[ML] Avoiding C++17 warnings for Windows builds #2120
Description
Our Windows build logs are currently warning of a number of C++ features we are using that are deprecated in C++17 and removed in C++20. We should try to stop using these before we upgrade to C++20 (although if we cannot then it may be possible to continue to use the deprecated features by defining certain macros).
In our own code we have:
C:\Users\jenkins\workspace\elastic+machine-learning+main+windows+debug\include\core/Concurrency.h(139,18): warning C4996: 'std::result_of_t': warning STL4014: std::result_of and std::result_of_t are deprecated in C++17. They are superseded by std::invoke_result and std::invoke_result_t. You can define _SILENCE_CXX17_RESULT_OF_DEPRECATION_WARNING or _SILENCE_ALL_CXX17_DEPRECATION_WARNINGS to acknowledge that you have received this warning.
C:\Users\jenkins\workspace\elastic+machine-learning+main+windows+debug\include\core/Concurrency.h(141,20): warning C4996: 'std::result_of_t': warning STL4014: std::result_of and std::result_of_t are deprecated in C++17. They are superseded by std::invoke_result and std::invoke_result_t. You can define _SILENCE_CXX17_RESULT_OF_DEPRECATION_WARNING or _SILENCE_ALL_CXX17_DEPRECATION_WARNINGS to acknowledge that you have received this warning.
CMemoryUsageTest.cc(244,29): warning C4996: 'std::allocator<void>::const_pointer': warning STL4010: Various members of std::allocator are deprecated in C++17. Use std::allocator_traits instead of accessing these members directly. You can define _SILENCE_CXX17_OLD_ALLOCATOR_MEMBERS_DEPRECATION_WARNING or _SILENCE_ALL_CXX17_DEPRECATION_WARNINGS to acknowledge that you have received this warning.
These should be easy to avoid by changing our code to comply with C++20.
In Boost we have:
C:\usr\local\include\boost-1_77\boost/variant/variant.hpp(1003,5): warning C4996: 'std::_Weak_result_type<std::_Weak_binary_args<ml::maths::common::`anonymous-namespace'::CBinaryVisitor<double,ml::maths::common::`anonymous-namespace'::SPdf>,void>,void>::result_type': warning STL4007: Many result_type typedefs and all argument_type, first_argument_type, and second_argument_type typedefs are deprecated in C++17. You can define _SILENCE_CXX17_ADAPTOR_TYPEDEFS_DEPRECATION_WARNING or _SILENCE_ALL_CXX17_DEPRECATION_WARNINGS to acknowledge that you have received this warning.
C:\usr\local\include\boost-1_77\boost/unordered/detail/implementation.hpp(1452,7): warning C4996: 'std::allocator<std::pair<const std::basic_string<char,std::char_traits<char>,std::allocator<char>>,std::basic_string<char,std::char_traits<char>,std::allocator<char>>>>::is_always_equal': warning STL4010: Various members of std::allocator are deprecated in C++17. Use std::allocator_traits instead of accessing these members directly. You can define _SILENCE_CXX17_OLD_ALLOCATOR_MEMBERS_DEPRECATION_WARNING or _SILENCE_ALL_CXX17_DEPRECATION_WARNINGS to acknowledge that you have received this warning.
For the boost::variant problem we should be able to switch to std::variant to avoid it. Our files that use boost::variant are:
model/CModelTools.h: boost::variant<maths::common::CJointProbabilityOfLessLikelySamples, maths::common::CProbabilityOfExtremeSample>;
maths/common/CBjkstUniqueValues.h: using TUInt32VecOrSketch = boost::variant<TUInt32Vec, SSketch>;
maths/common/CMixtureDistribution.h: boost::variant<boost::math::normal_distribution<>, boost::math::gamma_distribution<>, boost::math::lognormal_distribution<>>;
maths/common/CSetTools.h: using TSizeOrSizeSet = boost::variant<std::size_t, TSizeSet>;
maths/time_series/CCountMinSketch.h: using TUInt32FloatPrVecOrSketch = boost::variant<TUInt32FloatPrVec, SSketch>;
std::variant is not exactly the same as boost::variant, but it's close, so hopefully it won't be too much work to switch.
For the other Boost issue in the unordered detail, there's an issue open for it and somebody said they would fix it on a mailing list. They didn't follow up on their promise, but since it sounds like a 1 or 2 line fix and the mailing list states exactly what to do maybe we can contribute the fix to Boost.
Then there are a couple of issues with Eigen:
C:\Users\jenkins\workspace\elastic+machine-learning+main+windows+debug\3rd_party\eigen\Eigen\src/Core/util/Meta.h(349,2): warning C4996: 'std::equal_to<double>::result_type': warning STL4007: Many result_type typedefs and all argument_type, first_argument_type, and second_argument_type typedefs are deprecated in C++17. You can define _SILENCE_CXX17_ADAPTOR_TYPEDEFS_DEPRECATION_WARNING or _SILENCE_ALL_CXX17_DEPRECATION_WARNINGS to acknowledge that you have received this warning.
