TODO: Remove all this bullshit like rng.h, keep pointer, current boxed, and util; keep the error stuff and get a consistent framework for it but then work on the ruleset, rule definitions, making it ergonomic and `constinit`, etc. The actual real reason this project exists...
Then add things like `iterator<T>` to use for parsing, and `argument_{input/output}_stream` for transformations, etc. See notes on phone for custom handlers as well. Use as much aggregate (POD) init as possible, e.g:
```
// with rule ctor for `<size_t DN>(char8_t, const char &(desc)[DN], kind, attribute = {})`
//TODO: Overload case for `unique_ptr<T, D>` for `deleter` to apply (and contain) `D()` on back_cast()ed pointer T* instead of assuming default `delete`.
structdeleterfinal/*: public ErasedTypeDeleter XXX <- Is this useful, for unerasing the type, maybe?*/{
/// Base class which is used to store the layout of Box<T>'s allocated memory. Intended for use to determine if a certain `T`'s `this` pointer is inside a `Box<T>` or not.
static_assert(util::shares_layout<Box<_Complexdouble>,_Complexdouble*>,"Invalid Box<T> memory layout: More than just T*");
static_assert(util::shares_layout<Box<_Complexdouble>,std::unique_ptr<_Complexdouble>>,"Invalid Box<T> memory layout: More than just std::unique_ptr<T>");
template<typenameT,typenameR/*XXX: Maybe remove this, and put the requires after the definition, using `auto&& value` instead of `R&& value`*/>requires(std::derived_from<R,T>)
constexprvirtualtypes::MaybeOwned<Report>into_report()noexcept{/* TODO */}//TODO: Maybe just use `std::optional<std::reference_wrapper<Error>>`? Or Box<Report&>?
constexprvirtualtypes::MaybeOwned<Report>into_report()&noexcept{/* TODO */}//TODO: Maybe just use `std::optional<std::reference_wrapper<Error>>`? Or Box<Report&>?
{returntyped::Box<TracedError>{std::move(*this)};}//TODO: use `boxed::is_boxed_value()` to check if *this is boxed... Uhh, actually... XXX: No that doesn't work, since `Box<T>` doesn't make `T` visible to `boxed::ObjectBase`.... Eh...
#ifdef DEBUG /* We don't want to instantiate `null_optimise<>` *at all* if we don't have to outside of actual real use in `Option<T>`. This check is for development only since Option<Box<T>> is a bit of an awkward developmental case.
// Throws NullDerefException<?> for first type of `ptr` that is null (if any are).
// If: `All == true`: Instead will always throw `NullDerefException<T...> (all types) if any are, and will not assume every pointer is likely to be non-null.
/// Deducable version of `is_empty_pack<typename...>()`
///
/// Can be used as `is_empty_pack(pack...)` instead of the slightly more ugly `is_empty_pack<decltype(pack)...>()` though with (very unlikely, but possible) implications of actually "evaluating" the pack (despite the values themselves are `const&` and never used, they're still passed.)
/// HOWEVER: The latter should likely be preferred, since this function takes possible-nonconstexpr types as arguments, it cannot be immediate like the non-argument taking overload. See below comment for more information on that.
template<typename...A>
[[gnu::const]]//TODO: XXX: This mostly useless function won't mess up overload resolution for `is_empty_pack<types...>()`, right? If it will, either find a way to make sure it comes LAST, or remove it, it's not worth the hassle just so users (i.e. me) can avoid typeing `decltype` for once in my fuckin life... what a stupud keyword name srsl (see static_assert() below.)
andis_empty_pack<>()//XXX: This might also be ambiguous between the type and value one... Eh.... That's actually useful though, so removing this explicit (and pointless) *usage* of the function from the test would be fine if they don't clash in other ways too.
Avril