Fix/#655 eta expansion

[odersky]

Review by @lrytz or @retronym. Based on #673, so only the second commit is new.

[lrytz]

typo to -> that

[lrytz]

lgtm

[retronym]

In the test case, what was the too-narrow type that needed widening? Sharing that sort of detail makes review easier (and comes in handy when looking back at the history of this code.)

[retronym]

I just traced this:

methType=TermRef(ThisType(TypeRef(ThisType(TypeRef(NoPrefix,<empty>)),Test$)),foo)
methType.show=(str: => String)(i: Int)Test.Result(Test.foo)
methType.widen.show=(str: => String)(i: Int)Test.Result

After looking at the surrounding code, I would suggest that this widening should be done by the caller to liftArgs. This would be consistent with the way that adaptInterpolated/adapt passes the widened type as the argument for the methType parameter of etaExpand.

[odersky]

It was a TermRef. Stupid mistake, easy to make.

On Fri, Jun 19, 2015 at 10:46 PM, Jason Zaugg notifications@github.com
wrote:

In the test case, what was the too-narrow type that needed widening?
Sharing that sort of detail makes review easier (and comes in handy when
looking back at the history of this code.)

—
Reply to this email directly or view it on GitHub
#674 (comment).

Martin Odersky
EPFL

[retronym]

I see. In scalac we don't have to widen, because tree's have a symbol member to bind references to defintions, whereas dotc represents this with a NamedType instead.

[odersky]

@retronym I don't think there's a fixed convention whether we do widen in the caller or callee. The way it is in EtaExpansion is DRYer. If we move to caller, we have to duplicate the operation in the two callers to liftArgs.

[retronym]

In that case, the paramaeter name of methType is misleading.

[odersky]

@retronym Good point.