C:\Users\jenkins\workspace\elastic+machine-learning+main+windows+debug\3rd_party\eigen\Eigen\src/Core/util/Meta.h(349,2): warning C4996: 'std::equal_to<ml::core::CFloatStorage>::result_type': warning STL4007: Many result_type typedefs and all argument_type, first_argument_type, and second_argument_type typedefs are deprecated in C++17. You can define _SILENCE_CXX17_ADAPTOR_TYPEDEFS_DEPRECATION_WARNING or _SILENCE_ALL_CXX17_DEPRECATION_WARNINGS to acknowledge that you have received this warning.
Possibly we can upgrade Eigen to avoid these, or, if not, maybe we can contribute a fix there too.
Finally, there are a number of issues in PyTorch:
C:\usr\local\include\pytorch\ATen/core/List.h(106,40): warning C4996: 'std::iterator<std::random_access_iterator_tag,T,ptrdiff_t,T *,T &>': warning STL4015: The std::iterator class template (used as a base class to provide typedefs) is deprecated in C++17. (The <iterator> header is NOT deprecated.) The C++ Standard has never required user-defined iterators to derive from std::iterator. To fix this warning, stop deriving from std::iterator and start providing publicly accessible typedefs named iterator_category, value_type, difference_type, pointer, and reference. Note that value_type is required to be non-const, even for constant iterators. You can define _SILENCE_CXX17_ITERATOR_BASE_CLASS_DEPRECATION_WARNING or _SILENCE_ALL_CXX17_DEPRECATION_WARNINGS to acknowledge that you have received this warning.
C:\usr\local\include\pytorch\ATen/core/List.h(156,3): warning C4996: 'std::iterator<std::random_access_iterator_tag,T,ptrdiff_t,T *,T &>::difference_type': warning STL4015: The std::iterator class template (used as a base class to provide typedefs) is deprecated in C++17. (The <iterator> header is NOT deprecated.) The C++ Standard has never required user-defined iterators to derive from std::iterator. To fix this warning, stop deriving from std::iterator and start providing publicly accessible typedefs named iterator_category, value_type, difference_type, pointer, and reference. Note that value_type is required to be non-const, even for constant iterators. You can define _SILENCE_CXX17_ITERATOR_BASE_CLASS_DEPRECATION_WARNING or _SILENCE_ALL_CXX17_DEPRECATION_WARNINGS to acknowledge that you have received this warning.
C:\usr\local\include\pytorch\c10/util/Optional.h(1129,3): warning C4996: 'std::_Conditionally_enabled_hash<_Kty,true>::result_type': warning STL4007: Many result_type typedefs and all argument_type, first_argument_type, and second_argument_type typedefs are deprecated in C++17. You can define _SILENCE_CXX17_ADAPTOR_TYPEDEFS_DEPRECATION_WARNING or _SILENCE_ALL_CXX17_DEPRECATION_WARNINGS to acknowledge that you have received this warning.
C:\usr\local\include\pytorch\c10/util/TypeList.h(502,26): warning C4996: 'std::result_of_t': warning STL4014: std::result_of and std::result_of_t are deprecated in C++17. They are superseded by std::invoke_result and std::invoke_result_t. You can define _SILENCE_CXX17_RESULT_OF_DEPRECATION_WARNING or _SILENCE_ALL_CXX17_DEPRECATION_WARNINGS to acknowledge that you have received this warning.
C:\usr\local\include\pytorch\c10/util/irange.h(19,32): warning C4996: 'std::iterator<std::input_iterator_tag,I,ptrdiff_t,I *,I &>': warning STL4015: The std::iterator class template (used as a base class to provide typedefs) is deprecated in C++17. (The <iterator> header is NOT deprecated.) The C++ Standard has never required user-defined iterators to derive from std::iterator. To fix this warning, stop deriving from std::iterator and start providing publicly accessible typedefs named iterator_category, value_type, difference_type, pointer, and reference. Note that value_type is required to be non-const, even for constant iterators. You can define _SILENCE_CXX17_ITERATOR_BASE_CLASS_DEPRECATION_WARNING or _SILENCE_ALL_CXX17_DEPRECATION_WARNINGS to acknowledge that you have received this warning.
PyTorch itself builds with C++14, so won't see these. We want their headers to also work when we include them in C++20 code. Probably the first step is to see if there's already an issue open against PyTorch for this and open one if not. If it's too controversial for them to adjust their code to be compatible with C++20 then maybe this is where we can use the macros for allowing removed features to still be used, i.e. define _HAS_FEATURES_REMOVED_IN_CXX20, but only when building pytorch_inference.