Support gcc 12 as the C++ compiler

[robertmaynard]

Description

Corrects compilation issues found by gcc 12 and the new warnings it provides.

Three major classes of issues was found:

• incorrect lambda captures of the wrong container for the iterators being used
• incorrect implicit conversions
• warnings generated by using std::transform and std::initializer_lists

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[jrhemstad]

Why do we need to copy the init_list?

[robertmaynard]

Reproducer:

#include <algorithm>
#include <cstdint>
#include <vector>
#include <thrust/host_vector.h>


template <typename TypeParam, typename T>
thrust::host_vector<TypeParam>
make_type_param_vector(std::initializer_list<T> const& init_list) {
  // std::vector<T> input{init_list};
  std::vector<TypeParam> vec(init_list.size()); //or thrust::host_vector
  std::transform(std::cbegin(init_list), std::cend(init_list), std::begin(vec), [](auto const& e) {
    if constexpr (std::is_unsigned_v<TypeParam>) { return static_cast<TypeParam>(std::abs(e)); }
    return static_cast<TypeParam>(e);
  });
  return vec;
}


template <typename T> bool validate_A() {
  auto const input_column_valid_a = make_type_param_vector<uint8_t>({1, 0});
  auto const input_column_valid_b = make_type_param_vector<uint8_t>({0, 0});
  auto const input_column_valid_c = make_type_param_vector<T>({15, 16});
  return 0;
}
int main() {
  validate_A<float>();
  validate_A<double>();
  validate_A<int8_t>();
  validate_A<int16_t>();
  validate_A<int32_t>();
  validate_A<int64_t>();
  validate_A<uint8_t>();
  validate_A<uint16_t>();
  validate_A<uint32_t>();
  return 0;
}

Will generate warnings ( as errors ) of:

    inlined from 'thrust::host_vector<TypeParam> make_type_param_vector(const std::initializer_list<_Value>&) [with TypeParam = unsigned char; T = int]' at /home/rmaynard/Work/temp/sample2.cpp:11:34:
/home/rmaynard/miniconda3/envs/cuda_tk12/x86_64-conda-linux-gnu/include/c++/12.2.0/bits/new_allocator.h:137:55: note: at offset [29, 4611686018427387917] into destination object of size 3 allocated by 'operator new'
  137 |         return static_cast<_Tp*>(_GLIBCXX_OPERATOR_NEW(__n * sizeof(_Tp)));
      |                                                       ^
In function '_OIter std::transform(_IIter, _IIter, _OIter, _UnaryOperation) [with _IIter = const int*; _OIter = thrust::detail::normal_iterator<unsigned char*>; _UnaryOperation = make_type_param_vector<unsigned char, int>(const std::initializer_list<int>&)::<lambda(const auto:1&)>]',
    inlined from 'thrust::host_vector<TypeParam> make_type_param_vector(const std::initializer_list<_Value>&) [with TypeParam = unsigned char; T = int]' at /home/rmaynard/Work/temp/sample2.cpp:12:17:
/home/rmaynard/miniconda3/envs/cuda_tk12/x86_64-conda-linux-gnu/include/c++/12.2.0/bits/stl_algo.h:4263:19: error: writing 1 byte into a region of size 0 [-Werror=stringop-overflow=]
 4263 |         *__result = __unary_op(*__first);

I believe that this is a bug in gcc as it is easit to trigger when we have at least 3 calls to make_type_param_vector with input of the same length with no other calls of a differing length to make_type_param_vector. Copying the initializer_list to a std::vector avoids this bug.

[harrism]

Why so many reviewers assigned?

[robertmaynard]

Why so many reviewers assigned?

I believe that you and ttnghia got auto assigned, and everyone else wanted to see the gcc-12 changes.

[robertmaynard]

/merge