rfe: conwert SymMemb to smart references

[stsp]

c++23 will have smart references.
This will allow us to turn the sym wrappers
into a proxy objects.

[stsp]

Which would mean a gcc compatibility hopefully.
But its quite a big rework.
Whether or not the gcc compatibility worth
the trouble, is an open question. And when
this is going to be implemented in gcc, is difficult
to say.
clang just works...

Related material:
http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2017/p0352r1.pdf
http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2016/p0416r1.pdf

[stsp]

As became obvious from the failed #166
attempt, such wrappers should be
squashed into AsmXXX classes.

[stsp]

#56 (comment)
#56 (comment)

If we can use smart references, perhaps
the inability to modify temporaries will be
circumvented by relaying an access via an
internal pointer to the child object, rather
than the memcpy() to the inherited self.
#56 will use the explicit lifetime management,
so such pointers might be possible.
It however seems that smart references
are not happening in C++23.

[stsp]

If we can use smart references, perhaps
the inability to modify temporaries will be
circumvented by relaying an access via an
internal pointer to the child object, rather
than the memcpy() to the inherited self.

Not going to work, as in
this case the parent lookup
would still only work via
farhlp. Direct parent lookup
will not became possible,
making this proposal rather
useless.
So even here we'd still need
the memcpy() to inherited
self.

[stsp]

would still only work via
farhlp. Direct parent lookup
will not became possible,
making this proposal rather
useless.

And once farhlp_sta is removed
in #259, the only use-case for
this will be the _DOS_FP uses,
which are just 4 right now.
Of course if we could emit the
wrapper for dotted child, rather
than parent, then that would
still be interesting eg for replacing
#257, but that won't happen.

So I guess once #259 is landed,
this should be closed.

[stsp]

Likely no longer interesting.

[stsp]

A new paper have suddenly
appeared on that topic:
https://www.open-std.org/jtc1/sc22/wg21/docs/papers/2024/p3298r0.pdf
But with the new architecture
of assignable temporaries, it
is not clear if this proposal is
still needed here, and where
to apply it.
Will it allow to get rid of _DOS_FP()
macro? Likely not.
Will it allow to get rid of 1 of
the 2 sym wrappers? Probably,
although its a very small reward.
With the removal of farhlp_sta,
its probably absolutely uninteresting.
But the paper is much more complex
than the previous ones, and must
at least be carefully studied.