrality(const actual_type n)
{
   return ( (std::abs(n - std::floor(n)) < 1e-8) || (std::abs(n - std::ceil(n)) < 1e-8) );
}

// perform check only if non-trivial
template <typename target_type, typename actual_type>
constexpr typename std::enable_if<!std::is_same<target_type, actual_type>::value, bool>::type
CanTypeFitValue(const actual_type n)
{
   return test_upper_bound<target_type>(n) &&
          test_lower_bound<target_type>(n) &&
          test_integrality<target_type>(n);
}


// trivial case: actual_type == target_type
template <typename actual_type>
constexpr bool CanTypeFitValue(const actual_type)
{
   return true;
}

int main()
{
   int ns[] = {6, 1203032847, 2394857, -13423, 9324, -192992929};
   for ( const auto n : ns )
   {
      std::cout << std::setw(10) << n << "\t";
      std::cout << " " << CanTypeFitValue<int8_t>(n);
      std::cout << " " << CanTypeFitValue<uint8_t>(n);
      std::cout << " " << CanTypeFitValue<int16_t>(n);
      std::cout << " " << CanTypeFitValue<uint16_t>(n);
      std::cout << " " << CanTypeFitValue<int32_t>(n);
      std::cout << " " << CanTypeFitValue<uint32_t>(n);
      std::cout << " " << CanTypeFitValue<int64_t>(n);
      std::cout << " " << CanTypeFitValue<uint64_t>(n);
      std::cout << " " << CanTypeFitValue<float>(n);
      std::cout << " " << CanTypeFitValue<double>(n);
      std::cout << "\n";
   }
   std::cout << "\n";
   unsigned long long uss[] = {6, 1201146189143ull, 2397, 23};
   for ( const auto n : uss )
   {
      std::cout << std::setw(10) << n << "\t";
      std::cout << " " << CanTypeFitValue<int8_t>(n);
      std::cout << " " << CanTypeFitValue<uint8_t>(n);
      std::cout << " " << CanTypeFitValue<int16_t>(n);
      std::cout << " " << CanTypeFitValue<uint16_t>(n);
      std::cout << " " << CanTypeFitValue<int32_t>(n);
      std::cout << " " << CanTypeFitValue<uint32_t>(n);
      std::cout << " " << CanTypeFitValue<int64_t>(n);
      std::cout << " " << CanTypeFitValue<uint64_t>(n);
      std::cout << " " << CanTypeFitValue<float>(n);
      std::cout << " " << CanTypeFitValue<double>(n);
      std::cout << "\n";
   }
   std::cout << "\n";
   float fs[] = {0.0, 0.5, -0.5, 1.0, -1.0, 1e10, -1e10};
   for ( const auto f : fs )
   {
      std::cout << std::setw(10) << f << "\t";
      std::cout << " " << CanTypeFitValue<int8_t>(f);
      std::cout << " " << CanTypeFitValue<uint8_t>(f);
      std::cout << " " << CanTypeFitValue<int16_t>(f);
      std::cout << " " << CanTypeFitValue<uint16_t>(f);
      std::cout << " " << CanTypeFitValue<int32_t>(f);
      std::cout << " " << CanTypeFitValue<uint32_t>(f);
      std::cout << " " << CanTypeFitValue<int64_t>(f);
      std::cout << " " << CanTypeFitValue<uint64_t>(f);
      std::cout << " " << CanTypeFitValue<float>(f);
      std::cout << " " << CanTypeFitValue<double>(f);
      std::cout << "\n";
   }
}

This (new) version quickly decides (at compile time!) if checks are needed (concerning upper bound, lower bound and integrality) and uses the correct version (to avoid warnings about stupid >= 0 comparisons with unsigned types) (also at compile time). E.g. the integrality does not need to be checked if the target is float, the lower bound does not need to be checked if both types are unsigned etc.

The most obvious optimization (having equal types), is done with std::is_same.

This approach can also be extended to used-defined types with specialized templates. Checks such as std::is_integral will be negative on those types.

You can check that the assembler output is fairly small (except for the obvious case of floats) here or by invoking g++ with -S.