Allow @available(swift, ...) nested within @available(macOS, ...)

[jrose-apple]

Previously, we treated this as an attempt to widen the availability of a member beyond its context, but it's really a different axis of availability.

Also, warn about using @available on extensions without an OS, which we just completely ignore right now.

rdar://problem/32632327

[jrose-apple]

@swift-ci Please test

[jrose-apple]

@swift-ci Please test source compatibility

[Coeur]

What about a test for multiple attributes, like:

@available(macOS 10.12, *)
@available(swift 400)
extension DummyType {}

and

@available(swift 400)
@available(macOS 10.12, *)
extension DummyType {}

Should it warn?

[jrose-apple]

Ah, I see no reason why the existing code wouldn't work for that, but it's a good test to have! Thanks for the suggestion.

[jrose-apple]

@swift-ci Please test

[swift-ci]

Build failed
Swift Test OS X Platform
Git Sha - 75d107a5d47c9a08b554d3ef5b22e03c115bbcde

[swift-ci]

Build failed
Swift Test Linux Platform
Git Sha - 75d107a5d47c9a08b554d3ef5b22e03c115bbcde

[moiseev]

[jrose-apple]

Hm, it's the two failures that @davezarzycki was seeing earlier. Except I did rebase…

[jrose-apple]

@swift-ci Please test macOS

[davezarzycki]

Hi @jrose-apple – I may be grasping at straws here, but I'm seeing seemingly random but repeatable/inexplicable test time outs in other scenarios too. For example, @xedin made what a seemingly harmless change in #15134 that caused two simple/fast tests to start timing out on my Linux box. Do the timeouts in this pull request go away if you remove -swift-verion 3? That worked on my Linux box for #15134.

[xedin]

It seems like these failures are sporadic, I saw them happen couple of times on Linux CI even before my changes went in.

[jrose-apple]

I'm not sure that what you saw on Linux was related to the OpenCL and SceneKit tests on macOS, because it's always those two tests. I asked Mishal to take a look at the particular bot that's been running our tests—turns out it was one specific machine for at least my failure and one of yours—and he said he'd take it offline for now.

Linux tests have been bumping up against the timeouts for a while over the last few months. I wonder if there are one or two long-running tests there and/or if we've regressed on something, but no one's looked into it yet that I know of.

[DougGregor]

LGTM,