Sema: Prefer submodule imports for access-level-on-import diagnostics

[xymus]

Update the logic finding the import to point to in diagnostics to avoid relying on the implicit import of the top-level module when importing a clang submodule. A recent change avoids duplicating the implicit imports and broke the old logic that would only match the top-level implicit import. Instead let's rely on the imports of the submodules.

rdar://162151660

[xymus]

@swift-ci Please smoke test

[hnrklssn]

What happens if you have something like this, does the order matter then, or does it still find the correct submodule?

/--- Client_Clang_Submodules.swift

#if INVERT
private import ClangLib.Sub1
#endif

internal import ClangLib.Sub2

#if !INVERT
private import ClangLib.Sub1
#endif

public func dummyAPI(t2: SubType) {}

//--- module.modulemap

module ClangLib {
  header "ClangLib.h"

  explicit module Sub1 {
    header "Sub1.h"
  }

  explicit module Sub2 {
    header "Sub2.h"
  }
}

//--- ClangLib.h
struct ClangType {};

//--- Sub1.h
struct SubType {};

//--- Sub2.h
// no longer includes Sub1.h!

Just checking since now SubType is only accessible through the Sub1 import, but on the Swift side it's contained in ClangLib, which Sub2 also dumps its contents into.

Edit: also what about this case but with all imports being internal?

[xymus]

I would expect the same result even if it doesn't point to the right submodule (lookup wise). At this time access-level on imports looks at top-level modules only, so as far as it's concerned the different access-levels applied to submodules are transposed to the top module. This is something we may want to revisit in the future to properly track each submodule. We used to care more about the coarse grain dependencies but now we'd benefit from something more precise.

[xymus]

Updated the test for the newly required -verify-ignore-unrelated.

@swift-ci Please smoke test

[xymus]

@swift-ci Please smoke test