Support for Scala 2.12 (new trait encoding w/ default methods)

[dotta]

MiMa has intimate knowledge of how traits are encoded by the Scala compiler up to 2.11. Because the trait encoding changed in Scala 2.12, MiMa needs to be updated to correctly handle both Scala 2.11 (and previous) and Scala 2.12 (and future) binaries.

A good place to start is to figure out how the current test suite can be adapted to run against both Scala 2.11 and Scala 2.12 (when running the current test suite against the Scala 2.12, less errors will be reported because of the new trait encoding). My thought is that we could have two problems.txt file for each test, one that is used as the oracle when running MiMa in Scala 2.11 mode, and another one when running MiMa in Scala 2.12 mode. But, if you have a better idea, go for it. The bottom line is that I'd suggest to first start from updating the tests, as they will help you validate all changes you'll need to make in the MiMa code to support the new Scala 2.12 trait encoding. After updating the tests, I'd start by looking at ClassInfo implementation. For instance, the implementation of ClassInfo.concreteMethods is based on the current trait encoding, and it simply won't work as expected with binaries produced with Scala 2.12.

This ticket may subsume #114.

[szeiger]

@dotta Can you elaborate on what you think needs to be changed? I updated the tests for 2.12.0-RC1 in #123 and, as far as I can tell, the new test results are correct. There are some changes in the reports due to the new trait encoding (in particular, in many cases a trait is now seen as an interface because in 2.12 it is a pure interface) but they appear to be correct.

[dotta]

@szeiger Hey Stefan, sorry I didn't see your message. I don't have specifics in mind, but take this test as an example. I'd expect the test to report a problem when the code is compiled against Scala 2.11, while it should succeed against 2.12 (the reason why I believe it should succeed is because a trait concrete method is encoded into a Java8 default method, and hence it's a binary compatible evolution). Do you agree?

[szeiger]

I assumed this was because of forward compatibility checking, but apparently that's not tested in the functional-tests. I'll take a closer look at the bytecode to see if this test should succeed.

[dotta]

I don't follow your note about forward compatibility. The functional-tests are black-box testing MiMa's reporting functionality.

[szeiger]

Since #104 MiMa can test forwards and backwards, in which case the addition of a new trait method would still be an incompatible change (if the tests actually performed the checks both ways).

[dotta]

I see. There is no need to check both ways. In the functional-tests, we are testing MiMa reports binary incompatibilities as defined by the JVM. The fact that MiMa can be run to ensure both forward and backward binary compatibility is irrelevant, as the rules governing binary compatibility are the same, independently of what direction you are checking.

[szeiger]

I have added support for interface default methods to #123.

[dotta]

Great :)

[szeiger]

Fixed by #123