Potentially drop support for `vk::UniqueHandles` in the future

[M2-TE]

As shortly mentioned in #2358 (comment), use cases for vk::UniqueHandle could be superceded by the more versatile vk::raii::Handles. This might warrant deprecating them and eventually removing them from the headers.

One concern I wrote about in the abovementioned thread is that vk::UniqueHandles use the normal static/dynamic dispatchers, whereas vk::raii::Handles apparently use their own. The former could therefore be much more easily mixed together with the normal vk::Handles. It would be great if someone with more experience using the raii handles could comment on this.

[chrisprobst]

The newer vk::raii:: function loading mechanism is just inherently better and less complex than the old way of loading functions. vk::raii:: takes the pain out of things like loading multiple instances at once, which Volk can do but not cleanly (the device table is still loaded from a global function). I wish vk::raii:: would be more widespread and accepted. But the reality is, I guess, the few projects that have adopted Vulkan-Hpp actually use UniqueHandles (if at all) and therefore the old loading mechanism. In any way, dropping the old way and enforcing the new one means code breaks. Not sure if UniqueHandles could implemented with vk::raii:: under the hood? So that it would work transparently?

[M2-TE]

vk::raii still carries the downside of needing to keep additional pointers when compared to the simple vk::Handles. But yes, the function loading in raii is just much more comfortable to use. At least compared to unique handles, raii has no downside.

I am currently trying out a few ways to implement a sort of unified dispatch loader, which may unify standard and raii function loading.

[chrisprobst]

This sounds nice, good luck! Thanks for moving this library forward.

[Wunkolo]

I've personally only ever used the vk::Unique* handle-types and have had a good experience with them for expressing things like ownership and lifetimes of objects and having RAII-style clean-up and never actually used the vk::raii types. I had assumed they were doing something somewhat different to the Unique handles for them both to exist at the same time for all this time.
Would migrating to the vk::raii types basically be a drop-in replacement?

[asuessenbach]

Would migrating to the vk::raii types basically be a drop-in replacement?

That's a good question. Will make some experiments to figure out.

[pc-john]

Just please, do not break compatibility with the old code. I am the one that uses old Unique* classes, including their memory overhead they inherently carry with them. vk::raii approach does not improve this memory overhead, or actually, it is making it even worse. Few bytes is usually not a big deal unless huge number of objects is created. Anyway, it is overhead.