This PR uses the experimental symbol linkage margers feature in the Swift
compiler to emit metadata about tests (and exit tests) into a dedicated section
of the test executable being built. At runtime, we discover that section and
read out the tests from it.

This has several benefits over our current model, which involves walking Swift's
type metadata table looking for types that conform to a protocol:

1. We don't need to define that protocol as public API in Swift Testing,
1. We don't need to emit type metadata (much larger than what we really need)
   for every test function,
1. We don't need to duplicate a large chunk of the Swift ABI sources in order to
   walk the type metadata table correctly, and
1. Almost all the new code is written in Swift, whereas the code it is intended
   to replace could not be fully represented in Swift and needed to be written
   in C++.

The change also opens up the possibility of supporting generic types in the
future because we can emit metadata without needing to emit a nested type (which
is not always valid in a generic context.) That's a "future direction" and not
covered by this PR specifically.

I've defined a layout for entries in the new `swift5_tests` section that should
be flexible enough for us in the short-to-medium term and which lets us define
additional arbitrary test content record types. The layout of this section is
covered in depth in the new [TestContent.md](Documentation/ABI/TestContent.md)
article.

This functionality is only available if a test target enables the experimental
`"SymbolLinkageMarkers"` feature. We continue to emit protocol-conforming types
for now—that code will be removed if and when the experimental feature is
properly supported (modulo us adopting relevant changes to the feature's API.)

#735
swiftlang/swift#76698
swiftlang/swift#78411

This PR leverages the change at swiftlang/swift#78411 to let us look up Swift's metadata sections (specifically `"swift5_tests"`) by looking for the emitted ELF notes rather than by calling the Swift runtime function `swift_enumerateAllMetadataSections()` (which we'd like to remove.)

Resolves #735.

This PR uses the experimental symbol linkage margers feature in the Swift
compiler to emit metadata about tests (and exit tests) into a dedicated section
of the test executable being built. At runtime, we discover that section and
read out the tests from it.

This has several benefits over our current model, which involves walking Swift's
type metadata table looking for types that conform to a protocol:

1. We don't need to define that protocol as public API in Swift Testing,
1. We don't need to emit type metadata (much larger than what we really need)
   for every test function,
1. We don't need to duplicate a large chunk of the Swift ABI sources in order to
   walk the type metadata table correctly, and
1. Almost all the new code is written in Swift, whereas the code it is intended
   to replace could not be fully represented in Swift and needed to be written
   in C++.

The change also opens up the possibility of supporting generic types in the
future because we can emit metadata without needing to emit a nested type (which
is not always valid in a generic context.) That's a "future direction" and not
covered by this PR specifically.

I've defined a layout for entries in the new `swift5_tests` section that should
be flexible enough for us in the short-to-medium term and which lets us define
additional arbitrary test content record types. The layout of this section is
covered in depth in the new [TestContent.md](Documentation/ABI/TestContent.md)
article.

This functionality is only available if a test target enables the experimental
`"SymbolLinkageMarkers"` feature. We continue to emit protocol-conforming types
for now—that code will be removed if and when the experimental feature is
properly supported (modulo us adopting relevant changes to the feature's API.)

#735
swiftlang/swift#76698
swiftlang/swift#78411

This PR uses the experimental symbol linkage margers feature in the Swift
compiler to emit metadata about tests (and exit tests) into a dedicated section
of the test executable being built. At runtime, we discover that section and
read out the tests from it.

This has several benefits over our current model, which involves walking Swift's
type metadata table looking for types that conform to a protocol:

1. We don't need to define that protocol as public API in Swift Testing,
1. We don't need to emit type metadata (much larger than what we really need)
   for every test function,
1. We don't need to duplicate a large chunk of the Swift ABI sources in order to
   walk the type metadata table correctly, and
1. Almost all the new code is written in Swift, whereas the code it is intended
   to replace could not be fully represented in Swift and needed to be written
   in C++.

The change also opens up the possibility of supporting generic types in the
future because we can emit metadata without needing to emit a nested type (which
is not always valid in a generic context.) That's a "future direction" and not
covered by this PR specifically.

I've defined a layout for entries in the new `swift5_tests` section that should
be flexible enough for us in the short-to-medium term and which lets us define
additional arbitrary test content record types. The layout of this section is
covered in depth in the new [TestContent.md](Documentation/ABI/TestContent.md)
article.

This functionality is only available if a test target enables the experimental
`"SymbolLinkageMarkers"` feature. We continue to emit protocol-conforming types
for now—that code will be removed if and when the experimental feature is
properly supported (modulo us adopting relevant changes to the feature's API.)

#735
swiftlang/swift#76698
swiftlang/swift#78411

This PR uses the experimental symbol linkage margers feature in the Swift
compiler to emit metadata about tests (and exit tests) into a dedicated section
of the test executable being built. At runtime, we discover that section and
read out the tests from it.

This has several benefits over our current model, which involves walking Swift's
type metadata table looking for types that conform to a protocol:

1. We don't need to define that protocol as public API in Swift Testing,
1. We don't need to emit type metadata (much larger than what we really need)
   for every test function,
1. We don't need to duplicate a large chunk of the Swift ABI sources in order to
   walk the type metadata table correctly, and
1. Almost all the new code is written in Swift, whereas the code it is intended
   to replace could not be fully represented in Swift and needed to be written
   in C++.

The change also opens up the possibility of supporting generic types in the
future because we can emit metadata without needing to emit a nested type (which
is not always valid in a generic context.) That's a "future direction" and not
covered by this PR specifically.

I've defined a layout for entries in the new `swift5_tests` section that should
be flexible enough for us in the short-to-medium term and which lets us define
additional arbitrary test content record types. The layout of this section is
covered in depth in the new [TestContent.md](Documentation/ABI/TestContent.md)
article.

This functionality is only available if a test target enables the experimental
`"SymbolLinkageMarkers"` feature. We continue to emit protocol-conforming types
for now—that code will be removed if and when the experimental feature is
properly supported (modulo us adopting relevant changes to the feature's API.)

#735
swiftlang/swift#76698
swiftlang/swift#78411

This PR uses the experimental symbol linkage margers feature in the Swift
compiler to emit metadata about tests (and exit tests) into a dedicated section
of the test executable being built. At runtime, we discover that section and
read out the tests from it.

