PrintAsClang is supposed to emit declarations in the same order regardless of the compiler’s internal state, but we have repeatedly found that our current criteria are inadequate, resulting in non-functionality-affecting changes to generated header content. Add a diagnostic that’s emitted when this happens soliciting a bug report.
Since there *should* be no cases where the compiler fails to order declarations, this diagnostic is never actually emitted. Instead, we test this change by enabling `-verify` on nearly all PrintAsClang tests to make sure they are unaffected.
This did demonstrate a missing criterion that only mattered in C++ mode: extensions that varied only in their generic signature were not sorted stably. Add a sort criterion for this.
Find all the usages of `--enable-experimental-feature` or
`--enable-upcoming-feature` in the tests and replace some of the
`REQUIRES: asserts` to use `REQUIRES: swift-feature-Foo` instead, which
should correctly apply to depending on the asserts/noasserts mode of the
toolchain for each feature.
Remove some comments that talked about enabling asserts since they don't
apply anymore (but I might had miss some).
All this was done with an automated script, so some formatting weirdness
might happen, but I hope I fixed most of those.
There might be some tests that were `REQUIRES: asserts` that might run
in `noasserts` toolchains now. This will normally be because their
feature went from experimental to upcoming/base and the tests were not
updated.
Use the `%target-swift-5.1-abi-triple` substitution to compile the tests for
deployment to the minimum OS versions required for use of _Concurrency APIs,
instead of disabling availability checking.
It is really involved to change how methods and classes are emitted into
the header so this patch introduces the impression of nested structs
through using statements and still emits the structs themselves as top
level structs. It emits them in their own namespace to avoid name
collisions. This patch also had to change some names to be fully
qualified to avoid some name lookup errors in case of nested structs.
Moreover, nesting level of 3 and above requires C++17 because it relies
on nested namespaces. Only nested structs are supported, not nested
classes.
Since this patch is already started to grow quite big, I decided to put
it out for reviews and plan to address some of the shortcomings in a
follow-up PR.
rdar://118793469
C++ only support multiparameter operator[] in C++23 and up. Change the
code to protect such overloaded operators with a C++ language mode
check.
rdar://133539699
The generated header would not compile without these dependencies. Moreover
users probably expect all-public option to be the most permissive filter
including the maximal amount of declarations.
We export cdecl function declarations twice: for Objective-C and for
C++. When the code is compiled in Objective-C++ both of the declarations
are visible to the compiler. The generated header did not compile,
because only one of the declarations were noexcept. There are multiple
possible ways to fix this issue, one of them would make only C++
declarations visible in Objective-C++ mode. However, for this particular
problem I decided to also make the Objective-C functions SWIFT_NOEXCEPT.
This approach resolves the inconsistency that broke the code when
compiled in Objective-C++ mode. Moreover, Swift guarantees that those
cdecl declarations cannot raise errors, so in case we only generate the
C declarations and consume them from C++ or Objective-C++, those are the
correct declarations.
rdar://129550313
The generated thunks for functions can refer to some internal methods of
their arguments. As a result, those generated thunks should always be
after the definitions of the corresponding argument types. The printer
ordered the declarations by name, and Swift had the convention starting
types with upper case letters and functions with lower case letters.
This naming convention together with the ordering resulted in the
correct ordering in most of the cases. There were a couple of exceptions
when people diverged from the naming conventions or wanted to export
operators. This patch fixes this problem by always ordering type decls
before function decls.
rdar://129276354
Extend the previous commit’s support for functions that return Never to also properly generate code for *throwing* Never functions. This is a little subtle because:
• At the SWIFT_CALL level, throwing Never functions are *not* noreturn
• At the thunk level, throwing Never functions are noreturn *only* if you’re using exceptions; if you’re using swift::Expected, they should throw
• In either case, the compiler cannot statically prove that thunks are noreturn except on the error path, so we need to add an abort() call on the success path
Swift-to-C++ thunk printing for functions didn’t really take into account Swift’s `Never` type. This type maps to `SWIFT_NORETURN`, but it also requires other tweaks to code generation, such as omitting the `return` keyword. (Removing that requires minor changes to many tests.)
Fixes rdar://124137073.
This is an ABI breaking change for Windows. `WCHAR` on Windows is
mapped to `short` (`-fshort-wchar` makes it `unsigned short`). When C++
interop is enabled, `WCHAR` will be mapped to `wchar_t` which is then
mapped to `short` (or `unsigned short` if `-fshort-wchar` is specified).
Correct the mapping type to get the desired behaviour.
This macro applies always_inline in addition to inline. It also applies artificial, which lets debugger know that this is an artificial function. The used attribute is added in debug builds to ensure that the symbol is emitted in the binary so that LLDB can invoke it.
Each emitted declaration is annotated with the external_source_symbol with its own USR, to allow Clang's indexer to recognize this declaration as a Swift declaration with a specific USR
