When I use the subclassing API of tensorflow2 ,some strange problem arise: some layers can be reused,some can`t,why

Question

In the following two examples,that are all use the subclassing to construct the same model.Some strange problem occur about reused layers.the one can`t reused all la