This was a bug in 6.12.3.  I think the problem no longer occurs due to
the way sparks are treated as weak pointers, but it doesn't hurt to
test for tagged pointers anyway: better to do the test than have a
subtle invariant.