In some cases, access to shared data need not be synchronized, if concurrent access cannot result in incorrect behavior. The simplest example is read-only data. Since such data is never mutated, all threads will always read the same values regardless when they access the data.

In rare cases, shared data that is mutated may not require synchronization. However, understanding when this is the case requires a deep knowledge of how the interpreter and underlying software and hardware work. Consider the following example:

items = []
  flag = []
  
  def consume():
      while not flag:
          pass
      print('items is', items)
  
  def produce():
      consumer = threading.Thread(target=consume, args=())
      consumer.start()
      for i in range(10):
          items.append(i)
      flag.append('go')
  
  produce()
  

Here, the producer thread adds items to items, while the consumer waits until flag is non-empty. When the producer finishes adding items, it adds an element to flag, allowing the consumer to proceed.

In most Python implementations, this example will work correctly. However, a common optimization in other compilers and interpreters, and even the hardware itself, is to reorder operations within a single thread that do not depend on each other for data. In such a system, the statement flag.append('go') may be moved before the loop, since neither depends on the other for data. In general, you should avoid code like this unless you are certain that the underlying system won't reorder the relevant operations.