Yes. This makes sense. I would agree to Mockito replacements if:

1. Stubs implementations are declared in the test class.
2. Stub implementations are simple.

This will likely result in some duplication but that is an acceptable tradeoff.

Does someone made a study to use another mock library to see if there will be better performance ? (EasyMock/JMockit/...)

I didn't tried with other mock library. The proposed solution #2767 replaces mockito by stub and fake implementations, which leads to a 7% improvement of the build time (that is, much more than the 2% expected).

I didn't tried with other mock library. The proposed solution #2767 replaces mockito by stub and fake implementations, which leads to a 7% improvement of the build time (that is, much more than the 2% expected).

I understand the principle of using fake implementation. However, I think that if you have an interface with 25 methods, the test class will be poluate with a lot of fake implementation, with also a lot unuse code.

That is why I questioned the use of other library

I agree that readability is generally lower when using stub classes with a lot of methods... This may be improved by having specific stub classes that are shared between test cases, but @mpkorstanje wanted stubs implementations to be declared in the test class.

I didn't found a lot of mocking libraries performance comparisons, but according to mockito/mockito#3288, mockito and Easymock have more or less the same performance.