offa
commented
3 years ago Thanks for reporting! I can reproduce the issue using your example on GCC and Clang. From a first look into P0052 your example should compile.
I was able to compile the code with all combinations of noexcept / not-noexcet by providing a copy-overload for Wrapper::reset(), used in unique_resource's move-assignement operator:
void reset(const Wrapper<T>& newValue) noexcept(std::is_nothrow_assignable_v<T, const T&>)
{
value = newValue.value;
}Moreover, it seems there are some test cases missing …
Hi, and thank you for this implementation!
I've been playing around with it a little bit and found what seems to be a bug, but I'm not familiar enough with templates or this implementation to know whether I'm possibly using it incorrectly (or which fix is appropriate, if it is a bug).
I've found that when using a unique_resource with resource/deleter move operations that are not noexcept, I hit a build error since detail::Wrapper does not have a copy constructor and we don't .get() the value out of the other wrapper when we can't use std::move.
https://godbolt.org/z/9Px1Ec is a somewhat-minimal reproduction of the issue I'm seeing (based on the head of master). By removing either/both of the "noexcept" values on lines 715-716, it will generate errors in the unique_resource move assignment. I've found that I can get things building by adding .get() calls or by adding a copy constructor to detail::Wrapper, but I'm not sure which approach is better.
If I'm just using it incorrectly and no change is necessary, I'm sorry for the false bug report! :)