CWG3048 [stmt.expand] Zero "iterations" only possible for iterating and enumerating expansion statements

[jakubjelinek]

Full name of submitter (unless configured in github; will be published with the issue): Jakub Jelinek

Reference (section label): stmt.expand

Link to reflector thread (if any):

Issue description: Right now both iterating and enumerating expansion statements can have zero "iterations", the former if begin is the same as end, the latter for {}. But for destructuring expansion statements, I believe it has to have at least one iteration, because a valid structured binding declaration needs to have at least one name. While with structured binding packs one can have zero, i.e. if ihttps://eel.is/c++draft/stmt.expand#5.3 was instead constexpropt auto&& [...u] = expansion-initializer; and for-range-declaration = u...[i]; then it would allow also zero "itereations". Though, structured binding packs can only appear in templates, so such a change isn't possible, it would need to be some special case.

Suggested resolution:

[jensmaurer]

I'm reading the above as "the specification of destructuring expansion statements needs a special case for N=0, because the syntactic rewrite is not well-formed for that case".

[jakubjelinek]

Right. Whether it needs it or not (and can stay as is) will defer to CWG.

[jensmaurer]

CWG3048

[t3nsor]

The section label is wrong, it should be [stmt.expand].

I think this resolution isn't quite right because it doesn't check all the semantic constraints that an empty structured binding declaration would have (if written using a pack in a templated context). For example, it doesn't check constexpr if present and it doesn't try to instantiate std::tuple_size<E>. I think we're better off trying to find a way to say that we still treat it as a structured binding declaration even though it would be ill-formed to write it directly.

[jensmaurer]

In order to determine N, we need to instantiate tuple_size, I think. [dcl.struct.bind] p7

[jakubjelinek]

Yes. For the constexpr if present it could just be
constexpr_opt auto && range = expansion-initializer;
in the N == 0 case instead of (void) expansion-initializer;. Though, I think N should be introduced before it is used in the N == 0 case.

[jensmaurer]

I've fixed CWG3048. I haven't done anything about "N"; the general pattern is "rewrite ... where N is ..."
so N is already introduced too late.

[t3nsor]

In order to determine N, we need to instantiate tuple_size, I think. [dcl.struct.bind] p7

Right, my bad. It looks like [dcl.struct.bind] checks all the constraints before telling us the structured binding size, so the proposed resolution looks good now.