Date: Wed, 15 May 2024 07:57:36 +0100
On Tue, 14 May 2024, 23:54 Frederick Virchanza Gotham via Std-Proposals, <
std-proposals_at_[hidden]> wrote:
> I'm experimenting with a consteval implementation of the MD5 hash
> algorithm. I can write a switch statement with compile-time hashes as
> follows:
>
> switch ( md5(some_runtime_value) )
> {
> case md5("dog"): return 1;
> case md5("cat"): return 2;
> case md5("cow"): return 3;
> }
>
> The consteval implementation of 'md5' deals with a consteval array of
> char's -- which is fine when you're hashing strings. It's a little
> more complicated if you want to hash one of the following:
>
> int[3]
> tuple<int,char,bool>
> struct Monkey { int a, b; };
>
> It gets complicated when you want to use std::bit_cast to convert one
> of these three types to an std::array<char,N>, because std::bit_cast
> doesn't yield a compile-time constant if the source type isn't
> trivially copiable. (Most implementations of std::tuple are not
> trivially copiable -- not even when all of the internal types are).
>
> __builtin_bit_cast isn't much better.
Of course not, since it's just the implementation of std::bit_cast, and so
you should only be using std::bit_cast instead
Nor is __builtin_memcpy.
>
Ditto.
Why would you want to hash the indeterminate padding bits between members
anyway?
> If I want to hash an array of integers, I can use the following
> consteval function to turn it into a char array:
>
> template<typename T, std::size_t N> requires std::is_integral_v<T>
> constexpr auto FromIntegralArray( T const (&arg)[N] )
> {
> using std::size_t;
> std::array<char, sizeof(arg)> retval{};
> for ( size_t i = 0u; i < sizeof(arg); ++i )
> {
> size_t const a = i / sizeof(T),
> b = i % sizeof(T);
>
> retval[i] = arg[a] >> (CHAR_BIT * b);
> }
> return retval;
> }
>
> But then I would need to (try to) write a function that would work
> with floating points, tuples, and so on.
>
> So is anyone talking about making bit_cast more versatile? Or even
> allowing a consteval memcpy, so we can do this:
>
> template<typename T>
> constexpr auto FromIntegralArray( T const &arg )
> {
> using std::size_t;
> std::array<char, sizeof(arg)> retval{};
> std::memcpy( &retval.front(), &arg, sizeof retval );
> return retval;
> }
> --
> Std-Proposals mailing list
> Std-Proposals_at_[hidden]
> https://lists.isocpp.org/mailman/listinfo.cgi/std-proposals
>
std-proposals_at_[hidden]> wrote:
> I'm experimenting with a consteval implementation of the MD5 hash
> algorithm. I can write a switch statement with compile-time hashes as
> follows:
>
> switch ( md5(some_runtime_value) )
> {
> case md5("dog"): return 1;
> case md5("cat"): return 2;
> case md5("cow"): return 3;
> }
>
> The consteval implementation of 'md5' deals with a consteval array of
> char's -- which is fine when you're hashing strings. It's a little
> more complicated if you want to hash one of the following:
>
> int[3]
> tuple<int,char,bool>
> struct Monkey { int a, b; };
>
> It gets complicated when you want to use std::bit_cast to convert one
> of these three types to an std::array<char,N>, because std::bit_cast
> doesn't yield a compile-time constant if the source type isn't
> trivially copiable. (Most implementations of std::tuple are not
> trivially copiable -- not even when all of the internal types are).
>
> __builtin_bit_cast isn't much better.
Of course not, since it's just the implementation of std::bit_cast, and so
you should only be using std::bit_cast instead
Nor is __builtin_memcpy.
>
Ditto.
Why would you want to hash the indeterminate padding bits between members
anyway?
> If I want to hash an array of integers, I can use the following
> consteval function to turn it into a char array:
>
> template<typename T, std::size_t N> requires std::is_integral_v<T>
> constexpr auto FromIntegralArray( T const (&arg)[N] )
> {
> using std::size_t;
> std::array<char, sizeof(arg)> retval{};
> for ( size_t i = 0u; i < sizeof(arg); ++i )
> {
> size_t const a = i / sizeof(T),
> b = i % sizeof(T);
>
> retval[i] = arg[a] >> (CHAR_BIT * b);
> }
> return retval;
> }
>
> But then I would need to (try to) write a function that would work
> with floating points, tuples, and so on.
>
> So is anyone talking about making bit_cast more versatile? Or even
> allowing a consteval memcpy, so we can do this:
>
> template<typename T>
> constexpr auto FromIntegralArray( T const &arg )
> {
> using std::size_t;
> std::array<char, sizeof(arg)> retval{};
> std::memcpy( &retval.front(), &arg, sizeof retval );
> return retval;
> }
> --
> Std-Proposals mailing list
> Std-Proposals_at_[hidden]
> https://lists.isocpp.org/mailman/listinfo.cgi/std-proposals
>
Received on 2024-05-15 06:58:58