Allow @objc(Name) on enums

[lilyball]

This implements support for

@objc(ObjCEnum) enum SwiftEnum : Int {
    case One = 1, Two, Three
}

A new macro for the generated header is defined called SWIFT_ENUM_NAMED and the above declaration looks like

typedef SWIFT_ENUM_NAMED(NSInteger, ObjCEnum, "SwiftEnum") {
    ObjCEnumOne = 1,
    ObjCEnumTwo = 2,
    ObjCEnumThree = 3,
};

Since swift_name doesn't actually work properly yet when importing Obj-C enums, there's an XFAILed test that tests the clang importer here.

This does NOT cover support for @objc(Name) on the individual case declarations.

[lilyball]

I wasn't sure if this sort of test is the right way to do this, but I was using the test/ClangModules/enum-new.swift test as a prototype.

[jrose-apple]

Thanks for working on this!

[DougGregor]

FWIW, you should be able to add the flag -enable-swift-name-lookup-tables to this test to get it working now. The Swift name lookup tables aren't on by default just yet, but they address this problem already.

[lilyball]

I've updated the PR as per feedback. Tests are now part of the relevant commits, the enum constant prefix bit was pulled out into #743, and I've added relevant code to the resolve-cross-language test.

A couple of the tests have // FIXME comments for #743, because they use __attribute__((swift_name(...))) on enum constants in order to sidestep the enum constant prefix stripping. The tests will be unaffected by #743 but should ideally have the attributes removed once #743 lands.

[jrose-apple]

Looks great. Thank you!

I merged #743 just now. Do you want to fix these tests first or just do a follow-up?

[lilyball]

I'll fix the tests now.

[lilyball]

@jrose-apple Rebased and updated to remove the FIXMEs.

[jrose-apple]

Aren't these commits in the wrong order? Ideally every commit should be build-and-testable on its own, although I suppose that matters less in a merge.

[lilyball]

@jrose-apple GitHub bug, it's showing the commits in the wrong order. If you look at the branch they're ordered Sema, PrintObjC, ClangImporter. This bug can occur when rewriting commits, because it's sorting based on AuthorDate instead of actual topographical order or CommitDate.

[jrose-apple]

Weird, okay. Doing one more check on my machine, then I'll merge.

("It's not that I don't trust you…")

[lilyball]

Yeah, it's annoying. I filed a support ticket over a year ago about this, and they still haven't fixed it :/