This test would fail before if conditional breakpoints weren't working correctly, and the nature of the test (spinning up 10 threads, etc) opens the door to raciness. This patch vastly simplifies the test, removes the need for relying on conditional expression evaluation, and as a result makes the correctness of the test vastly easier to reason about and reduces flakiness. This test also used to pass on Windows only when the inferior was compiled with a specific version of MSVC, even though I could verify from inside of LLDB that thread-specific breakpoints don't work. With this patch, the test now correctly fails no matter what version of MSVC the inferior was compiled with.
Jim: Since the test still fails for me even after this patch, can you verify that
it does not regress anything on OSX? I think it's a better test overall, so as
long as it doesn't regress anything anywhere I think it should probably go in.