checked conversion function for C++20

Question

Similar to C++ int_cast<> function for checked casts?, but C++20, and with target type deduced from context.

The goal is to implement a runtime check that the value being converted can be represented in the target type, and throw an exception if it can't.

This basically works for my use case, but I wonder if there's anything obvious I've missed.

Use as

std::uint32_t x = 0x40000000;
std::int32_t y = checked_cast(x); // works
std::uint16_t z = checked_cast(x); // throws

Code:

#pragma once
#include <stdexcept>
#include <utility>
class bad_checked_cast : public std::logic_error
{
public:
 bad_checked_cast() : std::logic_error("Bad checked_cast") { }
};
template<typename From>
struct checked_cast_result_proxy
{
 template<typename To>
 operator To() const
 {
 if(std::in_range<To>(value))
 return static_cast<To>(value);
 else
 throw bad_checked_cast();
 }
 From value;
};
template<typename From>
auto checked_cast(From value) -> checked_cast_result_proxy<From>
{
 return checked_cast_result_proxy<From> { value };
}

Answer 1

There's an argument (no pun intended!) that it's not an error in the program logic for the value to be out of the convertible range. I would suggest using std::domain_error as the base of bad_checked_cast.

---

We should constrain the template to reject types that std::in_range() doesn't accept:

#include <concepts>
⋮
template<typename From>
 requires std::integral<From>
struct checked_cast_result_proxy
⋮
 template<typename To>
 requires std::integral<To>
 operator To() const

std::in_range cannot be used with

enums (including std::byte), char, char8_t, char16_t, char32_t, wchar_t and bool.

That's harder to enforce, as there isn't a concept that expresses that restriction (and the library I'm using allows invocation, but than fails at a static_assert).

---

I found it useful to have a plain function interface to collapse the proxy object, for when there isn't a specific enough context to do so (e.g. passing into another template function that requires an arithmetic type):

template<typename To, typename From>
 requires std::integral<From> && std::integral<To>
static To checked_cast_to(From from) { return checked_cast(from); }

That enabled me to test throwing with this:

TEST(CheckedCast, success)
{
 const std::uint32_t x = 0x40000000;
 std::int32_t y = checked_cast(x);
 EXPECT_EQ(x, y);
}
TEST(CheckedCast, fail)
{
 const std::uint32_t x = 0x40000000;
 EXPECT_THROW(checked_cast_to<std::uint16_t>(x), std::exception);
}

---

You could consider generalising to make it work with other (non-integer) arithmetic types (e.g. floating and complex numbers). That might be a lot of work that's not useful for you, though.

• Deduplicator: Allowing the caller to manually specify the target-type is a good idea. But using a different name for that is so inconvenient.