By the way, this discussion brings up another problem. I wasn't able to
write a test for this (to address the below concerns) because the test
suite for thread-naming supplies some sort of auto-generated subject
for threads with empty subjects. So we can't test behavior for dealing
with empty subjects.

Unfortunately, I don't know the history of the test suite, or why this
auto-subject decisicion was made, so I don't feel comfortable just
changing the behavior.