为什么即使在波动的修复之后，双重检查锁仍然不鼓励？

Question

Reference article on JMM

It is worth noting that broken techniques like double-checked locking are still broken under the new memory model, and that "fixing" double-checked locking was not one of the goals of the new memory model effort. (However, the new semantics of volatile allow one of the commonly proposed alternatives to double-checked locking to work correctly, although the technique is still discouraged. )

Related discussions:

• Why Double checked locking is 25% faster in Joshua Bloch Effective Java Example
• Why is double-checked locking broken in Java?

I can understand why it is broken without the volatile fix. but I don't know why it is still discouraged even after the fix. I noticed some disagreement in one of the discussions. but if it is really as the reference suggested ? what could be the reason this fixed version is still discouraged?

Answer 1

So first of all: with the new Java MM the double-checked locking idiom is not broken anymore. So you can safely use it in Java.

There are also situations where all other alternatives cannot be used (like enums or static initialization). For example, if you need access to a non-static reference in order to construct the singleton.

So it is safe to use as long as you have completely understood the idiom.