xt::xstrided_slice<std::ptrdiff_t> are not comparable

[jblespiau]

It's as simple as doing:

xt::all() == xt::all()

experimental/users/jblespiau/tensorfn/pybind11/pmap_function.cc:72:30: error: invalid operands to binary expression ('xt::xall_tag' and 'xt::xall_tag')
          bool c = xt::all() == xt::all();
                   ~~~~~~~~~ ^  ~~~~~~~~~
./third_party/xtensor/xutils.hpp:737:17: note: candidate template ignored: could not match 'tracking_allocator<T, AT, PT>' against 'xt::xall_tag'
    inline bool operator==(const tracking_allocator<T, AT, PT>&, const tracking_allocator<U, AU, PU>&)
                ^
./third_party/xtensor/xstorage.hpp:545:17: note: candidate template ignored: could not match 'uvector<T, A>' against 'xt::xall_tag'
    inline bool operator==(const uvector<T, A>& lhs, const uvector<T, A>& rhs)
                ^
./third_party/xtensor/xstorage.hpp:1270:17: note: candidate template ignored: could not match 'std::vector<T>' against 'xt::xall_tag'
    inline bool operator==(const std::vector<T>& lhs, const svector<T, N, A, Init>& rhs)
                ^
./third_party/xtensor/xstorage.hpp:1276:17: note: candidate template ignored: could not match 'svector<T, N, A, Init>' against 'xt::xall_tag'
    inline bool operator==(const svector<T, N, A, Init>& lhs, const std::vector<T>& rhs)
                ^
experimental/users/jblespiau/tensorfn/pybind11/pmap_function.cc:72:30: note: remaining 9 candidates omitted; pass -fshow-overloads=all to show them
          bool c = xt::all() == xt::all();

The type is defined as:

    template <class T>
    using xstrided_slice = xtl::variant<
        T,

        xrange_adaptor<placeholders::xtuph, T, T>,
        xrange_adaptor<T, placeholders::xtuph, T>,
        xrange_adaptor<T, T, placeholders::xtuph>,

        xrange_adaptor<T, placeholders::xtuph, placeholders::xtuph>,
        xrange_adaptor<placeholders::xtuph, T, placeholders::xtuph>,
        xrange_adaptor<placeholders::xtuph, placeholders::xtuph, T>,

        xrange_adaptor<T, T, T>,
        xrange_adaptor<placeholders::xtuph, placeholders::xtuph, placeholders::xtuph>,

        xrange<T>,
        xstepped_range<T>,

        xall_tag,
        xellipsis_tag,
        xnewaxis_tag
    >;

As a side-note, I was wondering if we should upgrade to C++ 17 to use std::variant instead of a backport. Another altnertive would be to use e.g. https://github.com/abseil/abseil-cpp/blob/master/absl/types/variant.h which will be an alias to std::variant when available.

[JohanMabille]

Hi,

Sorry for the late reply. The issue is indseed dut to the lack of coparison operators for the slices, that's something we should add in xtensor.

Regarding the upgrade to C++17, it has been discussed for xtensor. However, the XTL is also used by libraries which needs to support C++11/14, thus the need for a backport of std::variant. We are happy with this backport (easier to update than retrieving it from a huge library), but we could definitely add a "selection" layer here, that will simply use the standard implementation when available.