Peter Dimov
Peter Dimov
Opt-in with a macro sounds doable, we'll be able to supply the base classes this way, but we still need the member pointers.
> That's not the issue indeed, PFR will support extracting member pointers very soon. Out of curiosity, does this work, and if so, how is it implemented?
There seems to be a significant pushback in the committee against the idea of reflection on attributes. The objection is that attributes today can be silently ignored by the compiler,...
What's the intended use of `boost::hash`?
Aren't you going to store `task`s in the unordered set, though, instead of the raw coroutine_handle? In which case you'll need to specialize `boost::hash` anyway.
The test is overspecified; it should test that the hashes of t and t2 match and that they don't match the hash of t3 (a second task). We don't want...
On the other hand, for transparent hashing purposes, we do want to guarantee that `boost::hash()( h ) == boost::hash()( h.address() )`, and if `std::hash` is used, this won't hold. :-/
That's an interesting idea that can allow us to eliminate the various `hash_value` overloads for standard types, and is apparently also what Abseil does by default. But it's not quite...
The current implementation (on develop) can in principle be `constexpr`; it no longer uses rotations. The problem, however, is that we support forward-declaring `boost::hash_combine` and `boost::hash_range`, so that people can...
`value_init.hpp` uses TypeTraits and Core is not allowed to depend on TypeTraits.