rsmith added a comment.
In https://reviews.llvm.org/D47607#1118464, @EricWF wrote:
> One other concern I have is if there will be any performance impact to using
> `launder` all the time, but better safe than sorry.
In the short-to-medium term, I think we can tackle this by adding an attribute:
// ...
struct pair {
T first;
[[clang::not_actually_const]] U second;
};
... to directly express that the `second` field of a non-const `pair` is never
a const subobject, even if `U` is a `const`-qualified type. I think we'll need
this once we head down the path of `constexpr` containers, otherwise
`node_handle`s are not going to work in constant expression evaluation.
Then we can remove the `launder` calls for implementations that support the
attribute.
https://reviews.llvm.org/D47607
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
