Make singleton typecase patterns and type checks consistent #7027
Conversation
|
|
||
| def main(args: Array[String]): Unit = { | ||
| isAType("a") | ||
| println(new String("a").isInstanceOf["a"]) |
There was a problem hiding this comment.
I argue in #6996 that this should return true.
There was a problem hiding this comment.
Regardless of the result of that discussion, I think we can merge this PR. It's unlikely that we want FirstTransform to accidentally mangle some patterns.
| toTypeTree(tree) | ||
| } else if (tree.name != nme.WILDCARD) { | ||
| // We want to constant-fold _some_ idents here - for instance, @switch needs to see literals in patterns. | ||
| // However, constant-foldable wildcards can occur in patterns, for instance as `case _: "a"`; |
There was a problem hiding this comment.
I don't understand why we need to handle case _: "a" specially but not case x: "a"
There was a problem hiding this comment.
Oh, I guess because the latter is desugared to case x @ (_: "a") ? This is all pretty subtle :/
There was a problem hiding this comment.
Argh, I see where the comment is hilariously unclear. Give me a second.
| // AFAIK, constant-foldable wildcard idents can only occur in patterns, for instance as `case _: "a"`. | ||
| // Constant-folding that would result in `case "a": "a"`, which changes the meaning of the pattern. | ||
| // Note that we _do_ want to constant-fold idents in patterns that _aren't_ wildcards - | ||
| // for example, @switch annotation needs to see inlined literals and not indirect references. |
There was a problem hiding this comment.
@smarter I think this comment is at the precise level of detail that we want. I don't think going into the subtle differences of Bind/Ident nodes in patterns makes sense - if we did that consequently in the whole compiler, then more than half of the documentation would be permanently outdated.
|
@smarter @AleksanderBG Can we get a resolution on this so that we can close or merge for the release? |
|
I made the resolution myself. LGTM! |
|
This will need more work, see my comment in #6996 (comment). |
Fixes #6996.
We accidentally constant-folded wildcards in typecase pattern which tested for singleton types, which made their behaviour inconsistent with normal type checks.