Thanks for the quick fix, Mike!
It looks like another test case will need to be prepared; your fix makes
the above test work reliably, but the code that the minimization is from is
still crashing about 10% of the time with spurious nils, even with c5d262f
(your most recent fix), so there's still a lurking bug that didn't show up
before b82fc3d (Bump table allocations retroactively if they grow later on.)
Kat