| Issue |
75893
|
| Summary |
Clang gives incorrect error about constraint depending upon itself
|
| Labels |
clang
|
| Assignees |
|
| Reporter |
jacobsa
|
Here is a simple program where we construct a `std::function` from a callable type with a templated constructor. The constructor is compatible with any input where it can create its own copy from the input forwarding reference.
```c++
#include <functional>
#include <type_traits>
struct Callable {
Callable() = default;
Callable(const Callable&) = default;
Callable& operator=(const Callable&) = default;
template <typename F>
requires requires {
// We don't want this constructor to interfere with the copy
// constructor.
requires(!std::is_same_v<Callable, std::decay_t<F>>);
// It must be possible to make our own copy of the input from
// the input reference.
requires std::is_constructible_v<std::decay_t<F>, F>;
}
Callable(F&& f);
int operator()();
};
void foo() { std::function<int()> f = Callable{}; }
```
GCC [accepts](https://godbolt.org/z/PchvYqG68) this program as valid with `-std=c++20`, but clang [rejects it](https://godbolt.org/z/YzxE68Pfn) at trunk using `-std=c++20 -stdlib=libc++`, with the following error:
```
<source>:10:14: error: satisfaction of constraint 'requires { requires (!std::is_same_v<Callable, std::decay_t<F> >); requires std::is_constructible_v<std::decay_t<F>, F>; }' depends on itself
10 | requires requires {
| ^~~~~~~~~~
[...]
<snip>compressed_pair.h:129:22: note: while checking constraint satisfaction for template 'Callable<std::tuple<const Callable &>>' required here
```
But it's not true that this constraint depends upon itself. If you substitute in `std::tuple<const Callable &>` for `F`, you get:
```c++
requires(!std::is_same_v<Callable, std::decay_t<std::tuple<const Callable &>>>);
requires std::is_constructible_v<
std::decay_t<std::tuple<const Callable &>>,
std::tuple<const Callable &>>;
```
Neither of these should depend on anything to do with that constructor, as far as I can tell.
Note that clang [_does_ accept](https://godbolt.org/z/hs7s17zPs) this program if we don't use `-stdlib=libc++`. It's possible this is a bug in libc++, or it's just an accident of how the type traits are implemented in libc++, triggering a compiler bug that we don't see otherwise.
![]()
_______________________________________________
llvm-bugs mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
