As LDC needs it for its DMD-style inline asm parser. It was recently
converted to extern(D) in dlang/dmd!15728.

…/x86 kernel zone

… (dlang/dmd!15791)

This preserves the `CatAssignExp` in the AST until the glue layer where
an error is printed in case this expression is used with `-betterC`.
This is required to happen in the glue layer as the semantic analysis
doesn't correctly distinguish when code needs to be generated.

Signed-off-by: Teodor Dutu <teodor.dutu@gmail.com>

…ith align(N) > platform stack alignment (dlang/dmd!15820)

See: https://en.wikipedia.org/wiki/CPUID#EAX=80000008h:_Virtual_and_Physical_address_Sizes

This *might* fix Issue 24254, although I'd expect the read value for
that CPU to be 127 (*physical* cores minus 1), not the problematic 255.

…g/dmd!15861)

This assert is hit when compiling a private codebase. I'm not sure
how it comes about, but there are apparently cases where the primary
template instance has a non-root `minst`, but has been added to a
root module's members already (so `memberOf` NOT identical to `minst`),
at the time the primary instance's `minst` is swapped-out with a
root `minst` from a sibling instance (in `templateInstanceSemantic()`).

Previously, the primary instance was appended to 2 root modules in
such a case, most likely causing it to be codegen'd twice.

v2.106.0

To fix new fail_compilation/test24159.d.