This has several benefits over our current model, which involves walking Swift's
type metadata table looking for types that conform to a protocol:

1. We don't need to define that protocol as public API in Swift Testing,
1. We don't need to emit type metadata (much larger than what we really need)
   for every test function,
1. We don't need to duplicate a large chunk of the Swift ABI sources in order to
   walk the type metadata table correctly, and
1. Almost all the new code is written in Swift, whereas the code it is intended
   to replace could not be fully represented in Swift and needed to be written
   in C++.

The change also opens up the possibility of supporting generic types in the
future because we can emit metadata without needing to emit a nested type (which
is not always valid in a generic context.) That's a "future direction" and not
covered by this PR specifically.

I've defined a layout for entries in the new `swift5_tests` section that should
be flexible enough for us in the short-to-medium term and which lets us define
additional arbitrary test content record types. The layout of this section is
covered in depth in the new [TestContent.md](Documentation/ABI/TestContent.md)
article.

This functionality is only available if a test target enables the experimental
`"SymbolLinkageMarkers"` feature. We continue to emit protocol-conforming types
for now—that code will be removed if and when the experimental feature is
properly supported (modulo us adopting relevant changes to the feature's API.)

#735
swiftlang/swift#76698
swiftlang/swift#78411

This PR uses the experimental symbol linkage margers feature in the Swift
compiler to emit metadata about tests (and exit tests) into a dedicated section
of the test executable being built. At runtime, we discover that section and
read out the tests from it.

This has several benefits over our current model, which involves walking Swift's
type metadata table looking for types that conform to a protocol:

1. We don't need to define that protocol as public API in Swift Testing,
1. We don't need to emit type metadata (much larger than what we really need)
   for every test function,
1. We don't need to duplicate a large chunk of the Swift ABI sources in order to
   walk the type metadata table correctly, and
1. Almost all the new code is written in Swift, whereas the code it is intended
   to replace could not be fully represented in Swift and needed to be written
   in C++.

The change also opens up the possibility of supporting generic types in the
future because we can emit metadata without needing to emit a nested type (which
is not always valid in a generic context.) That's a "future direction" and not
covered by this PR specifically.

I've defined a layout for entries in the new `swift5_tests` section that should
be flexible enough for us in the short-to-medium term and which lets us define
additional arbitrary test content record types. The layout of this section is
covered in depth in the new [TestContent.md](Documentation/ABI/TestContent.md)
article.

This functionality is only available if a test target enables the experimental
`"SymbolLinkageMarkers"` feature. We continue to emit protocol-conforming types
for now—that code will be removed if and when the experimental feature is
properly supported (modulo us adopting relevant changes to the feature's API.)

#735
swiftlang/swift#76698
swiftlang/swift#78411

This PR uses the experimental symbol linkage margers feature in the Swift
compiler to emit metadata about tests (and exit tests) into a dedicated section
of the test executable being built. At runtime, we discover that section and
read out the tests from it.

This has several benefits over our current model, which involves walking Swift's
type metadata table looking for types that conform to a protocol:

1. We don't need to define that protocol as public API in Swift Testing,
1. We don't need to emit type metadata (much larger than what we really need)
   for every test function,
1. We don't need to duplicate a large chunk of the Swift ABI sources in order to
   walk the type metadata table correctly, and
1. Almost all the new code is written in Swift, whereas the code it is intended
   to replace could not be fully represented in Swift and needed to be written
   in C++.

The change also opens up the possibility of supporting generic types in the
future because we can emit metadata without needing to emit a nested type (which
is not always valid in a generic context.) That's a "future direction" and not
covered by this PR specifically.

I've defined a layout for entries in the new `swift5_tests` section that should
be flexible enough for us in the short-to-medium term and which lets us define
additional arbitrary test content record types. The layout of this section is
covered in depth in the new [TestContent.md](Documentation/ABI/TestContent.md)
article.

This functionality is only available if a test target enables the experimental
`"SymbolLinkageMarkers"` feature. We continue to emit protocol-conforming types
for now—that code will be removed if and when the experimental feature is
properly supported (modulo us adopting relevant changes to the feature's API.)

#735
swiftlang/swift#76698
swiftlang/swift#78411

This PR uses the experimental symbol linkage margers feature in the Swift
compiler to emit metadata about tests (and exit tests) into a dedicated section
of the test executable being built. At runtime, we discover that section and
read out the tests from it.

This has several benefits over our current model, which involves walking Swift's
type metadata table looking for types that conform to a protocol:

1. We don't need to define that protocol as public API in Swift Testing,
1. We don't need to emit type metadata (much larger than what we really need)
   for every test function,
1. We don't need to duplicate a large chunk of the Swift ABI sources in order to
   walk the type metadata table correctly, and
1. Almost all the new code is written in Swift, whereas the code it is intended
   to replace could not be fully represented in Swift and needed to be written
   in C++.

The change also opens up the possibility of supporting generic types in the
future because we can emit metadata without needing to emit a nested type (which
is not always valid in a generic context.) That's a "future direction" and not
covered by this PR specifically.

I've defined a layout for entries in the new `swift5_tests` section that should
be flexible enough for us in the short-to-medium term and which lets us define
additional arbitrary test content record types. The layout of this section is
covered in depth in the new [TestContent.md](Documentation/ABI/TestContent.md)
article.

This functionality is only available if a test target enables the experimental
`"SymbolLinkageMarkers"` feature. We continue to emit protocol-conforming types
for now—that code will be removed if and when the experimental feature is
properly supported (modulo us adopting relevant changes to the feature's API.)

#735
swiftlang/swift#76698
swiftlang/swift#78411

This PR uses the experimental symbol linkage margers feature in the Swift
compiler to emit metadata about tests (and exit tests) into a dedicated section
of the test executable being built. At runtime, we discover that section and
read out the tests from it.

This has several benefits over our current model, which involves walking Swift's
type metadata table looking for types that conform to a protocol:

1. We don't need to define that protocol as public API in Swift Testing,
1. We don't need to emit type metadata (much larger than what we really need)
   for every test function,
1. We don't need to duplicate a large chunk of the Swift ABI sources in order to
   walk the type metadata table correctly, and
1. Almost all the new code is written in Swift, whereas the code it is intended
   to replace could not be fully represented in Swift and needed to be written
   in C++.

