0:16:21hayleyAlso, that scheme is far too much of a batch operation to do CHANGE-CLASS.
0:25:04hayleyThat is, we don't want to have to walk the heap to replace the rack of one object. Taking the indirection and a the read barrier to handle CHANGE-CLASS incrementally, but be able to fix pointers eventually could work, but seems nasty itself.
0:46:34hayleyI found a paper on the issue in Smalltalk, which has similar perils due to the #become: method <https://dl.acm.org/doi/10.1145/2754169.2754186>
0:55:04hayleyThey require firing a read barrier before using "primitives", which would include EQ for us. They rely on forwarding objects always failing method dispatch, much like how obsolete instances fail method dispatch in SICL.