For now, I am mostly interested in the high-level approach. Is this something we would consider appropriate for a CTS test? Is there a lighter-weight way to test these changes?

Of course, if the high-level approach looks OK, I would be interested in detailed comments also. Thanks!

Note, if anybody wants to build and run this test, I was not seeing the intended unloading behavior when I compiled the tests with gcc. Specifically, the call to dlclose did not seem to close the dynamic library "plugin". I haven't fully identified why this is happening, but I suspect it is due to:

https://stackoverflow.com/questions/78603238/reloading-a-dynamically-loaded-library-with-dlclose-dlopen-works-with-clang-on

This doesn't cause any sort of test failure, but it also limits what the test is testing. When I compiled the test with Clang I saw the intended unloading behavior.

I haven't tried any compilation on Windows just yet.

It would be interesting to understand why it happens here and not in the loader test where I verified the behavior, at least on linux.

Aside from that, the test looks good to me.

Discussed in the March 17th teleconference.

We're going to try to figure out why the unloading isn't happening for gcc before merging.

We're also going to do some testing on Windows.