[clang] [Modules] Add warning for including header in module purview

[davidstone]

Given the following code

export module a;

#include <new>

When compiling with -std=c++26 -stdlib=libc++ -Weverything

all that we get as output is

In file included from <source>:3:
/opt/compiler-explorer/clang-assertions-trunk-20231009/bin/../include/c++/v1/new:217:66: error: declaration of 'operator new' in module a follows declaration in the global module
  217 | _LIBCPP_NODISCARD_AFTER_CXX17 _LIBCPP_OVERRIDABLE_FUNC_VIS void* operator new(std::size_t __sz) _THROW_BAD_ALLOC;
      |                                                                  ^
note: previous declaration is here
/opt/compiler-explorer/clang-assertions-trunk-20231009/bin/../include/c++/v1/new:219:36: error: declaration of 'operator delete' in module a follows declaration in the global module
  219 | _LIBCPP_OVERRIDABLE_FUNC_VIS void  operator delete(void* __p) _NOEXCEPT;
      |                                    ^
note: previous declaration is here
/opt/compiler-explorer/clang-assertions-trunk-20231009/bin/../include/c++/v1/new:225:66: error: declaration of 'operator new[]' in module a follows declaration in the global module
  225 | _LIBCPP_NODISCARD_AFTER_CXX17 _LIBCPP_OVERRIDABLE_FUNC_VIS void* operator new[](std::size_t __sz) _THROW_BAD_ALLOC;
      |                                                                  ^
note: previous declaration is here
/opt/compiler-explorer/clang-assertions-trunk-20231009/bin/../include/c++/v1/new:227:36: error: declaration of 'operator delete[]' in module a follows declaration in the global module
  227 | _LIBCPP_OVERRIDABLE_FUNC_VIS void  operator delete[](void* __p) _NOEXCEPT;
      |                                    ^
note: previous declaration is here
/opt/compiler-explorer/clang-assertions-trunk-20231009/bin/../include/c++/v1/new:234:66: error: declaration of 'operator new' in module a follows declaration in the global module
  234 | _LIBCPP_NODISCARD_AFTER_CXX17 _LIBCPP_OVERRIDABLE_FUNC_VIS void* operator new(std::size_t __sz, std::align_val_t) _THROW_BAD_ALLOC;
      |                                                                  ^
note: previous declaration is here
/opt/compiler-explorer/clang-assertions-trunk-20231009/bin/../include/c++/v1/new:236:36: error: declaration of 'operator delete' in module a follows declaration in the global module
  236 | _LIBCPP_OVERRIDABLE_FUNC_VIS void  operator delete(void* __p, std::align_val_t) _NOEXCEPT;
      |                                    ^
note: previous declaration is here
/opt/compiler-explorer/clang-assertions-trunk-20231009/bin/../include/c++/v1/new:242:66: error: declaration of 'operator new[]' in module a follows declaration in the global module
  242 | _LIBCPP_NODISCARD_AFTER_CXX17 _LIBCPP_OVERRIDABLE_FUNC_VIS void* operator new[](std::size_t __sz, std::align_val_t) _THROW_BAD_ALLOC;
      |                                                                  ^
note: previous declaration is here
/opt/compiler-explorer/clang-assertions-trunk-20231009/bin/../include/c++/v1/new:244:36: error: declaration of 'operator delete[]' in module a follows declaration in the global module
  244 | _LIBCPP_OVERRIDABLE_FUNC_VIS void  operator delete[](void* __p, std::align_val_t) _NOEXCEPT;
      |                                    ^
note: previous declaration is here
8 errors generated.
Compiler returned: 1

This is not obvious to users what they did wrong. The issue here is that the #include is in the module purview. It would be nice to have a general warning saying "Don't include there, type module; #include <whatever> instead.

[llvmbot]

@llvm/issue-subscribers-clang-modules

Given the following code

export module a;