The change also opens up the possibility of supporting generic types in the
future because we can emit metadata without needing to emit a nested type (which
is not always valid in a generic context.) That's a "future direction" and not
covered by this PR specifically.

I've defined a layout for entries in the new `swift5_tests` section that should
be flexible enough for us in the short-to-medium term and which lets us define
additional arbitrary test content record types. The layout of this section is
covered in depth in the new [TestContent.md](Documentation/ABI/TestContent.md)
article.

This functionality is only available if a test target enables the experimental
`"SymbolLinkageMarkers"` feature. We continue to emit protocol-conforming types
for now—that code will be removed if and when the experimental feature is
properly supported (modulo us adopting relevant changes to the feature's API.)

#735
swiftlang/swift#76698
swiftlang/swift#78411

This PR uses the experimental symbol linkage margers feature in the Swift
compiler to emit metadata about tests (and exit tests) into a dedicated section
of the test executable being built. At runtime, we discover that section and
read out the tests from it.

This has several benefits over our current model, which involves walking Swift's
type metadata table looking for types that conform to a protocol:

1. We don't need to define that protocol as public API in Swift Testing,
1. We don't need to emit type metadata (much larger than what we really need)
   for every test function,
1. We don't need to duplicate a large chunk of the Swift ABI sources in order to
   walk the type metadata table correctly, and
1. Almost all the new code is written in Swift, whereas the code it is intended
   to replace could not be fully represented in Swift and needed to be written
   in C++.

The change also opens up the possibility of supporting generic types in the
future because we can emit metadata without needing to emit a nested type (which
is not always valid in a generic context.) That's a "future direction" and not
covered by this PR specifically.

I've defined a layout for entries in the new `swift5_tests` section that should
be flexible enough for us in the short-to-medium term and which lets us define
additional arbitrary test content record types. The layout of this section is
covered in depth in the new [TestContent.md](Documentation/ABI/TestContent.md)
article.

This functionality is only available if a test target enables the experimental
`"SymbolLinkageMarkers"` feature. We continue to emit protocol-conforming types
for now—that code will be removed if and when the experimental feature is
properly supported (modulo us adopting relevant changes to the feature's API.)

#735
swiftlang/swift#76698
swiftlang/swift#78411

This PR uses the experimental symbol linkage margers feature in the Swift
compiler to emit metadata about tests (and exit tests) into a dedicated section
of the test executable being built. At runtime, we discover that section and
read out the tests from it.

This has several benefits over our current model, which involves walking Swift's
type metadata table looking for types that conform to a protocol:

1. We don't need to define that protocol as public API in Swift Testing,
1. We don't need to emit type metadata (much larger than what we really need)
   for every test function,
1. We don't need to duplicate a large chunk of the Swift ABI sources in order to
   walk the type metadata table correctly, and
1. Almost all the new code is written in Swift, whereas the code it is intended
   to replace could not be fully represented in Swift and needed to be written
   in C++.

The change also opens up the possibility of supporting generic types in the
future because we can emit metadata without needing to emit a nested type (which
is not always valid in a generic context.) That's a "future direction" and not
covered by this PR specifically.

I've defined a layout for entries in the new `swift5_tests` section that should
be flexible enough for us in the short-to-medium term and which lets us define
additional arbitrary test content record types. The layout of this section is
covered in depth in the new [TestContent.md](Documentation/ABI/TestContent.md)
article.

This functionality is only available if a test target enables the experimental
`"SymbolLinkageMarkers"` feature. We continue to emit protocol-conforming types
for now—that code will be removed if and when the experimental feature is
properly supported (modulo us adopting relevant changes to the feature's API.)

#735
swiftlang/swift#76698
swiftlang/swift#78411

This PR uses the experimental symbol linkage margers feature in the Swift
compiler to emit metadata about tests (and exit tests) into a dedicated section
of the test executable being built. At runtime, we discover that section and
read out the tests from it.

This has several benefits over our current model, which involves walking Swift's
type metadata table looking for types that conform to a protocol:

1. We don't need to define that protocol as public API in Swift Testing,
1. We don't need to emit type metadata (much larger than what we really need)
   for every test function,
1. We don't need to duplicate a large chunk of the Swift ABI sources in order to
   walk the type metadata table correctly, and
1. Almost all the new code is written in Swift, whereas the code it is intended
   to replace could not be fully represented in Swift and needed to be written
   in C++.

The change also opens up the possibility of supporting generic types in the
future because we can emit metadata without needing to emit a nested type (which
is not always valid in a generic context.) That's a "future direction" and not
covered by this PR specifically.

I've defined a layout for entries in the new `swift5_tests` section that should
be flexible enough for us in the short-to-medium term and which lets us define
additional arbitrary test content record types. The layout of this section is
covered in depth in the new [TestContent.md](Documentation/ABI/TestContent.md)
article.

This functionality is only available if a test target enables the experimental
`"SymbolLinkageMarkers"` feature. We continue to emit protocol-conforming types
for now—that code will be removed if and when the experimental feature is
properly supported (modulo us adopting relevant changes to the feature's API.)

#735
swiftlang/swift#76698
swiftlang/swift#78411

This PR uses the experimental symbol linkage margers feature in the Swift
compiler to emit metadata about tests (and exit tests) into a dedicated section
of the test executable being built. At runtime, we discover that section and
read out the tests from it.

This has several benefits over our current model, which involves walking Swift's
type metadata table looking for types that conform to a protocol:

1. We don't need to define that protocol as public API in Swift Testing,
1. We don't need to emit type metadata (much larger than what we really need)
   for every test function,
1. We don't need to duplicate a large chunk of the Swift ABI sources in order to
   walk the type metadata table correctly, and
1. Almost all the new code is written in Swift, whereas the code it is intended
   to replace could not be fully represented in Swift and needed to be written
   in C++.

The change also opens up the possibility of supporting generic types in the
future because we can emit metadata without needing to emit a nested type (which
is not always valid in a generic context.) That's a "future direction" and not
covered by this PR specifically.

I've defined a layout for entries in the new `swift5_tests` section that should
be flexible enough for us in the short-to-medium term and which lets us define
additional arbitrary test content record types. The layout of this section is
covered in depth in the new [TestContent.md](Documentation/ABI/TestContent.md)
article.

