Binary compatibility issue in 1.15.0

[larsrh]

Reported by @xuwei-k here: d4a3c25#r43785860

org.scalacheck.Gen.someOf

• 1.15.0 <T:Ljava/lang/Object;>(Lorg/scalacheck/Gen<TT;>;Lorg/scalacheck/Gen<TT;>;Lscala/collection/immutable/Seq<Lorg/scalacheck/Gen<TT;>;>;)Lorg/scalacheck/Gen<Lscala/collection/Seq<TT;>;>;

• 1.14.3 <T:Ljava/lang/Object;>(Lorg/scalacheck/Gen<TT;>;Lorg/scalacheck/Gen<TT;>;Lscala/collection/immutable/Seq<Lorg/scalacheck/Gen<TT;>;>;)Lorg/scalacheck/Gen<Lscala/collection/immutable/Seq<TT;>;>;

org.scalacheck.Gen.atLeastOne

• 1.15.0 <T:Ljava/lang/Object;>(Lorg/scalacheck/Gen<TT;>;Lorg/scalacheck/Gen<TT;>;Lscala/collection/immutable/Seq<Lorg/scalacheck/Gen<TT;>;>;)Lorg/scalacheck/Gen<Lscala/collection/Seq<TT;>;>;

• 1.14.3 <T:Ljava/lang/Object;>(Lorg/scalacheck/Gen<TT;>;Lorg/scalacheck/Gen<TT;>;Lscala/collection/immutable/Seq<Lorg/scalacheck/Gen<TT;>;>;)Lorg/scalacheck/Gen<Lscala/collection/immutable/Seq<TT;>;>;

[larsrh]

These changes come from @non (#603), could you please comment?

[ashawley]

Oscar raised his concern in a comment in #575, the original PR, and pointed out that type erasure means it's still binary compatible. So, this is only a source-compatibility breaking change?

[larsrh]

@johnynek Can you comment on this?

[johnynek]

1. I assume this project is running mima, if not it should. was that run? we should trust tools far more than humans in general to make these judgements.
2. to my eyes, it still appears the changed type is in the parameter of the result, and those are erased in JVM bytecode according to my understanding, so this appears binary (but not source) compatible to me. That said, maybe it is kind of bad kind of binary compatible because if you use the result it will result in a class cast exception... so maybe my argument is bad...

In 2.12 and earlier, I think immutable.Seq extends collection.Seq, but maybe that's not true in 2.13 so this differs a runtime exception...

If we are not running mima on all published versions, I think we need to update to do so, but even so, the version did bump to a new minor version, so maybe we just need to update the readme and say that there was a minor binary break in 2.13

[ashawley]

Mima is run on all versions, and even 2.13 after #482, but, yes, it only deals with binary compatible changes, IIUC.

[SethTisue]

At the moment I'm only glancing at this and not digging, but as per lightbend-labs/mima#299 shouldn't MiMa have warned on a generic-signature change?

[ashawley]

You're right, Seth. We disabled signature checking in 7bb0613. That's probably the issue.

[ashawley]

Actually, that wasn't the issue. Signature checking was enabled by default in 0.4.0, but then changed to opt-in in 0.7.0, see lightbend-labs/mima#471

[SethTisue]

@ashawley ah, good for you for finding that, I failed to remember it. (on 299, I added a comment with a link to 471, to aid future seekers.)

but even if we set mimaReportSignatureProblems to true here, won't that filter (the one in 7bb0613) still filter the problem out?

[ashawley]

That's right. Having both of these enabled didn't help. Turning them off, we do get the signature problems reported by MiMa.

[ashawley]

We can probably remove the return type signatures added for Gen.someOf and Gen.atLeastOne in 126168f that would at least make things compatible again with what was in 1.14.3. I believe this is because the types for scala.collection.Seq in 2.13 and later are different from 2.13 and earlier. Inferred types by the compiler is the only way to get it right. Otherwise, we'd probalby need to source-fork Gen.scala.

[johnynek]

You could just put those functions and type aliases in version dependent files and then call them from Gen.

[ashawley]

I predict Erik had an issue with lining up the types with Gen.pick which was that way because it had an explicit type signature of Gen[collection.Seq] that snuck in back during the 2.13 migration work. Unfortunately, we're probably stuck with that signature of Gen.pick.

[larsrh]

Are you running a release for this, @ashawley?

[ashawley]

No, I was not planning to do the 1.15.1 release. Would appreciate if you'd be willing to do it again. I'm not sure what others think, but it should be ready to go.

[larsrh]

OK, I will take care of it.

[larsrh]

I ran out of time today, but it's on my list for tomorrow.

[larsrh]

Done.