#include <new>

When compiling with -std=c++26 -stdlib=libc++ -Weverything

all that we get as output is

In file included from <source>:3:
/opt/compiler-explorer/clang-assertions-trunk-20231009/bin/../include/c++/v1/new:217:66: error: declaration of 'operator new' in module a follows declaration in the global module
  217 | _LIBCPP_NODISCARD_AFTER_CXX17 _LIBCPP_OVERRIDABLE_FUNC_VIS void* operator new(std::size_t __sz) _THROW_BAD_ALLOC;
      |                                                                  ^
note: previous declaration is here
/opt/compiler-explorer/clang-assertions-trunk-20231009/bin/../include/c++/v1/new:219:36: error: declaration of 'operator delete' in module a follows declaration in the global module
  219 | _LIBCPP_OVERRIDABLE_FUNC_VIS void  operator delete(void* __p) _NOEXCEPT;
      |                                    ^
note: previous declaration is here
/opt/compiler-explorer/clang-assertions-trunk-20231009/bin/../include/c++/v1/new:225:66: error: declaration of 'operator new[]' in module a follows declaration in the global module
  225 | _LIBCPP_NODISCARD_AFTER_CXX17 _LIBCPP_OVERRIDABLE_FUNC_VIS void* operator new[](std::size_t __sz) _THROW_BAD_ALLOC;
      |                                                                  ^
note: previous declaration is here
/opt/compiler-explorer/clang-assertions-trunk-20231009/bin/../include/c++/v1/new:227:36: error: declaration of 'operator delete[]' in module a follows declaration in the global module
  227 | _LIBCPP_OVERRIDABLE_FUNC_VIS void  operator delete[](void* __p) _NOEXCEPT;
      |                                    ^
note: previous declaration is here
/opt/compiler-explorer/clang-assertions-trunk-20231009/bin/../include/c++/v1/new:234:66: error: declaration of 'operator new' in module a follows declaration in the global module
  234 | _LIBCPP_NODISCARD_AFTER_CXX17 _LIBCPP_OVERRIDABLE_FUNC_VIS void* operator new(std::size_t __sz, std::align_val_t) _THROW_BAD_ALLOC;
      |                                                                  ^
note: previous declaration is here
/opt/compiler-explorer/clang-assertions-trunk-20231009/bin/../include/c++/v1/new:236:36: error: declaration of 'operator delete' in module a follows declaration in the global module
  236 | _LIBCPP_OVERRIDABLE_FUNC_VIS void  operator delete(void* __p, std::align_val_t) _NOEXCEPT;
      |                                    ^
note: previous declaration is here
/opt/compiler-explorer/clang-assertions-trunk-20231009/bin/../include/c++/v1/new:242:66: error: declaration of 'operator new[]' in module a follows declaration in the global module
  242 | _LIBCPP_NODISCARD_AFTER_CXX17 _LIBCPP_OVERRIDABLE_FUNC_VIS void* operator new[](std::size_t __sz, std::align_val_t) _THROW_BAD_ALLOC;
      |                                                                  ^
note: previous declaration is here
/opt/compiler-explorer/clang-assertions-trunk-20231009/bin/../include/c++/v1/new:244:36: error: declaration of 'operator delete[]' in module a follows declaration in the global module
  244 | _LIBCPP_OVERRIDABLE_FUNC_VIS void  operator delete[](void* __p, std::align_val_t) _NOEXCEPT;
      |                                    ^
note: previous declaration is here
8 errors generated.
Compiler returned: 1

This is not obvious to users what they did wrong. The issue here is that the #include is in the module purview. It would be nice to have a general warning saying "Don't include there, type module; #include <whatever> instead.

[ChuanqiXu9]

I feel like this is not so do-able thing. One the one hand, #include is not a thing in the semantic analysis. It belongs to the preprocessor. So it means we need to implement this warning in the preprocessor if we really want to do this. On the other hand, #include may happen in the module purview in a proper way. One use case is to define the enums. For example, we did this:

llvm-project/clang/include/clang/AST/ASTContext.h

Lines 1116 to 1136 in b5dffd4

	#define IMAGE_TYPE(ImgType, Id, SingletonId, Access, Suffix) \
	CanQualType SingletonId;
	#include "clang/Basic/OpenCLImageTypes.def"
	CanQualType OCLSamplerTy, OCLEventTy, OCLClkEventTy;
	CanQualType OCLQueueTy, OCLReserveIDTy;
	CanQualType IncompleteMatrixIdxTy;
	CanQualType OMPArraySectionTy, OMPArrayShapingTy, OMPIteratorTy;
	#define EXT_OPAQUE_TYPE(ExtType, Id, Ext) \
	CanQualType Id##Ty;
	#include "clang/Basic/OpenCLExtensionTypes.def"
	#define SVE_TYPE(Name, Id, SingletonId) \
	CanQualType SingletonId;
	#include "clang/Basic/AArch64SVEACLETypes.def"
	#define PPC_VECTOR_TYPE(Name, Id, Size) \
	CanQualType Id##Ty;
	#include "clang/Basic/PPCTypes.def"
	#define RVV_TYPE(Name, Id, SingletonId) \
	CanQualType SingletonId;
	#include "clang/Basic/RISCVVTypes.def"
	#define WASM_TYPE(Name, Id, SingletonId) CanQualType SingletonId;
	#include "clang/Basic/WebAssemblyReferenceTypes.def"

Also there is another practice to modularize a library:

// header.h
#ifndef USE_MODULES
#include "..."
#endif

EXPORT class A {...};

other declarations....

// a.cppm
module;
#define USE_MODULES
#define EXPORT export
#include "..."
export module a;
#include "header.h"
...

So it looks not good to prevent including in the module purview.

[koplas]

MSVC emits a warning for this: C5244

[ChuanqiXu9]

MSVC emits a warning for this: C5244

Got it. but that still can't work for the use cases I mentioned above... a hard decision to make ...

[iains]

+1 to @ChuanqiXu9, this seems like a very hard warning to get right (i.e. avoid many false-positives). It is certainly not obvious how the pre-processor would differentiate headers that are intended to be GMF from ones that are intended to be part of the named module.

I suppose we could warn only for a known set (i.e. the std lib).

[ChuanqiXu9]

+1 to @ChuanqiXu9, this seems like a very hard warning to get right (i.e. avoid many false-positives). It is certainly not obvious how the pre-processor would differentiate headers that are intended to be GMF from ones that are intended to be part of the named module.

I suppose we could warn only for a known set (i.e. the std lib).

Good idea. Sent #69555. Here I treat the headers within angles as the system headers. I think the judgement may make sense.