> It says that alloc() is O(1) and thus real-time...but then later on says that advance() must be called k times per alloc in order to avoid exhausting the free space while still having work on the gray list. This makes it not quite real-time; either you call advance() k times inside alloc (which it recommends) in which case it isn't constant latency but variable on the number of reachable objects from the object you are popping from the work list, or you risk running out of heap and have an eventual loop of draining the grey list to flip.

The original description is for CONS cells, which have exactly 2 pointers; the notes on generalization later in TFA state that if you want to support arbitrary sized objects, advance() needs to be made incremental for larger objects.