[clang] [clang][Sema] Improve error recovery for id-expressions referencing invalid decls (PR #81662)

[Nathan Ridge via cfe-commits]

HighCommander4 wrote:

Here is a reduced testcase for the OpenMP test failure:

```c++
#pragma omp declare target

static long double ld_return1e() { return 0; }

void external() {
  void *p1 = reinterpret_cast<void*>(&ld_return1e);
}

#pragma omp end declare target
```

When built with the following command:

> clang -cc1 -fopenmp -triple nvptx64-unknown-unknown -aux-triple 
> x86_64-unknown-linux test.cpp -fopenmp-is-target-device -fsyntax-only

Here are the diagnostics before this patch:

```console
test.cpp:6:39: error: 'ld_return1e' requires 128 bit size 'long double' type 
support, but target 'nvptx64-unknown-unknown' does not support it
    6 |   void *p1 = reinterpret_cast<void*>(&ld_return1e);
      |                                       ^
test.cpp:3:20: note: 'ld_return1e' defined here
    3 | static long double ld_return1e() { return 0; }
      |                    ^
1 error generated
```

and here are the diagnostics after this patch:

```console
test.cpp:6:39: error: 'ld_return1e' requires 128 bit size 'long double' type 
support, but target 'nvptx64-unknown-unknown' does not support it
    6 |   void *p1 = reinterpret_cast<void*>(&ld_return1e);
      |                                       ^
test.cpp:3:20: note: 'ld_return1e' defined here
    3 | static long double ld_return1e() { return 0; }
      |                    ^
test.cpp:3:20: error: 'ld_return1e' requires 128 bit size 'long double' type 
support, but target 'nvptx64-unknown-unknown' does not support it
    3 | static long double ld_return1e() { return 0; }
      |                    ^
test.cpp:6:39: note: called by 'external'
    6 |   void *p1 = reinterpret_cast<void*>(&ld_return1e);
      |                                       ^
test.cpp:3:20: note: 'ld_return1e' defined here
    3 | static long double ld_return1e() { return 0; }
      |                    ^
2 errors generated.
```

Basically, the patch somehow causes the same diagnostic to be issued a second 
time at a different location.

https://github.com/llvm/llvm-project/pull/81662
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits