The logic here is a little confusing and the PR misses the subtlety, but I think we can make it work. The is_sta() check is not really about checking for the STA as much as it is checking whether a call context needs to be captured at all. Reversing then does not quite achieve the original intent, as C++/WinRT can operate in a limited way without COM being loaded (at least initially).

So I suspect you can get the desired behavior without perturbing unintended behavior simply by updating is_sta to include NA or by making is_mta mean anything that can be served by the thread pool e.g. MTA or no apartment.

For example, the PR breaks the following sample:

IAsyncAction b()
{
    co_await resume_background();
}

IAsyncAction a()
{
    co_await b();
}

int main()
{
    a().get();
}

Pushed some changes.

Looks good. Just checking - we've confirmed this actually works with the API in question?

Looks good. Just checking - we've confirmed this actually works with the API in question?

Correct, this change make the API in question work as expected.

And you have a clean pass from run_tests.cmd?

And you have a clean pass from run_tests.cmd?

D:\Dev\cppwinrt>run_tests.cmd x64 Release
===============================================================================
All tests passed (1972 assertions in 62 test cases)

===============================================================================
All tests passed (8 assertions in 2 test cases)

===============================================================================
All tests passed (11 assertions in 2 test cases)

===============================================================================
All tests passed (3775 assertions in 363 test cases)

===============================================================================
All tests passed (3 assertions in 1 test case)

===============================================================================
All tests passed (4 assertions in 1 test case)


D:\Dev\cppwinrt>