There are no hard promises about when or even whether GC will reclaim a particular piece of garbage, precisely to enable optimizations and necessary compromises. Application availability is often more important than immediately collecting every possible piece of garbage. The use of tagging, generations, arenas, etc. all allow the GC to use heuristics and apply different collection regimes to different pieces of garbage.
But, different implementations of the same language and different versions of those implementations can all do different things. GC rarely says exactly when something will be collected. Only that it will eventually.
Example: https://jsfiddle.net/8cej4tpk/2/
Firefox GCs after about 8 seconds, Safari GCs immmediately, Chrome never (probably not until there's pressure). And, you'll find different behavior if you go check different versions of those browsers.
This is essentially what Python-the-language does. It does not require reference counting, or mark-sweep, or any garbage collection at all. The language specification says at https://docs.python.org/3/reference/datamodel.html#objects-v... :
"Objects are never explicitly destroyed; however, when they become unreachable they may be garbage-collected. An implementation is allowed to postpone garbage collection or omit it altogether — it is a matter of implementation quality how garbage collection is implemented, as long as no objects are collected that are still reachable."
That subjective language specification is quite different than its objective implementation in a Python implementation, which appear to be what you refer to now.