Yes, Peter, your summary is correct, except that the "do nothing trivial constructor" can be called, but likely won't.

Could you point me to what clarified it for you so I can incorporate it into the paper?

Thanks!


> On Jan 10, 2026, at 12:54 PM, Peter Bindels <dascandy_at_[hidden]> wrote:
>
> Hi Jody,
>
> On Sat, Jan 10, 2026 at 6:29 PM Jody Hagins <coachhagins_at_[hidden] <mailto:coachhagins_at_[hidden]>> wrote:
>> Hi Peter,
>>
>> Thanks for the feedback.
>>
>> I'm writing this off the top of my head, so forgive syntax issues.
>>
>> First, a terminology note: I'm talking about an "implicit-lifetime type," not "trivial lifetime." This distinction matters — it's about whether an object's lifetime can begin *implicitly* (via `memcpy`, `mmap`, `start_lifetime_as`).
>> <snip>
>> Does this help clarify the problem and proposed solution?
>
> Yes, completely.
>
>> I could have just said, "What Breno said."
>
> No, I did not get this distinction from what he said. I was aware of implicit lifetime types but didn't realize the rules were as you laid them out. So:
>
> - Implicit lifetime objects need a trivial constructor, but not necessarily one that you can call, or that you actually do call
> - std::atomic used to have that, but lost it in C++20
> - We want the problem from pre-C++-20 fixed, but implicit lifetime back
> - adding a do-nothing uncallable trivial constructor completely fixes it, so we should do that
>
> Correct?
>
> If so, sounds good to me.
>
> The specific example helps a lot to understand how it should work from a user point of view.
>
> Thanks,
> Peter