The current factset storage rests on the assumption that a nontrivial subset of top-level facts (including any subtree) will never change (though see #3956 for a promising, potential relaxation/improvement there). Given that, benchmark should probably attempt to support that assumption when mutating factsets.
The current factset storage rests on the assumption that a nontrivial subset of top-level facts (including any subtree) will never change (though see #3956 for a promising, potential relaxation/improvement there). Given that, benchmark should probably attempt to support that assumption when mutating factsets.
See also: #3955