This functionality is only available if a test target enables the experimental
`"SymbolLinkageMarkers"` feature. We continue to emit protocol-conforming types
for now—that code will be removed if and when the experimental feature is
properly supported (modulo us adopting relevant changes to the feature's API.)

#735
swiftlang/swift#76698
swiftlang/swift#78411

This PR uses the experimental symbol linkage margers feature in the Swift
compiler to emit metadata about tests (and exit tests) into a dedicated section
of the test executable being built. At runtime, we discover that section and
read out the tests from it.

This has several benefits over our current model, which involves walking Swift's
type metadata table looking for types that conform to a protocol:

1. We don't need to define that protocol as public API in Swift Testing,
1. We don't need to emit type metadata (much larger than what we really need)
   for every test function,
1. We don't need to duplicate a large chunk of the Swift ABI sources in order to
   walk the type metadata table correctly, and
1. Almost all the new code is written in Swift, whereas the code it is intended
   to replace could not be fully represented in Swift and needed to be written
   in C++.

The change also opens up the possibility of supporting generic types in the
future because we can emit metadata without needing to emit a nested type (which
is not always valid in a generic context.) That's a "future direction" and not
covered by this PR specifically.

I've defined a layout for entries in the new `swift5_tests` section that should
be flexible enough for us in the short-to-medium term and which lets us define
additional arbitrary test content record types. The layout of this section is
covered in depth in the new [TestContent.md](Documentation/ABI/TestContent.md)
article.

This functionality is only available if a test target enables the experimental
`"SymbolLinkageMarkers"` feature. We continue to emit protocol-conforming types
for now—that code will be removed if and when the experimental feature is
properly supported (modulo us adopting relevant changes to the feature's API.)

#735
swiftlang/swift#76698
swiftlang/swift#78411

This PR uses the experimental symbol linkage margers feature in the Swift
compiler to emit metadata about tests (and exit tests) into a dedicated section
of the test executable being built. At runtime, we discover that section and
read out the tests from it.

This has several benefits over our current model, which involves walking Swift's
type metadata table looking for types that conform to a protocol:

1. We don't need to define that protocol as public API in Swift Testing,
1. We don't need to emit type metadata (much larger than what we really need)
   for every test function,
1. We don't need to duplicate a large chunk of the Swift ABI sources in order to
   walk the type metadata table correctly, and
1. Almost all the new code is written in Swift, whereas the code it is intended
   to replace could not be fully represented in Swift and needed to be written
   in C++.

The change also opens up the possibility of supporting generic types in the
future because we can emit metadata without needing to emit a nested type (which
is not always valid in a generic context.) That's a "future direction" and not
covered by this PR specifically.

I've defined a layout for entries in the new `swift5_tests` section that should
be flexible enough for us in the short-to-medium term and which lets us define
additional arbitrary test content record types. The layout of this section is
covered in depth in the new [TestContent.md](Documentation/ABI/TestContent.md)
article.

This functionality is only available if a test target enables the experimental
`"SymbolLinkageMarkers"` feature. We continue to emit protocol-conforming types
for now—that code will be removed if and when the experimental feature is
properly supported (modulo us adopting relevant changes to the feature's API.)

#735
swiftlang/swift#76698
swiftlang/swift#78411

This PR uses the experimental symbol linkage margers feature in the Swift
compiler to emit metadata about tests (and exit tests) into a dedicated section
of the test executable being built. At runtime, we discover that section and
read out the tests from it.

This has several benefits over our current model, which involves walking Swift's
type metadata table looking for types that conform to a protocol:

1. We don't need to define that protocol as public API in Swift Testing,
1. We don't need to emit type metadata (much larger than what we really need)
   for every test function,
1. We don't need to duplicate a large chunk of the Swift ABI sources in order to
   walk the type metadata table correctly, and
1. Almost all the new code is written in Swift, whereas the code it is intended
   to replace could not be fully represented in Swift and needed to be written
   in C++.

The change also opens up the possibility of supporting generic types in the
future because we can emit metadata without needing to emit a nested type (which
is not always valid in a generic context.) That's a "future direction" and not
covered by this PR specifically.

I've defined a layout for entries in the new `swift5_tests` section that should
be flexible enough for us in the short-to-medium term and which lets us define
additional arbitrary test content record types. The layout of this section is
covered in depth in the new [TestContent.md](Documentation/ABI/TestContent.md)
article.

This functionality is only available if a test target enables the experimental
`"SymbolLinkageMarkers"` feature. We continue to emit protocol-conforming types
for now—that code will be removed if and when the experimental feature is
properly supported (modulo us adopting relevant changes to the feature's API.)

#735
swiftlang/swift#76698
swiftlang/swift#78411

This PR uses the experimental symbol linkage margers feature in the Swift
compiler to emit metadata about tests (and exit tests) into a dedicated section
of the test executable being built. At runtime, we discover that section and
read out the tests from it.

This has several benefits over our current model, which involves walking Swift's
type metadata table looking for types that conform to a protocol:

1. We don't need to define that protocol as public API in Swift Testing,
1. We don't need to emit type metadata (much larger than what we really need)
   for every test function,
1. We don't need to duplicate a large chunk of the Swift ABI sources in order to
   walk the type metadata table correctly, and
1. Almost all the new code is written in Swift, whereas the code it is intended
   to replace could not be fully represented in Swift and needed to be written
   in C++.

The change also opens up the possibility of supporting generic types in the
future because we can emit metadata without needing to emit a nested type (which
is not always valid in a generic context.) That's a "future direction" and not
covered by this PR specifically.

I've defined a layout for entries in the new `swift5_tests` section that should
be flexible enough for us in the short-to-medium term and which lets us define
additional arbitrary test content record types. The layout of this section is
covered in depth in the new [TestContent.md](Documentation/ABI/TestContent.md)
article.

This functionality is only available if a test target enables the experimental
`"SymbolLinkageMarkers"` feature. We continue to emit protocol-conforming types
for now—that code will be removed if and when the experimental feature is
properly supported (modulo us adopting relevant changes to the feature's API.)

#735
swiftlang/swift#76698
swiftlang/swift#78411

This PR uses the experimental symbol linkage margers feature in the Swift
compiler to emit metadata about tests (and exit tests) into a dedicated section
of the test executable being built. At runtime, we discover that section and
read out the tests from it.

