Allison Piper

No problem -- it might be a little while before I can dig into it, but it's on my radar 👍

Thrust will be deprecating device-side launch in the near future, so we won't be fixing this.

Actually, I'll leave this open for now. We may be able to keep things working here.

Eventually `thrust::complex` will just be a wrapper around libcu++'s complex implementation, which does provide constexpr constructors. This will happen sometime after Thrust 2.0. In the meantime, would switching to `cuda::std::complex`...

Can you share a bit more about the usecase here? Using `transform_iterator` with a functor that modifies the input sequence seems problematic, and I'm not sure this should be allowed.

The async algorithms are no longer being actively developed, as we've shifted focus towards developing a new async framework based on [senders + receivers](https://isocpp.org/files/papers/P2300R5.html). It's unlikely that the `thrust::async` algorithms...

Interesting -- I remember fixing this issue in 7e6f33b167b4f3b5eff01eb3ded96b3693f5a162, which was caught by [this test](https://github.com/NVIDIA/thrust/blob/bd63dd1664ba9d626f06c6b055a12a200e11cb82/testing/scan.cu#L515-L562). Unfortunately, that only tests a trivially small input and needs a much clearer failure behavior....

run tests

run tests

run tests