SI-1786 incorporate defined bounds in inference

Also fixes SI-5459.

Look, you don't have to redeclare the bounds, isn't it exciting?
For instance, there are eight places in JavaMirrors with this:

  jTypeVariable[_ <: GenericDeclaration]

After this code is in starr, those can look like this:

  jTypeVariable[_]

Since TypeVariable's definition looks like this:

  interface TypeVariable<D extends GenericDeclaration>

We already know that!

1 files changed, 48 insertions, 0 deletions

diff --git a/test/files/pos/t5459.scala b/test/files/pos/t5459.scala
new file mode 100644
index 0000000000..971e6f896d
--- /dev/null
+++ b/test/files/pos/t5459.scala
@@ -0,0 +1,48 @@
+trait A1
+trait A2
+trait A3
+trait L1 extends A1 with A2 with A3
+
+object Test {
+  trait T1[-A <: A1]
+  trait T2[-A >: L1]
+  trait T3[ A <: A1]
+  trait T4[ A >: L1]
+  trait T5[+A <: A1]
+  trait T6[+A >: L1]
+
+  def f1(x: T1[_]) = x
+  def f2(x: T2[_]) = x
+  def f3(x: T3[_]) = x
+  def f4(x: T4[_]) = x
+  def f5(x: T5[_]) = x
+  def f6(x: T6[_]) = x
+  // a.scala:22: error: type arguments [Any] do not conform to trait T5's type parameter bounds [+A <: A1]
+  //   def f5(x: T5[_]) = x
+  //       ^
+
+  def g1(x: T1[_ <: A1]) = x
+  def g2(x: T2[_ >: L1]) = x
+  def g3(x: T3[_ <: A1]) = x
+  def g4(x: T4[_ >: L1]) = x
+  def g5(x: T5[_ <: A1]) = x
+  def g6(x: T6[_ >: L1]) = x
+
+  def q1(x: T1[_ >: L1]) = x
+  def q2(x: T2[_ <: A1]) = x
+  def q3(x: T3[_ >: L1]) = x
+  def q4(x: T4[_ <: A1]) = x
+  def q5(x: T5[_ >: L1]) = x
+  def q6(x: T6[_ <: A1]) = x
+  // a.scala:41: error: type arguments [Any] do not conform to trait T5's type parameter bounds [+A <: A1]
+  //   def q5(x: T5[_ >: L1]) = x
+  //       ^
+  // two errors found
+
+  def h1(x: T1[_ >: L1 <: A1]) = x
+  def h2(x: T2[_ >: L1 <: A1]) = x
+  def h3(x: T3[_ >: L1 <: A1]) = x
+  def h4(x: T4[_ >: L1 <: A1]) = x
+  def h5(x: T5[_ >: L1 <: A1]) = x
+  def h6(x: T6[_ >: L1 <: A1]) = x
+}