* result: try out some experimental shenanigans
* result: sketch out some more shenanigans
* result: see what it looks like to convert kernel to use result conds instead of guards
* make rest of kernel use experimental new macro-ing
* Work around Clang's incomplete C++20 support for omitting typename
* vapours: fix Clang error about missing return in constexpr function
* stratosphere: fix call to non-constexpr strlen in constexpr function
strlen being constexpr is a non-compliant GCC extension; Clang
explicitly rejects it: https://reviews.llvm.org/D23692
* stratosphere: add a bunch of missing override specifiers
* stratosphere: work around Clang consteval bug
Minimal example: https://godbolt.org/z/MoM64v93M
The issue seems to be that Clang does not consider f(x) to be a
constant expression if x comes from a template argument that isn't
a non-type auto template argument (???)
We can work around this by relaxing GetMessageHeaderForCheck (by using
constexpr instead of consteval). This produces no functional changes
because the result of GetMessageHeaderForCheck() is assigned to a
constexpr variable, so the result is guaranteed to be computed
at compile-time.
* stratosphere: fix missing require clauses in definitions
GCC not requiring the require clauses to be repeated for member
definitions is actually a compiler bug:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96830
Clang rejects declarations with missing require clauses.
* Fix ALWAYS_INLINE_LAMBDA and parameter list relative order
While GCC doesn't seem to care about the position of the always_inline
attribute relative to the parameter list, Clang is very picky
and requires the attribute to appear after the parameter list
(and before a trailing return type)
* stratosphere: fix static constexpr member variable with incomplete type
GCC accepts this for some reason (because of the lambda?) but Clang
correctly rejects this.
Nintendo used to do what we were doing because the function wasn't directly in the handler table,
but we've always been directly in the handler table, so we were trashing the last four arguments to light ipc
when called from aarch32. Nothing uses this, but needed to be fixed.
This is an optimization that saves the most common type of virtual call in the kernel (DynamicCast)
by storing class token as a member, rather than getting it via virtual call every time.
This does not currently cost any memory space on 64-bit targets, due to pre-existing padding space.
This optimization can be turned off via a compile-time flag for accuracy.