This has several benefits over our current model, which involves walking Swift's
type metadata table looking for types that conform to a protocol:

1. We don't need to define that protocol as public API in Swift Testing,
1. We don't need to emit type metadata (much larger than what we really need)
   for every test function,
1. We don't need to duplicate a large chunk of the Swift ABI sources in order to
   walk the type metadata table correctly, and
1. Almost all the new code is written in Swift, whereas the code it is intended
   to replace could not be fully represented in Swift and needed to be written
   in C++.

The change also opens up the possibility of supporting generic types in the
future because we can emit metadata without needing to emit a nested type (which
is not always valid in a generic context.) That's a "future direction" and not
covered by this PR specifically.

I've defined a layout for entries in the new `swift5_tests` section that should
be flexible enough for us in the short-to-medium term and which lets us define
additional arbitrary test content record types. The layout of this section is
covered in depth in the new [TestContent.md](Documentation/ABI/TestContent.md)
article.

This functionality is only available if a test target enables the experimental
`"SymbolLinkageMarkers"` feature. We continue to emit protocol-conforming types
for now—that code will be removed if and when the experimental feature is
properly supported (modulo us adopting relevant changes to the feature's API.)

#735
swiftlang/swift#76698
swiftlang/swift#78411

This PR uses the experimental symbol linkage margers feature in the Swift
compiler to emit metadata about tests (and exit tests) into a dedicated section
of the test executable being built. At runtime, we discover that section and
read out the tests from it.

This has several benefits over our current model, which involves walking Swift's
type metadata table looking for types that conform to a protocol:

1. We don't need to define that protocol as public API in Swift Testing,
1. We don't need to emit type metadata (much larger than what we really need)
   for every test function,
1. We don't need to duplicate a large chunk of the Swift ABI sources in order to
   walk the type metadata table correctly, and
1. Almost all the new code is written in Swift, whereas the code it is intended
   to replace could not be fully represented in Swift and needed to be written
   in C++.

The change also opens up the possibility of supporting generic types in the
future because we can emit metadata without needing to emit a nested type (which
is not always valid in a generic context.) That's a "future direction" and not
covered by this PR specifically.

I've defined a layout for entries in the new `swift5_tests` section that should
be flexible enough for us in the short-to-medium term and which lets us define
additional arbitrary test content record types. The layout of this section is
covered in depth in the new [TestContent.md](Documentation/ABI/TestContent.md)
article.

This functionality is only available if a test target enables the experimental
`"SymbolLinkageMarkers"` feature. We continue to emit protocol-conforming types
for now—that code will be removed if and when the experimental feature is
properly supported (modulo us adopting relevant changes to the feature's API.)

#735
swiftlang/swift#76698
swiftlang/swift#78411

This PR uses the experimental symbol linkage margers feature in the Swift
compiler to emit metadata about tests (and exit tests) into a dedicated section
of the test executable being built. At runtime, we discover that section and
read out the tests from it.

This has several benefits over our current model, which involves walking Swift's
type metadata table looking for types that conform to a protocol:

1. We don't need to define that protocol as public API in Swift Testing,
1. We don't need to emit type metadata (much larger than what we really need)
   for every test function,
1. We don't need to duplicate a large chunk of the Swift ABI sources in order to
   walk the type metadata table correctly, and
1. Almost all the new code is written in Swift, whereas the code it is intended
   to replace could not be fully represented in Swift and needed to be written
   in C++.

The change also opens up the possibility of supporting generic types in the
future because we can emit metadata without needing to emit a nested type (which
is not always valid in a generic context.) That's a "future direction" and not
covered by this PR specifically.

I've defined a layout for entries in the new `swift5_tests` section that should
be flexible enough for us in the short-to-medium term and which lets us define
additional arbitrary test content record types. The layout of this section is
covered in depth in the new [TestContent.md](Documentation/ABI/TestContent.md)
article.

This functionality is only available if a test target enables the experimental
`"SymbolLinkageMarkers"` feature. We continue to emit protocol-conforming types
for now—that code will be removed if and when the experimental feature is
properly supported (modulo us adopting relevant changes to the feature's API.)

#735
swiftlang/swift#76698
swiftlang/swift#78411

This PR uses the experimental symbol linkage margers feature in the Swift
compiler to emit metadata about tests (and exit tests) into a dedicated section
of the test executable being built. At runtime, we discover that section and
read out the tests from it.

This has several benefits over our current model, which involves walking Swift's
type metadata table looking for types that conform to a protocol:

1. We don't need to define that protocol as public API in Swift Testing,
1. We don't need to emit type metadata (much larger than what we really need)
   for every test function,
1. We don't need to duplicate a large chunk of the Swift ABI sources in order to
   walk the type metadata table correctly, and
1. Almost all the new code is written in Swift, whereas the code it is intended
   to replace could not be fully represented in Swift and needed to be written
   in C++.

The change also opens up the possibility of supporting generic types in the
future because we can emit metadata without needing to emit a nested type (which
is not always valid in a generic context.) That's a "future direction" and not
covered by this PR specifically.

I've defined a layout for entries in the new `swift5_tests` section that should
be flexible enough for us in the short-to-medium term and which lets us define
additional arbitrary test content record types. The layout of this section is
covered in depth in the new [TestContent.md](Documentation/ABI/TestContent.md)
article.

This functionality is only available if a test target enables the experimental
`"SymbolLinkageMarkers"` feature. We continue to emit protocol-conforming types
for now—that code will be removed if and when the experimental feature is
properly supported (modulo us adopting relevant changes to the feature's API.)

#735
swiftlang/swift#76698
swiftlang/swift#78411

This PR uses the experimental symbol linkage margers feature in the Swift
compiler to emit metadata about tests (and exit tests) into a dedicated section
of the test executable being built. At runtime, we discover that section and
read out the tests from it.

This has several benefits over our current model, which involves walking Swift's
type metadata table looking for types that conform to a protocol:

1. We don't need to define that protocol as public API in Swift Testing,
1. We don't need to emit type metadata (much larger than what we really need)
   for every test function,
1. We don't need to duplicate a large chunk of the Swift ABI sources in order to
   walk the type metadata table correctly, and
1. Almost all the new code is written in Swift, whereas the code it is intended
   to replace could not be fully represented in Swift and needed to be written
   in C++.

