How can one format double as hex?
double t = 1.123;
fmt::format("{:x}", t);
It throws exception "invalid type specifier".
I would like to get string 3ff1f7ced916872b
Asked
Active
Viewed 200 times
2
TylerDense
- 45
- 5
-
That's an odd use case I don't know if `fmt` supports. Consider using brute-force to `reinterpret_cast` to `uint64_t`. – user4581301 Sep 13 '22 at 00:20
-
@user4581301 I tried, compiler says `error: invalid cast from type 'double' to type 'uint64_t'` – TylerDense Sep 13 '22 at 00:21
-
1@user4581301: That would violate the [strict aliasing rule](https://stackoverflow.com/q/98650/12149471). I recommend using [`std::bit_cast`](https://en.cppreference.com/w/cpp/numeric/bit_cast) instead. – Andreas Wenzel Sep 13 '22 at 00:22
-
@AndreasWenzel has the better solution. My pitch requires you to dive into pointers, cast the pointers, and then dereference your way back to values. It's, as Andreas says, illegal, but it's the sort of illegal that will only fail on a particularly strange system. – user4581301 Sep 13 '22 at 00:26
-
@AndreasWenzel Failed to build with gcc, standart c++20. `bit_cast is not a member of std` – TylerDense Sep 13 '22 at 00:30
-
@TylerDense: Did you `#include
`, as specified in the linked documentation? – Andreas Wenzel Sep 13 '22 at 00:31 -
@TylerDense: Did you compile with `-std=c++20`? – Andreas Wenzel Sep 13 '22 at 00:33
-
`double t = 1.123; uint64_t x; memcpy(&x, &t, sizeof x)); fmt::format("{:x}", x);` – Eljay Sep 13 '22 at 00:37
-
2If all else fails there's always `memcpy`. https://godbolt.org/z/9jETPbs3G – Retired Ninja Sep 13 '22 at 00:37
-
@TylerDense: `std::bit_cast` works with gcc 12.2. [Demo](https://godbolt.org/z/475b5WrGE) – Andreas Wenzel Sep 13 '22 at 00:39
-
@AndreasWenzel is there c++17 way? – TylerDense Sep 13 '22 at 01:37
-
1@TylerDense: Yes, I believe the `memcpy` solution mentioned by someone else complies with ISO C++17. – Andreas Wenzel Sep 13 '22 at 02:04
1 Answers
2
You can use std::bit_cast to cast double into an appropriately sized integer and format that in hexadecimal, e.g. assuming IEEE754 double:
double t = 1.123;
auto s = fmt::format("{:x}", std::bit_cast<uint64_t>(t));
// s == "3ff1f7ced916872b"
godbolt: https://godbolt.org/z/ehKTrMz7M
vitaut
- 49,672
- 25
- 199
- 336
-
That requires c++20. Is there another option but memcpy to uint64_t in c++17? – TylerDense Sep 13 '22 at 19:18
-
2You can implement a fallback `bit_cast` via `memcpy` as shown in https://en.cppreference.com/w/cpp/numeric/bit_cast. – vitaut Sep 13 '22 at 19:43