call of overloaded ‘for_each_value_in_order<(nda::shape<>::rank() - 1)>(nda::shape<>::index_type, void (&)(const int&, int&), std::pair<const int*, std::tuple<> >&, std::pair<int*, std::tuple<> >&)’ is ambiguous

[Mr-doooog]

mkdir -p obj/test g++ -I. -c -o obj/test/algorithm.o test/algorithm.cpp -O2 -ffast-math -fstrict-aliasing -march=native -std=c++14 -Wall In file included from test/algorithm.cpp:15: ./array.h: In instantiation of ‘void nda::for_each_value_in_order(const Shape&, const ShapeA&, PtrA, const ShapeB&, PtrB, Fn&&) [with Shape = nda::shape<>; ShapeA = nda::shape<>; PtrA = const int*; ShapeB = nda::shape<>; PtrB = int*; Fn = void (&)(const int&, int&); <template-parameter-1-7> = void]’: ./array.h:1796:28: required from ‘static void nda::copy_shape_traits<ShapeSrc, ShapeDst>::for_each_value(const ShapeSrc&, TSrc, const ShapeDst&, TDst, Fn&&) [with Fn = void (&)(const int&, int&); TSrc = const int*; TDst = int*; ShapeSrc = nda::shape<>; ShapeDst = nda::shape<>]’ ./array.h:2598:56: required from ‘void nda::copy(const nda::array_ref<TSrc, ShapeSrc>&, const nda::array_ref<TDst, ShapeDst>&) [with TSrc = const int; TDst = int; ShapeSrc = nda::shape<>; ShapeDst = nda::shape<>; <template-parameter-1-5> = void]’ ./array.h:2614:7: required from ‘void nda::copy(const nda::array<TSrc, ShapeSrc, AllocSrc>&, nda::array<TDst, ShapeDst, AllocDst>&) [with TSrc = int; TDst = int; ShapeSrc = nda::shape<>; ShapeDst = nda::shape<>; AllocSrc = std::allocator<int>; AllocDst = std::allocator<int>; <template-parameter-1-7> = void]’ test/algorithm.cpp:76:12: required from here ./array.h:1602:55: error: call of overloaded ‘for_each_value_in_order<(nda::shape<>::rank() - 1)>(nda::shape<>::index_type, void (&)(const int&, int&), std::pair<const int*, std::tuple<> >&, std::pair<int*, std::tuple<> >&)’ is ambiguous 1602 | internal::for_each_value_in_order<Shape::rank() - 1>(shape.extent(), fn, a, b); | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~ ./array.h:1361:41: note: candidate: ‘void nda::internal::for_each_value_in_order(const ExtentType&, Fn&&, Ptrs ...) [with long unsigned int D = 18446744073709551615; ExtentType = std::tuple<>; Fn = void (&)(const int&, int&); Ptrs = {std::pair<const int*, std::tuple<> >, std::pair<int*, std::tuple<> >}]’ 1361 | NDARRAY_INLINE NDARRAY_HOST_DEVICE void for_each_value_in_order( | ^~~~~~~~~~~~~~~~~~~~~~~ ./array.h:1369:41: note: candidate: ‘void nda::internal::for_each_value_in_order(const std::tuple<>&, Fn&&, Ptrs ...) [with long unsigned int D = 18446744073709551615; Fn = void (&)(const int&, int&); Ptrs = {std::pair<const int*, std::tuple<> >, std::pair<int*, std::tuple<> >}]’ 1369 | NDARRAY_INLINE NDARRAY_HOST_DEVICE void for_each_value_in_order( | ^~~~~~~~~~~~~~~~~~~~~~~ ./array.h: In instantiation of ‘void nda::for_each_value_in_order(const Shape&, Ptr, Fn&&) [with Shape = nda::shape<>; Ptr = int*; Fn = nda::generate(const nda::array_ref<T, Shape>&, Generator&&) [with T = int; Shape = nda::shape<>; Generator = int (&)(); <template-parameter-1-4> = int]::<lambda(int&)>&; <template-parameter-1-4> = void]’: ./array.h:1771:28: required from ‘static void nda::shape_traits<Shape>::for_each_value(const Shape&, Ptr, Fn&&) [with Ptr = int*; Fn = nda::generate(const nda::array_ref<T, Shape>&, Generator&&) [with T = int; Shape = nda::shape<>; Generator = int (&)(); <template-parameter-1-4> = int]::<lambda(int&)>&; Shape = nda::shape<>]’ ./array.h:1978:38: required from ‘void nda::array_ref<T, Shape>::for_each_value(Fn&&) const [with Fn = nda::generate(const nda::array_ref<T, Shape>&, Generator&&) [with T = int; Shape = nda::shape<>; Generator = int (&)(); <template-parameter-1-4> = int]::<lambda(int&)>; <template-parameter-2-2> = void; T = int; Shape = nda::shape<>]’ ./array.h:2747:3: required from ‘void nda::generate(const nda::array_ref<T, Shape>&, Generator&&) [with T = int; Shape = nda::shape<>; Generator = int (&)(); <template-parameter-1-4> = int]’ ./array.h:2752:11: required from ‘void nda::generate(nda::array<T, Shape, Alloc>&, Generator&&) [with T = int; Shape = nda::shape<>; Alloc = std::allocator<int>; Generator = int (&)(); <template-parameter-1-5> = int]’ test/algorithm.cpp:73:19: required from here ./array.h:1585:55: error: call of overloaded ‘for_each_value_in_order<(nda::shape<>::rank() - 1)>(nda::shape<>::index_type, nda::generate(const nda::array_ref<T, Shape>&, Generator&&) [with T = int; Shape = nda::shape<>; Generator = int (&)(); <template-parameter-1-4> = int]::<lambda(int&)>&, std::pair<int*, std::tuple<> >&)’ is ambiguous 1585 | internal::for_each_value_in_order<Shape::rank() - 1>(shape.extent(), fn, base_and_stride); | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ./array.h:1361:41: note: candidate: ‘void nda::internal::for_each_value_in_order(const ExtentType&, Fn&&, Ptrs ...) [with long unsigned int D = 18446744073709551615; ExtentType = std::tuple<>; Fn = nda::generate(const nda::array_ref<T, Shape>&, Generator&&) [with T = int; Shape = nda::shape<>; Generator = int (&)(); <template-parameter-1-4> = int]::<lambda(int&)>&; Ptrs = {std::pair<int*, std::tuple<> >}]’ 1361 | NDARRAY_INLINE NDARRAY_HOST_DEVICE void for_each_value_in_order( | ^~~~~~~~~~~~~~~~~~~~~~~ ./array.h:1369:41: note: candidate: ‘void nda::internal::for_each_value_in_order(const std::tuple<>&, Fn&&, Ptrs ...) [with long unsigned int D = 18446744073709551615; Fn = nda::generate(const nda::array_ref<T, Shape>&, Generator&&) [with T = int; Shape = nda::shape<>; Generator = int (&)(); <template-parameter-1-4> = int]::<lambda(int&)>&; Ptrs = {std::pair<int*, std::tuple<> >}]’ 1369 | NDARRAY_INLINE NDARRAY_HOST_DEVICE void for_each_value_in_order( | ^~~~~~~~~~~~~~~~~~~~~~~ ./array.h: In instantiation of ‘void nda::for_each_value_in_order(const Shape&, const ShapeA&, PtrA, const ShapeB&, PtrB, Fn&&) [with Shape = nda::shape<>; ShapeA = nda::shape<>; PtrA = const int*; ShapeB = nda::shape<>; PtrB = const int*; Fn = nda::array_ref<T, Shape>::operator!=(const nda::array_ref<T, Shape>&) const [with T = const int; Shape = nda::shape<>]::<lambda(nda::array_ref<const int, nda::shape<> >::const_reference, nda::array_ref<const int, nda::shape<> >::const_reference)>&; <template-parameter-1-7> = void]’: ./array.h:1796:28: required from ‘static void nda::copy_shape_traits<ShapeSrc, ShapeDst>::for_each_value(const ShapeSrc&, TSrc, const ShapeDst&, TDst, Fn&&) [with Fn = nda::array_ref<T, Shape>::operator!=(const nda::array_ref<T, Shape>&) const [with T = const int; Shape = nda::shape<>]::<lambda(nda::array_ref<const int, nda::shape<> >::const_reference, nda::array_ref<const int, nda::shape<> >::const_reference)>; TSrc = const int*; TDst = const int*; ShapeSrc = nda::shape<>; ShapeDst = nda::shape<>]’

