[span.cons] thinks that to_address() can throw, but it can't

[StephanTLavavej]

N4842 22.7.3.2 [span.cons]/6 and /9 say: "Throws: When and what to_address(first) throws."

But 20.10.2 [memory.syn] and 20.10.4 [pointer.conversion] require both overloads of to_address() to be noexcept:

template<class T> constexpr T* to_address(T* p) noexcept;
template<class Ptr> auto to_address(const Ptr& p) noexcept;

Therefore, [span.cons]/6 and [span.cons]/9 should be removed editorially.

[CaseyCarter]

[span.cons]/9 fails to account for the fact that last - first can potentially throw. We should probably file an LWG issue to handle all of this.

EDIT: And by "we", I probably mean "me".

[CaseyCarter]

This is now LWG 3358.

[jensmaurer]

Looks like we can editorially replace the phrases about to_address with "Throws: Nothing.", but of course the "last-first" issue needs to be addressed non-editorially.

[burblebee]

The changes applied to fix this (in PR #3546) conflict with those approved in LWG3358 (the drafting note in LWG3358 suggests that only p6 should have been modified). I assume LWG3358 is what we really want here?

[jwakely]

The changes applied here fix the editorial issue that "Throws: [thing that cannot throw]" should be "Throws: Nothing.". LWG 3358 addresses the normative defect that last - first should be considered. So they do conflict, but both changes are correct 🙂

The LWG 3358 changes supersede this editorial change and are what we want.

[CaseyCarter]

+1 to Jonathan's comments, and apologies for not rebasing the wording of LWG 3358 onto the result of the editorial change to avoid this confusion in the first place.