[C++20] [Modules] The entity with internal linkage shouldn't be accessed in other TU even if they are in the same module

[ChuanqiXu9]

Reproducer:

export module m:a;
static int a = 32;

export module m:b;
import :a;
int b() {
    return a;
}

Currently the reproducer is accepted. But we expect the compiler to reject it and saying that the partition b can't access the static variable 'a' in partition a.

[llvmbot]

@llvm/issue-subscribers-clang-modules

[iains]

Also, I believe we should reject this simpler case where the implementation implicitly imports the interface.

export module S;
static void f() {}

module S;

void foo() {
  f ();
}

[iains]

https://reviews.llvm.org/D126959 fixes some of the cases without any change to Sema::isModuleUnitOfCurrentTU since it makes the implementation unambiguously a different TU.

[iains]

(there is still more to do, of course, I also plan to update https://reviews.llvm.org/D145965 to remove the changes to Sema::isModuleUnitOfCurrentTU [which @ChuanqiXu9 is intending to handle separately])

[cor3ntin]

@iains @ChuanqiXu9 someone was asking me about this, are you still working on it?

[iains]

@iains @ChuanqiXu9 someone was asking me about this, are you still working on it?

I am not, for the foreseeable future.

[ChuanqiXu9]

@iains @ChuanqiXu9 someone was asking me about this, are you still working on it?

This is still on my TODO list. But it may not be too soon given my time schedule.