[jiawen]

This is very hard to parse. What did you run as your top level command? Can you also say what platform you're on and what version of gcc you're using?

[dsharlet]

This just looks like make test, but yes, the gcc version would be helpful. I remember testing gcc for a while via Travis so it at least worked at some point with some version of gcc.

On Sun, Jul 10, 2022 at 4:07 PM Jiawen (Kevin) Chen < @.***> wrote:

This is very hard to parse. What did you run as your top level command? Can you also say what platform you're on and what version of gcc you're using?

— Reply to this email directly, view it on GitHub https://github.com/dsharlet/array/issues/67#issuecomment-1179818212, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAQUSAALDJKPOTXIETBNL43VTNJT3ANCNFSM523YNWUA . You are receiving this because you are subscribed to this thread.Message ID: @.***>

[dsharlet]

I was able to reproduce this on gcc 9.4 on Ubuntu. Unfortunately this seems like a pretty basic compiler bug and I'm not sure if there is a reasonable workaround. I think according to the standard, this is clearly not an ambiguous overload, and later versions of GCC don't have this problem (neither does clang).

[dsharlet]

I actually did find a workaround for this and other issues that old GCC versions struggled with. It was actually make_compact_dims that seemed harder to help with ambiguous overloads for, but then I realized that could be implemented without relying on overloading at all. #73 should fix this.