hacking std::copy to use pointers

jeremy_k

I'm confused by that code snippet. The following compiles with a selection of compilers for C++11 and later.

#include <cstdint>
#include <algorithm>
#if _cplusplus >= 202002L
#include <span>
#endif

void test(uint8_t *input, size_t length, uint8_t *output)
{
#if _cplusplus >= 202002L
    std::span s{input, length};
    std::copy(s.begin(), s.end(), output);
#endif
    std::copy(input, input + length, output);
}

std::span is a C++20 non-owning wrapper container for contiguous sequences, which might meet your last request.

I don't understand how std::copy() using a separately passed length is superior to memcpy(). I do understand following arbitrary rules to avoid spending time discussing arbitrary rules.

Was the original formulation of the question referring to C++20 ranges?

Kent-Dorfman

No c++20...It is to conform to AUTOSAR "Guidelines for the use of the C++14 language in critical and safety related systems". It's a coding safety standard where much of it makes sense and some of it is pure OCD on the part of the members of the standards committee. LOL Point is, trying to slowly phase out C-ish code but still working on low level embedded systems so "pointers are king". Standard requires documentation and justification each time a rule is violated.

A custom allocator for std::vector may be a work-around but I had hoped something already existed for this use-case.

Yes, this compiles but give erroneous results.

void test(const uint8_t *input, const size_t length, uint8_t *output) {
    std::copy(input, input + length, output);
}

and this is the other real use-case that fails...moving into or out of the destPtr buffer

uint32_t ip = ::htonl((127U << 24U) | 1U);
std::copy(&ip, &ip + sizeof(ip), destPtr);

JKSH

@Kent-Dorfman said in hacking std::copy to use pointers:

Is there a way to hack std::copy so that it can take simple uint8_t* as range limits

You don't need to hack it; the code you posted in your original post should already work (assuming that ADDRESS, DESTINATION, and length are correct).

Yes, this compiles but give erroneous results.

void test(const uint8_t *input, const size_t length, uint8_t *output) {
    std::copy(input, input + length, output);
}

Erroneous how?

and this is the other real use-case that fails...moving into or out of the destPtr buffer
uint32_t ip = ::htonl((127U << 24U) | 1U);
std::copy(&ip, &ip + sizeof(ip), destPtr);

This code will try to copy 16 bytes, not 4 bytes. You want std::copy(&ip, &ip + 1, destPtr);

Kent-Dorfman

This post is deleted!

Kent-Dorfman

std::copy(&ip, &ip + sizeof(ip), destPtr); was definitely a goof on my part.

Kent-Dorfman

another fly in the ointment that makes std::copy unsuitable for moving memory buffers around:

Given the above example of copying the bytes of a uint32_t to an arbitrary memory location, the destPtr (being of type uint8_t*) wont allow std::copy to copy the whole range from the input iteratores. It only copies one byte! Some would argue that destPtr should then be of type uint32_t* but that could cause other problems since there is no guarantee that the destination address would be word aligned...IOW, I don't necessarily want it to be aligned, but instead packed.

going back to using using <cstring> methods that aren't so picky about how you address memory.

JKSH

@Kent-Dorfman You can cast your input pointers to <uint8_t*> (and change your pointer arithmetic for the end iterator) before passing them to std::copy.

going back to using using <cstring> methods

I thought that was strongly discouraged?

Kent-Dorfman

I understand why, but std::copy is just too particular about operand types for generic RAM (network buffer) manipulation. Going back to <cstring> methods that use void*...yeah, it violates the consortium rules but this is embedded programming, not high level algorithmic stuff..and rank has its privilege. I'm the one driving the conformance level so nobody is gonna complain if I make their life easier. and in the end, violations just have to be documented, if I leave the rule turned on. at this point I'm inclined to disable that particular rule as being overly strict for nodes with no safety critical functions.

jeremy_k

memmove() or memcpy() sound like the right abstraction for a block of memory. std::copy() deals a range of items that may be noncontiguous or have copy constructors. Using it on a buffer presumably means breaking that abstraction to get char sized chunks, while having limited potential to utilize the iterator abstraction. I guess it could make sense with a scatter/gather interface.

fcarney

I just don't get where it fails from this discussion:

#include <cstdint>
#include <algorithm>
#include <cassert>

int main(int argc, char *argv[])
{
    constexpr size_t length = 0x20U;
    uint8_t* ptr = new uint8_t[length];
    for(size_t count=0; count<length; ++count)
        ptr[count] = count;
    uint8_t* dest = new uint8_t[length];
    std::copy(ptr, ptr+length, dest);
    for(size_t count=0; count<length; ++count){
        assert(ptr[count] == dest[count]);
    }
    delete[] ptr;
    delete[] dest;

    return 0;
}

I tested in both c++11 and c++17.