Ronan Keryell

> [gcc.gnu.org/bugzilla/show_bug.cgi?id=96720](https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96720) Thanks for digging into this. Actually I was not able to trigger the bug by playing around https://godbolt.org/z/oM7v41 Perhaps it happens only withrange-v3 0-11-0 Version 0.11.0 "Thanks, ISO"...

So it looks like it is somehow solved in a more recent environment. Just wait for the next version to land into Debian/unstable and this is caught by Ubuntu? Honestly...

It is not obvious how to fix this because as the documentation pointed says: > -Wformat-security > If -Wformat is specified, also warn about uses of format functions that represent...

I have just modernized the code around the `printf`, but not enough to get rid of the `printf` completely unfortunately. There are already cleaner alternatives like https://github.com/gabime/spdlog so I do...

Yes. And perhaps it is possible to write a small script to update most of the current log API to something more modern and adjust manually the remainers? Or dive...

For the serializing part you might rely on FlatBuffer or Cap'n Proto if you want something efficient while not handling with pointer directly. But this might be overkill for this...

In ```c++ void finish_distribution(const uint64_t n, particle_t p[const n]) ``` you could just use `particle_t p[]` or if you are courageous `sycl::span` from SYCL 2020. ```c++ d = sycl::default_selector{}.select_device(); ```...

> Meaning that starting asynchronous work in the lambda and not waiting for it is undefined behavior. > And this is also the case for the host_task, these two are...

It may be related to another issue in the documentation configuration: it looks like the basic types (Core Library) are not "doxygened" on http://www.boost.org/doc/libs/1_61_0/libs/compute/doc/html/boost_compute/reference.html#boost_compute.reference.api_overview compared to http://boostorg.github.io/compute/boost_compute/reference.html#boost_compute.reference.api_overview

I think that in a real application the `compute::context::get_devices` should not be the critical path, so having a more correct version does not seem like a problem.