Regression since 2.12.7: missing subtyping proofs involving singleton types

[LPTK]

The following works in 2.12.6, but fails in both 2.12.7 and 2.12.8:

class A extends B
class B
val a = new A
implicitly[a.type <:< a.type with B]

There is no reason it should not work, though. The a.type <:< a.type with B is true and recognized by the compiler, as the following demonstrates, which works in all versions:

conforms[a.type with B] : a.type <:< a.type with B

These subtyping relationships are sometimes useful; for a concrete example see: https://github.com/epfldata/squid/blob/7df295499b7371980b5574f87e0f56f2c20e3a5d/core/src/main/scala/squid/quasi/QuasiBase.scala#L133-L136

[SethTisue]

do we have a guess what scala/scala PR might have altered this behavior? perhaps we'll need to bisect.

@milessabin does this look like your territory...?

[milessabin]

It's possible. I'll take a look.

[milessabin]

@SethTisue I'm afraid I'm not going to have time to look at this for the next week or so.