I'm not very sure about the static_cast<T::type*> ones.

And that's exactly why LWG haven't expressed any desire to rush into this.

One of the major implementations hasn't even implemented the feature yet.

I'm not very sure about the static_cast<T::type*> ones.

And that's exactly why LWG haven't expressed any desire to rush into this.

One of the major implementations hasn't even implemented the feature yet.

GCC and MSVC accept this. Clang's implementation under review has test cases for this case https://reviews.llvm.org/D53847:

dynamic_cast<T::tclass *>(new B)
const_cast<const T::type *>(Ptr)

I think I understand what does "smallest enclosing" mean here now; http://wg21.link/p1787 said this instead:

A qualified or unqualified name is said to be in a type-only context if it is the terminal name of [...]
a type-specifier of a [...] type-id of a static_­cast, const_­cast, reinterpret_­cast, or dynamic_­cast
[temp.res.general]/4.2

See also https://www.open-std.org/jtc1/sc22/wg21/docs/papers/2020/p2150r0.html
My intent is to revisit this paper as an editorial NB comment against C++23, where LWG review each option to omit typename in different contexts, before setting the editorial policy. I expect this will be too disruptive for C++23 at this point, but hope to get the policy review to apply to the C++26 Draft immediately it is created, before any new work starts.

Could you please split this up into smaller commits or PRs? Please reopen if and when you want to resume this.