| Issue |
126600
|
| Summary |
False positive lifetime GSL warning for no-capture lambda?
|
| Labels |
clang:diagnostics
|
| Assignees |
|
| Reporter |
smeenai
|
For the code in https://godbolt.org/z/v919qqe6o:
```
struct [[gsl::Pointer]] function_ref {
template <typename Callable>
function_ref(Callable &&callable [[clang::lifetimebound]]) : ref(callable) {}
void (*ref)();
};
void f();
void g(function_ref &r) {
r = []() { f(); };
}
```
Clang trunk produces a warning:
```
warning: object backing the pointer r will be destroyed at the end of the full-_expression_ [-Wdangling-assignment-gsl]
9 | r = []() { f(); };
```
Clang seems to consider the no-capture lambda as a temporary that will be destroyed at the end of the _expression_. I'm not a standards expert and I haven't found anything definitive through searching, but in practice no-capture lambdas will turn into ordinary functions, so there's no actual issue here. Should we actually be warning for this?
https://github.com/llvm/llvm-project/commit/a3b4d9147406cbd90090466a9b2b9bb2e9f6000c is a commit that ran into this and changed the code slightly to avoid it (the code change made there would result in a stack-use-after-free were this an actual issue, but it avoids the warning). https://github.com/llvm/llvm-project/pull/126140 is a high-firing instance of this warning, and we'd appreciate some guidance.
![]()
_______________________________________________
llvm-bugs mailing list
llvm-bugs@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
