tensorflow 装饰器 @tf.function 的奇怪行为

Question

有人可以解释这段代码的行为：

class Count:
    def __init__(self):
        self.count=0
    def increment(self):
        self.count+=1
        tf.print(self.count)

class SurCount:
    def __init__(self,count):
        self.count=count

    @tf.function 
    def step(self):
        self.count.increment()

    def start(self):
        for i in range(10):
            self.step()

count=Count()
surCount=SurCount(count)
surCount.start()

返回 1,1,1,1,1...（计数器不增加）。

下一个代码有效：它返回 1,2,3,...

class SurCount:
    def __init__(self,count):
        self.count=count

    @tf.function 
    def start(self):
        for i in range(10):
            self.count.increment()

count=Count()
surCount=SurCount(count)
surCount.start()