P0843R14 `<inplace_vector>`

[StephanTLavavej]

WG21-P0843R14 <inplace_vector>
LWG-4147 Precondition on inplace_vector::emplace
LWG-4396 Improve inplace_vector(from_range_t, R&& rg)

Feature-test macro:

#define __cpp_lib_inplace_vector 202406L

This was also patched by the Core paper WG21-P3074R7 "Trivial unions", which added:

#define __cpp_lib_constexpr_inplace_vector 202502L

Note: We're focused on implementing the remaining library-only features in C++23. Until that's done, we will NOT be accepting PRs for C++26 features.

[vasama]

I have implemented this here: https://github.com/vasama/MSSTL/tree/feature/inplace_vector

When you start accepting C++26 features, I will be happy to reopen and update my PR.
Please comment in this issue so that I receive a notification about it.

[frederick-vs-ja]

A proxy object for container/iterator debugging will make the container totally non-trivial, which is fine for all containers but array and inplace_vector. The problem is more severe for inplace_vector because the valid range of an inplace_vector object can vary among its lifetime.

Do we want to add a proxy object (but shouldn't be dynamically allocated if added) for debugging when it's allowed to do so?

[StephanTLavavej]

Dynamically allocated proxy objects are a bane, and we should avoid adding any more.

There's no point in container-internal proxy objects - their whole purpose is to allow iterators to find their parents after swapping/moving. (Note the distinction between them and sentinel nodes. Container-internal sentinel nodes work because they don't remain valid after swapping/moving.)

[vasama]

In my implementation the debug iterator currently contains a pointer directly to the size member of the inplace_vector, which is a null pointer in the case of zero capacity. This is used by the iterator itself to detect out-of-bounds accesses. The vector itself detects its own iterators using the data pointer also stored in the iterator.