Is capy::delay thread-safe?

[infotraining]

    capy::thread_pool thd_pool;
    capy::Executor auto ex = thd_pool.get_executor();
    // ex.context().make_service<capy::detail::timer_service>(); // this line commented out causes ThreadSanitizer warning
    std::latch done(no_of_tasks);

    auto delay_task = [&](size_t index) -> capy::task<void> {
        co_await capy::delay(10ms * index);
    };

    for (int i = 0; i < no_of_tasks; ++i)
    {
        capy::run_async(thd_pool.get_executor(), [&]() { done.count_down(); }, [&](std::exception_ptr) { done.count_down(); })(delay_task(i));
    }

    thd_pool.join();
    done.wait();

Tests using capy::delay run on a thread_pool{1}. When the pool contains more threads, ThreadSanitizer reports a warning.
Should the timer_service- or any other service used by the execution context - be created before running any asynchronous tasks, or is it safely and lazily created on demand in await_suspend() (in this case inside delay_awaitable with ts_ = &env->executor.context().use_service<detail::timer_service>())?

[mvandeberg]

I ran your example and did find a problem with >1 threads in the pool caused by a refactor I had done during development.

It has been fixed along with a proper multi-threaded test to catch a regression in the future in PR #230.

I was not, however, able to reproduce the Tsan warning locally. Perhaps the sanitizer was picking up on the same bug? Can you run Tsan on develop and let me know if the warning is still popping up? If so, please add the Tsan warning to the issue. If it appears to be fixed, please close.