The change also opens up the possibility of supporting generic types in the
future because we can emit metadata without needing to emit a nested type (which
is not always valid in a generic context.) That's a "future direction" and not
covered by this PR specifically.

I've defined a layout for entries in the new `swift5_tests` section that should
be flexible enough for us in the short-to-medium term and which lets us define
additional arbitrary test content record types. The layout of this section is
covered in depth in the new [TestContent.md](Documentation/ABI/TestContent.md)
article.

This functionality is only available if a test target enables the experimental
`"SymbolLinkageMarkers"` feature. We continue to emit protocol-conforming types
for now—that code will be removed if and when the experimental feature is
properly supported (modulo us adopting relevant changes to the feature's API.)

#735
swiftlang/swift#76698
swiftlang/swift#78411

This PR uses the experimental symbol linkage margers feature in the Swift
compiler to emit metadata about tests (and exit tests) into a dedicated section
of the test executable being built. At runtime, we discover that section and
read out the tests from it.

This has several benefits over our current model, which involves walking Swift's
type metadata table looking for types that conform to a protocol:

1. We don't need to define that protocol as public API in Swift Testing,
1. We don't need to emit type metadata (much larger than what we really need)
   for every test function,
1. We don't need to duplicate a large chunk of the Swift ABI sources in order to
   walk the type metadata table correctly, and
1. Almost all the new code is written in Swift, whereas the code it is intended
   to replace could not be fully represented in Swift and needed to be written
   in C++.

The change also opens up the possibility of supporting generic types in the
future because we can emit metadata without needing to emit a nested type (which
is not always valid in a generic context.) That's a "future direction" and not
covered by this PR specifically.

I've defined a layout for entries in the new `swift5_tests` section that should
be flexible enough for us in the short-to-medium term and which lets us define
additional arbitrary test content record types. The layout of this section is
covered in depth in the new [TestContent.md](Documentation/ABI/TestContent.md)
article.

This functionality is only available if a test target enables the experimental
`"SymbolLinkageMarkers"` feature. We continue to emit protocol-conforming types
for now—that code will be removed if and when the experimental feature is
properly supported (modulo us adopting relevant changes to the feature's API.)

#735
swiftlang/swift#76698
swiftlang/swift#78411

This PR uses the experimental symbol linkage margers feature in the Swift
compiler to emit metadata about tests (and exit tests) into a dedicated section
of the test executable being built. At runtime, we discover that section and
read out the tests from it.

This has several benefits over our current model, which involves walking Swift's
type metadata table looking for types that conform to a protocol:

1. We don't need to define that protocol as public API in Swift Testing,
1. We don't need to emit type metadata (much larger than what we really need)
   for every test function,
1. We don't need to duplicate a large chunk of the Swift ABI sources in order to
   walk the type metadata table correctly, and
1. Almost all the new code is written in Swift, whereas the code it is intended
   to replace could not be fully represented in Swift and needed to be written
   in C++.

The change also opens up the possibility of supporting generic types in the
future because we can emit metadata without needing to emit a nested type (which
is not always valid in a generic context.) That's a "future direction" and not
covered by this PR specifically.

I've defined a layout for entries in the new `swift5_tests` section that should
be flexible enough for us in the short-to-medium term and which lets us define
additional arbitrary test content record types. The layout of this section is
covered in depth in the new [TestContent.md](Documentation/ABI/TestContent.md)
article.

This functionality is only available if a test target enables the experimental
`"SymbolLinkageMarkers"` feature. We continue to emit protocol-conforming types
for now—that code will be removed if and when the experimental feature is
properly supported (modulo us adopting relevant changes to the feature's API.)

#735
swiftlang/swift#76698
swiftlang/swift#78411

This PR uses the experimental symbol linkage margers feature in the Swift
compiler to emit metadata about tests (and exit tests) into a dedicated section
of the test executable being built. At runtime, we discover that section and
read out the tests from it.

This has several benefits over our current model, which involves walking Swift's
type metadata table looking for types that conform to a protocol:

1. We don't need to define that protocol as public API in Swift Testing,
1. We don't need to emit type metadata (much larger than what we really need)
   for every test function,
1. We don't need to duplicate a large chunk of the Swift ABI sources in order to
   walk the type metadata table correctly, and
1. Almost all the new code is written in Swift, whereas the code it is intended
   to replace could not be fully represented in Swift and needed to be written
   in C++.

The change also opens up the possibility of supporting generic types in the
future because we can emit metadata without needing to emit a nested type (which
is not always valid in a generic context.) That's a "future direction" and not
covered by this PR specifically.

I've defined a layout for entries in the new `swift5_tests` section that should
be flexible enough for us in the short-to-medium term and which lets us define
additional arbitrary test content record types. The layout of this section is
covered in depth in the new [TestContent.md](Documentation/ABI/TestContent.md)
article.

This functionality is only available if a test target enables the experimental
`"SymbolLinkageMarkers"` feature. We continue to emit protocol-conforming types
for now—that code will be removed if and when the experimental feature is
properly supported (modulo us adopting relevant changes to the feature's API.)

#735
swiftlang/swift#76698
swiftlang/swift#78411

This PR uses the experimental symbol linkage margers feature in the Swift
compiler to emit metadata about tests (and exit tests) into a dedicated section
of the test executable being built. At runtime, we discover that section and
read out the tests from it.

This has several benefits over our current model, which involves walking Swift's
type metadata table looking for types that conform to a protocol:

1. We don't need to define that protocol as public API in Swift Testing,
1. We don't need to emit type metadata (much larger than what we really need)
   for every test function,
1. We don't need to duplicate a large chunk of the Swift ABI sources in order to
   walk the type metadata table correctly, and
1. Almost all the new code is written in Swift, whereas the code it is intended
   to replace could not be fully represented in Swift and needed to be written
   in C++.

The change also opens up the possibility of supporting generic types in the
future because we can emit metadata without needing to emit a nested type (which
is not always valid in a generic context.) That's a "future direction" and not
covered by this PR specifically.

