Add definition for _Bool in C++ mode

[etcwilde]

_Bool is not defined when importing Foundation with C++ interop enabled
resulting in the missing type. This patch defines _Bool to bool when
compiling for C++.

I'm not entire sure how to test this. The test itself is fairly simple, it's just

import Foundation

with the invocation swiftc -Xcc -std=c++17 -enable-experimental-cxx-interop, but I don't know how to tell XCTest to do that so I would like pointers there.

rdar://99713881

[compnerd]

I wonder if we want to use stdbool.h here as well as we do on the other platforms.

[etcwilde]

So include stdbool.h, and then fix the uses of _Bool through the codebase to use bool?

[compnerd]

Yeah, that seems like a better approach - using the standard to use the proper spelling irrespective of the language.

[ian-twilightcoder]

Are there any other uses of _Bool in CF?

[etcwilde]

This is the list I have. I just need to find a cycle or two to go through and change them and run the tests.

./CoreFoundation/Base.subproj/ForSwiftFoundationOnly.h:static inline _Bool _resizeConditionalAllocationBuffer(_ConditionalAllocationBuffer *_Nonnull buffer, size_t amt) {
./CoreFoundation/Base.subproj/ForSwiftFoundationOnly.h:static inline _Bool _withStackOrHeapBuffer(size_t amount, void (__attribute__((noescape)) ^ _Nonnull applier)(_ConditionalAllocationBuffer *_Nonnull)) {
./CoreFoundation/Base.subproj/ForSwiftFoundationOnly.h:static inline _Bool _withStackOrHeapBuffer(size_t amount, void (__attribute__((noescape)) ^ _Nonnull applier)(_ConditionalAllocationBuffer *_Nonnull)) {
./CoreFoundation/Base.subproj/ForSwiftFoundationOnly.h:static inline _Bool _withStackOrHeapBufferWithResultInArguments(size_t amount, void (__attribute__((noescape)) ^ _Nonnull applier)(void *_Nonnull memory, size_t capacity, _Bool onStack)) {
./CoreFoundation/Base.subproj/ForSwiftFoundationOnly.h:static _Bool const _CFHasRenameat2 = 1;
./CoreFoundation/Base.subproj/ForSwiftFoundationOnly.h:static _Bool const _CFHasRenameat2 = 0;
./CoreFoundation/Base.subproj/CFBase.h:    typedef _Bool                   Boolean;
./CoreFoundation/Base.subproj/ForFoundationOnly.h:#define BOOL _Bool
./CoreFoundation/Base.subproj/CFOverflow.h:    static _Bool __os_warn_unused(_Bool x) __attribute__((__warn_unused_result__));
./CoreFoundation/Base.subproj/CFOverflow.h:    static _Bool __os_warn_unused(_Bool x) { return x; }
./Darwin/shims/FoundationOverlayShims.h:    _Bool onStack;
./Darwin/shims/FoundationOverlayShims.h:static inline _Bool _resizeConditionalAllocationBuffer(_ConditionalAllocationBuffer *_Nonnull buffer, size_t amt) {
./Darwin/shims/FoundationOverlayShims.h:static inline _Bool _withStackOrHeapBuffer(size_t amount, void (__attribute__((noescape)) ^ _Nonnull applier)(_ConditionalAllocationBuffer *_Nonnull)) {
./CoreFoundation/PlugIn.subproj/CFBundle_Grok.c:static inline _Bool _CF_isspace(int c) {

[parkera]

I agree with @compnerd's approach.

[etcwilde]

Alright, I've gone though and replaced uses of _Bool with bool and included stdbool.h.

[etcwilde]

@swift-ci please test

[compnerd]

Thanks!

[compnerd]

ForFoundationOnly.h L89-102 needs adjustment.

[etcwilde]

@swift-ci please test

[etcwilde]

ForFoundationOnly.h L89-102 needs adjustment.

Yeah, the issue is that BOOL isn't always _Bool, sometimes it's unsigned char. I just added #define BOOL bool here;

#ifdef __OBJC__
#define NSISARGTYPE id _Nullable
#else
#define NSISARGTYPE void * _Nullable
#define BOOL bool
#endif

[ian-twilightcoder]

ForFoundationOnly.h L89-102 needs adjustment.

Yeah, the issue is that BOOL isn't always _Bool, sometimes it's unsigned char. I just added #define BOOL bool here;

#ifdef __OBJC__
#define NSISARGTYPE id _Nullable
#else
#define NSISARGTYPE void * _Nullable
#define BOOL bool
#endif

That seems really sketchy, why does BOOL need to be defined at all in CF?

[etcwilde]

That seems really sketchy, why does BOOL need to be defined at all in CF?

I think it's for ObjC compatibility since we're also building this code without ObjC.
The ForFoundationOnly.h only defines BOOL when __OBJC__ isn't defined.

// Arguments to these are id, but this header is non-Objc
#ifdef __OBJC__
#define NSISARGTYPE id _Nullable
#else
#define NSISARGTYPE void * _Nullable
#define BOOL _Bool
#endif

[etcwilde]

Yeah, the BOOL definition came in on the Monterey CoreFoundation sync-up; d8e8a8b#diff-32adebf83b3fcbdf66de2d896086ed7702b8cb7083cedb9594f092a11f513618

[ian-twilightcoder]

That seems really sketchy, why does BOOL need to be defined at all in CF?

I think it's for ObjC compatibility since we're also building this code without ObjC. The ForFoundationOnly.h only defines BOOL when __OBJC__ isn't defined.

// Arguments to these are id, but this header is non-Objc
#ifdef __OBJC__
#define NSISARGTYPE id _Nullable
#else
#define NSISARGTYPE void * _Nullable
#define BOOL _Bool
#endif

Including <objc/objc.h> would be a lot better, I think it's ok to do that from pure C. Declaring it yourself is likely to eventually cause module problems

[etcwilde]

Including <objc/objc.h> would be a lot better, I think it's ok to do that from pure C. Declaring it yourself is likely to eventually cause module problems

The definition was in the original code, I'm just trying to allow C++ to work, so I think that's maybe for another PR?
Windows and Linux come with objc/objc.h and it doesn't look like corelib-foundation is providing it. How are we distributing that header for those cases?

[ian-twilightcoder]

Including <objc/objc.h> would be a lot better, I think it's ok to do that from pure C. Declaring it yourself is likely to eventually cause module problems

The definition was in the original code, I'm just trying to allow C++ to work, so I think that's maybe for another PR? Windows and Linux come with objc/objc.h and it doesn't look like corelib-foundation is providing it. How are we distributing that header for those cases?

Yeah true, not really related to this PR. I'm not sure what provides the objc runtime on Windows and Linux.