SI-7655 Enforce stable id patterns conform to scrutinee type #2742
Conversation
ghost
assigned 
This was only being done if the stable id pattern was a direct reference to a module. This commit distinguishes stable ID patterns with the recently added facility in TreeInfo. (Actually, this might be overkill in adapt and we could probably get away with just checking for `_: Ident | _: Select`.) It seems, however, from the quantity of changes required to bootstrap the distro under this regime, that this change overreaches. We might need to check if the spec is correct here. We might be able to keep some of the change (such as aliases for modules such as scala.Right) as an error, and move the rest under -Xlint.
|
I'll leave this open so just to get a build. But we can't really merge it as is, based on the quantity of changes needed to bootstrap under this interpretation of the spec. |
|
Okay, that was near enough to a green test run. Closing the PR pending discussion on the the best approach for backwards compatibility / convenience vs safety etc. |
| @@ -188,9 +188,9 @@ object Futures { | |||
| }) | |||
|
|
|||
| val partFuns = unsetFts.map((p: Pair[Int, Future[Any]]) => { | |||
| val FutCh = p._2.inputChannel | |||
| val FutCh = p._2.inputChannel.asInstanceOf[Channel[Any]] | |||
There was a problem hiding this comment.
doesn't p._2 : Future[Any] imply p._2.inputChannel: Channel[Any]?
abstract class Future[+T] extends Responder[T] with Function0[T] {
// ...
def inputChannel: InputChannel[T]
There was a problem hiding this comment.
You're right, that is a left-over from my first round of trying to fix the problem. The actual fix was to use guards.
|
It seems this is one of those occasions where we should change the spec to match the implementation. I expect the stricter version to be pretty painful to migrate to, and not sure about the benefit that it brings, as type information doesn't really tend to tell you much about what |
|
Yeah, I agree. The spec is too strict here to be useful. There is definitely a need for some static checking though. Maybe that will be under Xlint. See SI-7211 for example. I'll ask Martin for a clarification on this next time I see him. |
This was only being done if the stable id pattern was a direct
reference to a module.
This commit distinguishes stable ID patterns with the recently
added facility in TreeInfo. (Actually, this might be overkill
in adapt and we could probably get away with just checking for
_: Ident | _: Select.)Review by @adriaanm.
Opinions on the particular way to check for the stable ID pattern are welcome.