Fix CUDA 13.1 compile failure: use trailing return types for dependent name resolution

[Copilot]

CUDA 13.1's cudafe++ fails to match the leading return type ComputeGraphVar<XxxView<T>>& (written before the scope qualifier) against the declaration's ComputeGraphVar<VarType>& where VarType is a dependent type alias. This affects all four ComputeGraphVar partial specialization .inl files.

• Use trailing return types so VarType resolves within class scope:

// Before
template <typename T>
MUDA_INLINE ComputeGraphVar<VarView<T>>& ComputeGraphVar<VarView<T>>::operator=(const RWView& view)

// After
template <typename T>
MUDA_INLINE auto ComputeGraphVar<VarView<T>>::operator=(const RWView& view) -> ComputeGraphVar<VarType>&

• Applied to graph_var_view.inl, graph_buffer_view.inl, graph_buffer_2d_view.inl, graph_buffer_3d_view.inl

Original prompt

---

This section details on the original issue you should resolve

<issue_title>Compile failure about the dependent names resolve using CUDA 13.1</issue_title>
<issue_description>## What happened

Environment:

• GCC 15.2.1
• CUDA release 13.1, V13.1.115
• CMAKE 4.2.3

Compiling muda with the tools above:

mkdir build
cd build
cmkae -S .. -B .
make

nvcc reports:

/home/ll06/src/muda/src/muda/buffer/details/graph_var_view.inl(10): error: declaration is incompatible with "muda::ComputeGraphVar<muda::ComputeGraphVar<muda::VarView<T>>::VarType> &muda::ComputeGraphVar<muda::VarView<T>>::operator=(const muda::ComputeGraphVar<muda::VarView<T>>::RWView &)" (declared at line 48 of /home/ll06/src/muda/src/muda/buffer/graph_var_view.h)
  inline ComputeGraphVar<VarView<T>>& ComputeGraphVar<VarView<T>>::operator=(const RWView& view)
                                                                   ^

/home/ll06/src/muda/src/muda/buffer/details/graph_buffer_view.inl(10): error: declaration is incompatible with "muda::ComputeGraphVar<muda::ComputeGraphVar<muda::BufferView<T>>::VarType> &muda::ComputeGraphVar<muda::BufferView<T>>::operator=(const muda::ComputeGraphVar<muda::BufferView<T>>::RWView &)" (declared at line 51 of /home/ll06/src/muda/src/muda/buffer/graph_buffer_view.h)
  inline ComputeGraphVar<BufferView<T>>& ComputeGraphVar<BufferView<T>>::operator=(const RWView& view)
                                                                         ^

/home/ll06/src/muda/src/muda/buffer/details/graph_buffer_2d_view.inl(10): error: declaration is incompatible with "muda::ComputeGraphVar<muda::ComputeGraphVar<muda::Buffer2DView<T>>::VarType> &muda::ComputeGraphVar<muda::Buffer2DView<T>>::operator=(const muda::ComputeGraphVar<muda::Buffer2DView<T>>::RWView &)" (declared at line 51 of /home/ll06/src/muda/src/muda/buffer/graph_buffer_2d_view.h)
  inline ComputeGraphVar<Buffer2DView<T>>& ComputeGraphVar<Buffer2DView<T>>::operator=(const RWView& view)
                                                                             ^

/home/ll06/src/muda/src/muda/buffer/details/graph_buffer_3d_view.inl(10): error: declaration is incompatible with "muda::ComputeGraphVar<muda::ComputeGraphVar<muda::Buffer3DView<T>>::VarType> &muda::ComputeGraphVar<muda::Buffer3DView<T>>::operator=(const muda::ComputeGraphVar<muda::Buffer3DView<T>>::RWView &)" (declared at line 51 of /home/ll06/src/muda/src/muda/buffer/graph_buffer_3d_view.h)
  inline ComputeGraphVar<Buffer3DView<T>>& ComputeGraphVar<Buffer3DView<T>>::operator=(const RWView& view)

According to the test, the following files have the same problem:

• graph_buffer_2d_view.inl
• graph_buffer_3d_view.inl
• graph_buffer_view.inl
• graph_var_view.inl

Possible reason

Here is a snippet that could make nvcc report the same error:

// Common pattern of these file.
namespace muda {

template <bool IsConst, typename T> struct VarView {};


template <typename T>
using var_view = VarView<false, T>;

template <typename T>
class ComputeGraphVar {
public:
    using VarType = T;
    using RWView = T;

    ComputeGraphVar<VarType>& operator=(const RWView& view) {}
};

template <typename T>
class ComputeGraphVar<var_view<T>> {

using VarType = var_view<T>;
using RWView = var_view<T>;

    ComputeGraphVar<VarType>& operator=(const RWView& view);
};

template <typename U>
inline ComputeGraphVar<var_view<U>>&
ComputeGraphVar<var_view<U>>::operator=(const var_view<U>& view) {
    return *this;
}

}

int main() {
    muda::ComputeGraphVar<muda::var_view<int>> v;
    return 0;
}

Compile the snippet with command

$ nvcc -o exp exp.cu --verbose
#$ "/usr/bin"/g++ -D__CUDA_ARCH_LIST__=750 -E -x c++ -D__CUDACC__ -D__NVCC__  "-I/opt/cuda/bin/../targets/x86_64-linux/include"   "-isystem" "/opt/cuda/bin/../targets/x86_64-linux/include/cccl"    -D__CUDACC_VER_MAJOR__=13 -D__CUDACC_VER_MINOR__=1 -D__CUDACC_VER_BUILD__=115 -D__CUDA_API_VER_MAJOR__=13 -D__CUDA_API_VER_MINOR__=1 -D__NVCC_DIAG_PRAGMA_SUPPORT__=1 -D__CUDACC_DEVICE_ATOMIC_BUILTINS__=1 -include "cuda_runtime.h" -m64 "exp.cu" -o "/tmp/tmpxft_00002618_00000000-5_exp.cpp4.ii" 
#$ cudafe++ --c++17 --static-host-stub --device-hidden-visibility --gnu_version=150201 --display_error_number --orig_src_file_name "exp.cu" --orig_src_path_name "/home/ll06/src/muda/build/tmp/exp.cu" --allow_managed  --m64 --parse_templates --gen_c_file_name "/tmp/tmpxft_00002618_00000000-6_exp.cudafe1.cpp" --stub_file_name "tmpxft_00002618_00000000-6_exp.cudafe1.stub.c" --gen_module_id_file --module_id_file_name "/tmp/tmpxft_00002618_00000000-4_exp.module_id" "/tmp/tmpxft_00002618_00000000-5_exp.cpp4.ii" 
exp.cu(30): error: declaration is incompatible with "muda::ComputeGraphVar<muda::ComputeGraphVar<muda::var_view<T>>::VarType> &muda::ComputeGraphVar<muda::var_view<T>>::operator=(const muda::ComputeGraphVar<muda::var_view<T>>::RWView &)" (declared at line 25)
  ComputeGraphVar<var_view<U>>::operator=(const var_view<U>& view)...

</details>



<!-- START COPILOT CODING AGENT SUFFIX -->

- Fixes MuGdxy/muda#74

<!-- START COPILOT CODING AGENT TIPS -->
---

✨ Let Copilot coding agent [set things up for you](https://github.com/MuGdxy/muda/issues/new?title=✨+Set+up+Copilot+instructions&body=Configure%20instructions%20for%20this%20repository%20as%20documented%20in%20%5BBest%20practices%20for%20Copilot%20coding%20agent%20in%20your%20repository%5D%28https://gh.io/copilot-coding-agent-tips%29%2E%0A%0A%3COnboard%20this%20repo%3E&assignees=copilot) — coding agent works faster and does higher quality work when set up for your repo.