Fix for SI-6499, regression in type inference.

I can't do any better than a reproduced comment:

For some reason which is still a bit fuzzy, we must let Nothing
through as a lower bound despite the fact that Nothing is always
a lower bound. My current supposition is that the side-effecting
type constraint accumulation mechanism depends on these subtype
tests being performed to make forward progress when there are
mutally recursive type vars. See pos/t6367 and pos/t6499 for the
competing test cases.

1 files changed, 3 insertions, 0 deletions

diff --git a/test/files/pos/t6499.scala b/test/files/pos/t6499.scala
new file mode 100644
index 0000000000..db376572ee
--- /dev/null
+++ b/test/files/pos/t6499.scala
@@ -0,0 +1,3 @@
+object Test {
+  Map(): Map[_, Int] with Map[_, Int]
+}