This is strange. CodaFi (JIRA User), any insights?

@swift-ci create

This is the remnant of an ancient would-be feature that even extends to bridging the collection through Objective-C. It is converting the 'is' check into an as? check, but isn't including the comparator value so we're diagnosing it. I'm torn as to whether this can be removed altogether since it doesn't seem like it ever worked...

@DougGregor?

More concerning to me is that asking the runtime

[Dog]() is [Cat]

returns true despite a diagnosis to the contrary:

<REPL Input>:1:9: warning: cast from '[Dog]' to unrelated type '[Cat]' always fails
[Dog]() is [Cat]
~~~~~~~ ^  ~~~~~
// r1 : Bool = true

That last sounds like the problem with nil Optionals being freely convertible. @swiftix, @jckarter?

I checked that the cast is not eliminated by the optimizer. Thus is must be an issue with the runtime. @jckarter could you have a look?

At least to get the diagnostic QoI side of this #11441

A [Dog] is a [Cat] because arrays are covariant. The runtime is correct, but the warning is wrong. The errors from the original bug report might be remnants from when array covariance wasn't fully integrated with the runtime dynamic cast system, so weren't supported in pattern matches, but that shouldn't be a problem anymore, and we should accept the patterns.

The runtime issue is unrelated to the switch issue. I just factored the 'is' check the pattern effectively performs into its own distinct expression

[Dog]() is [Cat]

If covariance applied to this expression, we could derive a contradiction.