clangd gets confused by -U... in command line

[jpalus]

For input file test.c:

#include <stdio.h>

#ifndef MSG
#define MSG "hello world"
#endif

void hello() {
  printf(MSG);
}

and compile_commands.json:

[
  {      
    "directory": "/tmp",      
    "command": "clang -UMSG -c -o test.o test.c'",            
    "file": "test.c",            
    "output": "test.o"            
  }      
]

clangd reports false positive:

$ clangd --check=test.c
...
I[21:05:56.452] Indexing headers...
I[21:05:56.472] Building AST...
E[21:05:56.491] [undeclared_var_use] Line 8: use of undeclared identifier 'MSG'
I[21:05:56.491] Indexing AST...
I[21:05:56.492] Building inlay hints
I[21:05:56.492] Building semantic highlighting
I[21:05:56.492] Testing features at each token (may be slow in large files)
I[21:05:56.493] Found definition heuristically using nearby identifier MSG
I[21:05:56.493] Found definition heuristically using nearby identifier MSG
I[21:05:56.501] All checks completed, 1 errors

[llvmbot]

@llvm/issue-subscribers-clangd

Author: Jan Palus (jpalus)

For input file `test.c`: ```c #include <stdio.h>

#ifndef MSG
#define MSG "hello world"
#endif

void hello() {
printf(MSG);
}

and `compile_commands.json`:
```json
[
  {      
    "directory": "/tmp",      
    "command": "clang -UMSG -c -o test.o test.c'",            
    "file": "test.c",            
    "output": "test.o"            
  }      
]

clangd reports false positive:

$ clangd --check=test.c
...
I[21:05:56.452] Indexing headers...
I[21:05:56.472] Building AST...
E[21:05:56.491] [undeclared_var_use] Line 8: use of undeclared identifier 'MSG'
I[21:05:56.491] Indexing AST...
I[21:05:56.492] Building inlay hints
I[21:05:56.492] Building semantic highlighting
I[21:05:56.492] Testing features at each token (may be slow in large files)
I[21:05:56.493] Found definition heuristically using nearby identifier MSG
I[21:05:56.493] Found definition heuristically using nearby identifier MSG
I[21:05:56.501] All checks completed, 1 errors

[HighCommander4]

I believe this has the same underlying cause as clangd/clangd#1396: "clang first loads the preamble and then applies the builtin buffer".

Here, the #define MSG becomes part of the preamble, and the -UMSG becomes part of the builtin buffer.

A workaround is to add a non-preprocessor declaration between the includes and the #define MSG to force the preamble to end before the #define MSG:

#include <stdio.h>

struct EndPreamble {}; // dummy declaration to force the end of the preamble

#ifndef MSG
#define MSG "hello world"
#endif

void hello() {
  printf(MSG);
}

[HighCommander4]

Duplicate of clangd/clangd#1396