[BUG]: thrust::tuple<Eigen::Vector3f> copy Constructor fails in template dedcution

[miwechner]

Is this a duplicate?

• [x] I confirmed there appear to be no duplicate issues for this bug and that I agree to the Code of Conduct

Type of Bug

Compile-time Error

Component

Thrust

Describe the bug

Trying to call the copy constuctor on a thrust::tuple<Eigen::Vector3f> results in a compile time error. However, copy assignment strangely still works. Not sure whats so special about Eigen::Matrix types that it causes it to trigger a deduction with an iterator type. Using cuda::std::tuple instead of thrust::tuple also works.

Originally had the problem on MSVC 19.41 with CUDA 12.6 buried inside a call to thrust::sorty_by_key with a zip iterator as the Output Iterator for the values, but was also easily reproducible on godbolt.

How to Reproduce

See godbolt link

Expected behavior

thrust::tuple<Eigen::Vector3f> should be copy constructed

Reproduction link

https://godbolt.org/z/n6M7ET6jf

Operating System

No response

nvidia-smi output

No response

NVCC version

nvcc version 12.6.1

[bernhardmgruber]

Thank you for reporting this issue! We had a time earlier this year where we implemented thrust::tuple in terms of cuda::std::tuple, but had to keep a few workarounds in. This caused errors like the one you observed in other projects as well, so we eventually reverted the change.

However, replacing thrust::tuple completely by cuda::std::tuple eventually worked (only an alias remains in Thrust), and this error is gone now on the current main branch: https://godbolt.org/z/7qxcG818P (you can select CCCL trunk on compiler explorer). So I kindly ask you to either wait for the next CUDA Toolkit shipping the change, or using CCCL directly from GitHub.