I guess that was also a workaround for EDG. We'll need to reduce a repro to send to the Edge team and conditionally (or unconditionally) restore the workaround with a transition comment.

We'll need to reduce a repro to send to the Edge team

Reduced to:

struct S {
    int x, y;
};
void f(void* ptr) {
    S(13, 42); // Ok
    new S(13, 42); // error: no instance of constructor "S::S" matches
}

Which I filed as VSO-2228186 and added to #1621.

I'm mirroring this to the MSVC-internal repo - please notify me if any further changes are pushed.

/clr hates this technique, reduced and reported as VSO-2231268 "/clr /std:c++20 rejects sneaky technique to access private data members".

Instead of guarding this test coverage for /clr, I'm now inclined to abandon the private data member hack, and merge only the TRANSITION comments for EDG. I think this is adding too much test complexity and the cost for product code is minimal.

Thanks for helping us find and report so many compiler bugs! 🐞 ⏱️ 😹