I've defined a layout for entries in the new `swift5_tests` section that should
be flexible enough for us in the short-to-medium term and which lets us define
additional arbitrary test content record types. The layout of this section is
covered in depth in the new [TestContent.md](Documentation/ABI/TestContent.md)
article.

This functionality is only available if a test target enables the experimental
`"SymbolLinkageMarkers"` feature. We continue to emit protocol-conforming types
for now—that code will be removed if and when the experimental feature is
properly supported (modulo us adopting relevant changes to the feature's API.)

#735
swiftlang/swift#76698
swiftlang/swift#78411

This PR uses the experimental symbol linkage margers feature in the Swift
compiler to emit metadata about tests (and exit tests) into a dedicated section
of the test executable being built. At runtime, we discover that section and
read out the tests from it.

This has several benefits over our current model, which involves walking Swift's
type metadata table looking for types that conform to a protocol:

1. We don't need to define that protocol as public API in Swift Testing,
1. We don't need to emit type metadata (much larger than what we really need)
   for every test function,
1. We don't need to duplicate a large chunk of the Swift ABI sources in order to
   walk the type metadata table correctly, and
1. Almost all the new code is written in Swift, whereas the code it is intended
   to replace could not be fully represented in Swift and needed to be written
   in C++.

The change also opens up the possibility of supporting generic types in the
future because we can emit metadata without needing to emit a nested type (which
is not always valid in a generic context.) That's a "future direction" and not
covered by this PR specifically.

I've defined a layout for entries in the new `swift5_tests` section that should
be flexible enough for us in the short-to-medium term and which lets us define
additional arbitrary test content record types. The layout of this section is
covered in depth in the new [TestContent.md](Documentation/ABI/TestContent.md)
article.

This functionality is only available if a test target enables the experimental
`"SymbolLinkageMarkers"` feature. We continue to emit protocol-conforming types
for now—that code will be removed if and when the experimental feature is
properly supported (modulo us adopting relevant changes to the feature's API.)

#735
swiftlang/swift#76698
swiftlang/swift#78411

This PR uses the experimental symbol linkage margers feature in the Swift
compiler to emit metadata about tests (and exit tests) into a dedicated section
of the test executable being built. At runtime, we discover that section and
read out the tests from it.

This has several benefits over our current model, which involves walking Swift's
type metadata table looking for types that conform to a protocol:

1. We don't need to define that protocol as public API in Swift Testing,
1. We don't need to emit type metadata (much larger than what we really need)
   for every test function,
1. We don't need to duplicate a large chunk of the Swift ABI sources in order to
   walk the type metadata table correctly, and
1. Almost all the new code is written in Swift, whereas the code it is intended
   to replace could not be fully represented in Swift and needed to be written
   in C++.

The change also opens up the possibility of supporting generic types in the
future because we can emit metadata without needing to emit a nested type (which
is not always valid in a generic context.) That's a "future direction" and not
covered by this PR specifically.

I've defined a layout for entries in the new `swift5_tests` section that should
be flexible enough for us in the short-to-medium term and which lets us define
additional arbitrary test content record types. The layout of this section is
covered in depth in the new [TestContent.md](Documentation/ABI/TestContent.md)
article.

This functionality is only available if a test target enables the experimental
`"SymbolLinkageMarkers"` feature. We continue to emit protocol-conforming types
for now—that code will be removed if and when the experimental feature is
properly supported (modulo us adopting relevant changes to the feature's API.)

#735
swiftlang/swift#76698
swiftlang/swift#78411

This PR uses the experimental symbol linkage margers feature in the Swift
compiler to emit metadata about tests (and exit tests) into a dedicated section
of the test executable being built. At runtime, we discover that section and
read out the tests from it.

This has several benefits over our current model, which involves walking Swift's
type metadata table looking for types that conform to a protocol:

1. We don't need to define that protocol as public API in Swift Testing,
1. We don't need to emit type metadata (much larger than what we really need)
   for every test function,
1. We don't need to duplicate a large chunk of the Swift ABI sources in order to
   walk the type metadata table correctly, and
1. Almost all the new code is written in Swift, whereas the code it is intended
   to replace could not be fully represented in Swift and needed to be written
   in C++.

The change also opens up the possibility of supporting generic types in the
future because we can emit metadata without needing to emit a nested type (which
is not always valid in a generic context.) That's a "future direction" and not
covered by this PR specifically.

I've defined a layout for entries in the new `swift5_tests` section that should
be flexible enough for us in the short-to-medium term and which lets us define
additional arbitrary test content record types. The layout of this section is
covered in depth in the new [TestContent.md](Documentation/ABI/TestContent.md)
article.

This functionality is only available if a test target enables the experimental
`"SymbolLinkageMarkers"` feature. We continue to emit protocol-conforming types
for now—that code will be removed if and when the experimental feature is
properly supported (modulo us adopting relevant changes to the feature's API.)

#735
swiftlang/swift#76698
swiftlang/swift#78411

This PR uses the experimental symbol linkage margers feature in the Swift
compiler to emit metadata about tests (and exit tests) into a dedicated section
of the test executable being built. At runtime, we discover that section and
read out the tests from it.

This has several benefits over our current model, which involves walking Swift's
type metadata table looking for types that conform to a protocol:

1. We don't need to define that protocol as public API in Swift Testing,
1. We don't need to emit type metadata (much larger than what we really need)
   for every test function,
1. We don't need to duplicate a large chunk of the Swift ABI sources in order to
   walk the type metadata table correctly, and
1. Almost all the new code is written in Swift, whereas the code it is intended
   to replace could not be fully represented in Swift and needed to be written
   in C++.

The change also opens up the possibility of supporting generic types in the
future because we can emit metadata without needing to emit a nested type (which
is not always valid in a generic context.) That's a "future direction" and not
covered by this PR specifically.

I've defined a layout for entries in the new `swift5_tests` section that should
be flexible enough for us in the short-to-medium term and which lets us define
additional arbitrary test content record types. The layout of this section is
covered in depth in the new [TestContent.md](Documentation/ABI/TestContent.md)
article.

This functionality is only available if a test target enables the experimental
`"SymbolLinkageMarkers"` feature. We continue to emit protocol-conforming types
for now—that code will be removed if and when the experimental feature is
properly supported (modulo us adopting relevant changes to the feature's API.)

