Ambiguous implicit values

[ashawley]

Seth ran across an issue in the Scala build when upgrading to ScalaCheck 1.15.1 from 1.14.3 in scala/scala#9301.

Seems it fails to compile because of an ambiguous implicit:

[error] /home/jenkins/workspace/scala-2.13.x-validate-main@2/test/scalacheck/scala/collection/FloatFormatTest.scala:86:17: ambiguous implicit values:
[error]  both method buildableArrayList in object Buildable of type [T]org.scalacheck.util.Buildable[T,java.util.ArrayList[T]]
[error]  and method buildableSeq in object Buildable of type [T]org.scalacheck.util.Buildable[T,scala.collection.Seq[T]]
[error]  match expected type org.scalacheck.util.Buildable[String,C]
[error]     Gen.sequence(bogoparts).map(_.asScala.mkString)
[error]                 ^
[error] /home/jenkins/workspace/scala-2.13.x-validate-main@2/test/scalacheck/scala/collection/IntegralParseTest.scala:123:33: ambiguous implicit values:
[error]  both method buildableArrayList in object Buildable of type [T]org.scalacheck.util.Buildable[T,java.util.ArrayList[T]]
[error]  and method buildableSeq in object Buildable of type [T]org.scalacheck.util.Buildable[T,scala.collection.Seq[T]]
[error]  match expected type org.scalacheck.util.Buildable[Char,C]
[error]     val sequenced = Gen.sequence(listOfGens)
[error]                                 ^
[error] two errors found

[ashawley]

The only thing I can think of is the addition of buildableSeq as part of #575.

[SethTisue]

I worked around it as follows:

-    import scala.collection.JavaConverters._
-
-    Gen.sequence(bogoparts).map(_.asScala.mkString)
+    // type annotation shouldn't be necessary? see typelevel/scalacheck#721
+    Gen.sequence[List[String], String](bogoparts).map(_.mkString)

-    val sequenced = Gen.sequence(listOfGens)
-    sequenced.map(l => scala.collection.JavaConverters.asScalaBuffer(l).mkString)
+    // type annotation shouldn't be necessary? see typelevel/scalacheck#721
+    Gen.sequence[List[Char], Char](listOfGens).map(_.mkString)

I don't know why JavaConverters dropped out of the picture, but perhaps it's a clue? 🤷

[ashawley]

I presume the JavaConverters is just a consequence of the domain of the test.

Not a lot of people use Gen.sequence but it seems like we need to fix the ambiguity. I'm not sure if we need both buildables or not. Fixing it would be a binary compatible change. This would be to fix a bug, so that's ok?