Issue 208655
Summary [compiler-rt][tysan][libc++] False positives with global/static struct containing STL containers
Labels libc++
Assignees
Reporter onurcan-isler-oss
    # global/static `struct` offset and layout
#### Flags: `-stdlib=libc++ -fsanitize=type -O1`
#### Compiler: x86-64 clang 22.1.0
#### Link: https://godbolt.org/z/x4nM6K5Mv
#### Code: 
```.cpp
#include <vector>

struct Registry {
    std::vector<int> arr;
    char temp_byte;
    bool bool_var = false;
};

int main() {
    static Registry r;
    r.arr.push_back(0); // reserve is also repro
    r.bool_var = true;
}
```
#### Logs:
```
==1==ERROR: TypeSanitizer: type-aliasing-violation on address 0x62c0c8b4df08 (pc 0x62c0c81f3179 bp 0x7ffdf94d1420 sp 0x7ffdf94d13b0 tid 1)
WRITE of size 1 at 0x62c0c8b4df08 with type bool (in Registry at offset 32) accesses part of an existing object of type p1 int (in std::__1::vector<int, std::__1::allocator<int>> at offset 0) that starts at offset -32
    #0 0x62c0c81f3178  (/app/output.s+0x36178)
```
#### Explanation:

- If we remove the “static” keyword in Registry initialization, it will not reproduce.
- If we replace the vector array with three int pointers, it will not reproduce.
- If we remove the vector function, it will not reproduce.
- If we reorder struct variables it will not reproduce:
```.cpp
#include <vector>

struct Registry {
    char temp_byte;
    std::vector<int> arr;
    bool bool_var = false;
};

int main() {
    static Registry r;
    r.arr.push_back(0);
    r.bool_var = true;
}
```
- Global `struct` also fires:
```.cpp
#include <vector>

struct Registry {
    std::vector<int> arr;
    char temp_byte;
    bool bool_var = false;
};

Registry r;

int main() {
    r.arr.push_back(0); // reserve is also repro
    r.bool_var = true;
}
```
Might be related to global/static STL containers FP in 20.1.0: https://godbolt.org/z/q3TsrK5c4
But fixed in 22.1.0: https://godbolt.org/z/1qE3z6eea
_______________________________________________
llvm-bugs mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs

Reply via email to