Add swift5_tests to MetadataSections.

[grynspan]

This PR adds a swift5_tests property to MetadataSections and populates it in the static constructors used on Linux and Windows. This section will be used by swift-testing to store test metadata for discovery at runtime (see swiftlang/swift-testing#210.)

This change is necessary because on Linux and Windows, section boundaries are not discoverable at runtime and so must be noted on a per-image basis at compile time. (This is also why MetadataSections and the static constructors are present in the first place.)

There are no compiler changes needed here. The compiler does not need special knowledge of the tests section emitted by swift-testing. The compiler does not itself need to emit anything into this section.

This change does not affect platforms that use the Mach-O binary format (i.e. Darwin) because on those platforms, there is API to find sections and segments at runtime.

[grynspan]

@swift-ci please smoke test

[grynspan]

That "ready for review" button is too easy to hit in the mobile app.

[compnerd]

Addition of this is fine. Where is the content emitted? The Windows case is particularly interesting as that requires specific section naming (the content itself must be emitted into the B group so that the boundaries are properly ordered).

[grynspan]

Addition of this is fine. Where is the content emitted? The Windows case is particularly interesting as that requires specific section naming (the content itself must be emitted into the B group so that the boundaries are properly ordered).

I believe it's going in the "C" section along with all the other emitted Swift section symbols. Mind helping me make sense of the macros in the SwiftRT-COFF source?

[grynspan]

Worth noting the plan for the ABI for this section is to emit data only, no executable code, so I think it should be fine to do the same as what other sections do?

[grynspan]

Re-reading the Windows section documentation here: I think we're fine?

[compnerd]

Addition of this is fine. Where is the content emitted? The Windows case is particularly interesting as that requires specific section naming (the content itself must be emitted into the B group so that the boundaries are properly ordered).

I believe it's going in the "C" section along with all the other emitted Swift section symbols. Mind helping me make sense of the macros in the SwiftRT-COFF source?

The section should be fine (sw5test). However, the linker table generation relies on grouping to get the start and end symbols ordered properly. The start symbol is emitted into sw5test$A and the end symbol into swtest$C. The content of the (unsorted) table is emitted into group B - sw5test$B. This allows the linker to sort the groups properly as they are sorted lexicographically. Since the emission is not in the compiler, I just wanted to cross-check that the table contents were emitted into the proper group.

[grynspan]

As of right now, the content of the section is not emitted by any component. swift-testing will (hopefully/eventually) start emitting data in a future PR.

So I think all is well.