If a C++ namespace has redeclarations in a bridging header, printing AST for the namespace would crash the compiler. This is because such a redeclaration would not have an owning Clang module, and the AST printer did not account for that.
This change fixes the crash.
rdar://151715540
(cherry picked from commit cc9c51deea)
Explanation: We did not handle this declaration kind. This PR makes sure we
mangle it the same way we do for the target declaration.
Issue: rdar://152841420
Risk: Low, the fix is small, localized, and straightforward.
Testing: Regression test added.
Original PR: #82144
Reviewer: @egorzhdan @hnrklssn @j-hui
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.
With the upcoming `MemberImportVisibility` feature enabled, code built with Cxx
interop also enabled could be rejected by the compiler with cryptic errors
about the `__ObjC` module not being imported. This is the result of a
surprising implementation detail of Cxx interop. When importing C++ namespaces
and their members, the Clang importer puts these declarations in the Clang
header import module (a.k.a. the bridging header module, `__ObjC`). C++
namespaces don't have a logical modular home in the Swift AST because they can
span multiple modules, so it's understandable why this implementation was
chosen. However, the concrete members of namespaces also get placed in the
`__ObjC` module too, and this really confuses things.
To work around this idiosyncrasy of Cxx interop, I've introduced
`Decl::getModuleContextForNameLookup()` which returns the module that a
declaration would ideally belong to if Cxx interop didn't have this behavior.
This alternative to `Decl::getModuleContext()` is now used everywhere that
`MemberImportVisibility` rules are enforced to provide consistency.
Additionally, I found that I also had to further special-case the header import
module for Cxx interop because it turns out that there are some additional
declarations, beyond imported namespaces, that also live there and need to be
implicitly visible in every source file. The `__ObjC` module is not implicitly
imported in source files when Cxx interop is enabled, so these declarations are
not deemed visible under normal name lookup rules. When I tried to add an
implicit import of `__ObjC` when Cxx interop is enabled, it broke a bunch
tests. So for now, when a decl really belongs to the `__ObjC` module in Cxx
interop mode, we just always allow it to be referenced.
This Cxx interop behavior really needs a re-think in my opinion, but that will
require larger discussions.
Resolves rdar://136600598.
When importing a C++ class template instantiation, Swift translates the template parameter type names from C++ into their Swift equivalent.
For instance, `basic_string<wchar_t, char_traits<wchar_t>, allocator<wchar_t>>` gets imported as `basic_string<Scalar, char_traits<Scalar>, allocator<Scalar>>`: `wchar_t` is imported as `CWideChar`, which is a typealias for `Scalar` on most platforms including Darwin. Notice that Swift goes through the `CWideChar` typealias on the specific platform. Another instantiation `basic_string<uint32_t, char_traits<uint32_t>, allocator<uint32_t>>` also gets imported as `basic_string<Scalar, char_traits<Scalar>, allocator<Scalar>>`: `uint32_t` is also imported as `Scalar`. This is problematic because we have two distinct C++ types that have the same name in Swift.
This change makes sure Swift doesn't go through typealiases when emitting names of template parameters, so `wchar_t` would now get printed as `CWideChar`, `int` would get printed as `CInt`, etc.
This also encourages clients to use the correct type (`CInt`, `CWideChar`, etc) instead of relying on platform-specific typealiases.
rdar://115673622
This ensures that when we're printing multiple module interfaces, we separate out
the namespaces and their members correctly between the produced module interfaces.
If a non-member operator is declared in a C++ namespace, we previously imported it as a static member of the enum that represents the C++ namespace.
This is not always correct under Swift rules for operators. In pure Swift, this code is valid:
```
public protocol UnsafeCxxRandomAccessIterator {
static func +=(lhs: inout Self, rhs: Int)
}
enum std {
public struct A : UnsafeCxxRandomAccessIterator {
public static func += (lhs: inout A, rhs: Int) {
}
}
}
```
but this is not valid:
```
public protocol UnsafeCxxRandomAccessIterator {
static func +=(lhs: inout Self, rhs: Int)
}
enum std {
public struct A : UnsafeCxxRandomAccessIterator {}
public static func += (lhs: inout A, rhs: Int) {}
}
// error: Member operator '+=' must have at least one argument of type 'std'
```
This caused assertion failures in SILGen when conforming C++ iterator types to `UnsafeCxxRandomAccessIterator`.
Make sure nested namespaces are added to the SwiftLookupTable when they are declared in a non-canonical redecl of the parent namespace.
This was blocking C++ interop adoption in SwiftCompilerSources.
This does not include subscript operators.
Before this is re-enabled operators need to be re-implemented. Right now they are the source of a lot of bugs. They cause frequent crashes and mis compiles. Also, templated operators insert a lot of names into global lookup which causes problems.
They also don't work on Windows.
Previously a namespace declaration was imported along with all of its redeclarations, and their members were added to a single Swift extension. This was problematic when a single namespace is declared in multiple modules – the extension belonged to only one of them.
For an example of this, try printing a module interface for `std.string`/`std.iosfwd` – it will be empty, even though the declarations from those modules are actually imported into Swift correctly.
This change makes sure that when we're importing different redeclarations of the same namespace, we're adding them as separate extensions to appropriate modules.
This fixes the issue that prevents `std::string::size_type` from being imported into Swift: `size_type` is imported before its parent struct, and we're skipping it when importing the struct afterwards.
This is caused by an out-of-line decl for `std::string::npos`:
```cpp
template<class _CharT, class _Traits, class _Allocator>
_LIBCPP_FUNC_VIS
const typename basic_string<_CharT, _Traits, _Allocator>::size_type
basic_string<_CharT, _Traits, _Allocator>::npos;
```
When importing `npos`, we first import `size_type`, which triggers the issue.
C++ namespaces are module-independent, but enums are owned by their module's in Swift. So, to prevent declaring two enums with the same name, this patch implements a new approach to namespaces: enums with extensions.
Here's an example:
```
// Module A
namespace N { void test1(); }
// Module B
namespace N { void test2(); }
// __ObjC module
enum N { }
// Swift module A
extension N { func test1() }
// Swift module B
extension N { func test1() }
```
Thanks to @gribozavr for the great idea.