#735
swiftlang/swift#76698
swiftlang/swift#78411

This PR uses the experimental symbol linkage margers feature in the Swift
compiler to emit metadata about tests (and exit tests) into a dedicated section
of the test executable being built. At runtime, we discover that section and
read out the tests from it.

This has several benefits over our current model, which involves walking Swift's
type metadata table looking for types that conform to a protocol:

1. We don't need to define that protocol as public API in Swift Testing,
1. We don't need to emit type metadata (much larger than what we really need)
   for every test function,
1. We don't need to duplicate a large chunk of the Swift ABI sources in order to
   walk the type metadata table correctly, and
1. Almost all the new code is written in Swift, whereas the code it is intended
   to replace could not be fully represented in Swift and needed to be written
   in C++.

The change also opens up the possibility of supporting generic types in the
future because we can emit metadata without needing to emit a nested type (which
is not always valid in a generic context.) That's a "future direction" and not
covered by this PR specifically.

I've defined a layout for entries in the new `swift5_tests` section that should
be flexible enough for us in the short-to-medium term and which lets us define
additional arbitrary test content record types. The layout of this section is
covered in depth in the new [TestContent.md](Documentation/ABI/TestContent.md)
article.

This functionality is only available if a test target enables the experimental
`"SymbolLinkageMarkers"` feature. We continue to emit protocol-conforming types
for now—that code will be removed if and when the experimental feature is
properly supported (modulo us adopting relevant changes to the feature's API.)

#735
swiftlang/swift#76698
swiftlang/swift#78411

This PR uses the experimental symbol linkage margers feature in the Swift
compiler to emit metadata about tests (and exit tests) into a dedicated section
of the test executable being built. At runtime, we discover that section and
read out the tests from it.

This has several benefits over our current model, which involves walking Swift's
type metadata table looking for types that conform to a protocol:

1. We don't need to define that protocol as public API in Swift Testing,
1. We don't need to emit type metadata (much larger than what we really need)
   for every test function,
1. We don't need to duplicate a large chunk of the Swift ABI sources in order to
   walk the type metadata table correctly, and
1. Almost all the new code is written in Swift, whereas the code it is intended
   to replace could not be fully represented in Swift and needed to be written
   in C++.

The change also opens up the possibility of supporting generic types in the
future because we can emit metadata without needing to emit a nested type (which
is not always valid in a generic context.) That's a "future direction" and not
covered by this PR specifically.

I've defined a layout for entries in the new `swift5_tests` section that should
be flexible enough for us in the short-to-medium term and which lets us define
additional arbitrary test content record types. The layout of this section is
covered in depth in the new [TestContent.md](Documentation/ABI/TestContent.md)
article.

This functionality is only available if a test target enables the experimental
`"SymbolLinkageMarkers"` feature. We continue to emit protocol-conforming types
for now—that code will be removed if and when the experimental feature is
properly supported (modulo us adopting relevant changes to the feature's API.)

#735
swiftlang/swift#76698
swiftlang/swift#78411

This PR uses the experimental symbol linkage margers feature in the Swift
compiler to emit metadata about tests (and exit tests) into a dedicated section
of the test executable being built. At runtime, we discover that section and
read out the tests from it.

This has several benefits over our current model, which involves walking Swift's
type metadata table looking for types that conform to a protocol:

1. We don't need to define that protocol as public API in Swift Testing,
1. We don't need to emit type metadata (much larger than what we really need)
   for every test function,
1. We don't need to duplicate a large chunk of the Swift ABI sources in order to
   walk the type metadata table correctly, and
1. Almost all the new code is written in Swift, whereas the code it is intended
   to replace could not be fully represented in Swift and needed to be written
   in C++.

The change also opens up the possibility of supporting generic types in the
future because we can emit metadata without needing to emit a nested type (which
is not always valid in a generic context.) That's a "future direction" and not
covered by this PR specifically.

I've defined a layout for entries in the new `swift5_tests` section that should
be flexible enough for us in the short-to-medium term and which lets us define
additional arbitrary test content record types. The layout of this section is
covered in depth in the new [TestContent.md](Documentation/ABI/TestContent.md)
article.

This functionality is only available if a test target enables the experimental
`"SymbolLinkageMarkers"` feature. We continue to emit protocol-conforming types
for now—that code will be removed if and when the experimental feature is
properly supported (modulo us adopting relevant changes to the feature's API.)

#735
swiftlang/swift#76698
swiftlang/swift#78411

This PR uses the experimental symbol linkage margers feature in the
Swift compiler to emit metadata about tests (and exit tests) into a
dedicated section of the test executable being built. At runtime, we
discover that section and read out the tests from it.

This has several benefits over our current model, which involves walking
Swift's type metadata table looking for types that conform to a
protocol:

1. We don't need to define that protocol as public API in Swift Testing,
1. We don't need to emit type metadata (much larger than what we really
need) for every test function,
1. We don't need to duplicate a large chunk of the Swift ABI sources in
order to walk the type metadata table correctly, and
1. Almost all the new code is written in Swift, whereas the code it is
intended to replace could not be fully represented in Swift and needed
to be written in C++.

This change will be necessary to support Embedded Swift because there is
no type metadata section emitted for embedded targets.

The change also opens up the possibility of supporting generic types in
the future because we can emit metadata without needing to emit a nested
type (which is not always valid in a generic context.) That's a "future
direction" and not covered by this PR specifically.

I've defined a layout for entries in the new `swift5_tests` section that
should be flexible enough for us in the short-to-medium term and which
lets us define additional arbitrary test content record types. The
layout of this section is covered in depth in the new
[TestContent.md](https://github.com/swiftlang/swift-testing/blob/main/Documentation/ABI/TestContent.md)
article.

This functionality is only available if a test target enables the
experimental `"SymbolLinkageMarkers"` feature and only if Swift Testing
is used as a package (not as a toolchain component.) We continue to emit
protocol-conforming types for now—that code will be removed if and when
the experimental feature is properly supported (modulo us adopting
relevant changes to the feature's API.)

## See Also

#735
#764
swiftlang/swift#76698
swiftlang/swift#78411

### Checklist:

- [x] Code and documentation should follow the style of the [Style
Guide](https://github.com/apple/swift-testing/blob/main/Documentation/StyleGuide.md).
- [x] If public symbols are renamed or modified, DocC references should
be updated.