This doesn't look right.

"noexcept" means no exception escapes, but violating a precondition (e.g. when moving a deleter throws) is still undefined behavior.

Can you show which normative statement implies the struck text?

There's none. My assumption was that the precondition existed to prevent users from examining the unique_ptr object after such a throw, or alternatively, to not have to define the state in such a case. That shouldn't be possible after it was marked noexcept, though. I can't think of any other reason why to declare UB, so it doesn't make sense to me that it's still a precondition.

This is definitely not editorial, though. If you agree, I'll close this PR and its issue, and consider submitting an LWG issue.

Well, for once, having to catch all exceptions to call terminate for a noexcept violation is overhead you might want to avoid. (We don't require the deleter's move operation to be declared noexcept, we just require that it does not throw an exception for this particular move invocation.)

Anyway, this is not editorial, so I'd appreciate if you closed this issue and submitted an LWG issue if you believe there is a defect in the specification (I don't believe so).

Well, for once, having to catch all exceptions to call terminate for a noexcept violation is overhead you might want to avoid. (We don't require the deleter's move operation to be declared noexcept, we just require that it does not throw an exception for this particular move invocation.)

The compiler is going to provide thenoexcept guarantee anyway, unless unique_ptr is decorated with some non-standard extension to tell the compiler that an exception cannot happen. In practice, nobody does that, so the terminating branch is still emitted by the compiler (although a good implementation can arrange for the terminate with no overhead anyway).

Anyway, this is not editorial, so I'd appreciate if you closed this issue and submitted an LWG issue if you believe there is a defect in the specification (I don't believe so).

Yes, this should be an LWG issue.

It would be nice to have the link to this draft PULL request as part of the corresponding LWG issue submission text.

It would be nice to have the link to this draft PULL request as part of the corresponding LWG issue submission text.

Would that be the case even though there are more signatures that are noexcept yet have the precondition that an operation doesn't exit with an exception? This PR wouldn't suficce as a Proposed Resolution.

I only recommend that when an LWG issue is submitted as a consequence and strong relation to an issue/pull request from this project, it would be nice to refer to that issue/pull request in the text of the LWG issue submission.