CWG3218 [temp.variadic] Expanding packs created within a pack expansion

[Halalaluyafail3]

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

Reference (section label): [temp.variadic]

Issue description:

Consider the following:

#include<tuple>
void bar(auto...){}
void foo(auto x){
    bar([&]{auto[...y]=x;return y;}()...);
}
int main(){
    foo(std::tuple());
}

Clang rejects this, GCC is OK with it except for the actual instantiation of foo which causes an ICE. A second example is with fold expressions:

#include<tuple>
void baz(auto z){
    (...,[&]{auto[...w]=z;w;}());
}
int main(){
    baz(std::tuple());
}

There does not seem to be anything that forbids these examples, even though I do not think it was intended to allow them.

[jensmaurer]

This problem should also appear for init-capture packs, right?

It seems the pack in question already went out-of-scope, yet the rule about "contains an unexpanded pack" is formally satisfied.

[feedab1e]

Yes, and also for arguments, so

void foo(auto) { ([]<auto ...xs>{xs;}, ...); }

and

void bar(auto) { ([](auto ...xs){xs;}, ...); }

are also problematic.

[feedab1e]

There is also an adjacent case of

struct x { int a, b, c; };
template<int> void foo() {
  auto [...xs] = x{xs...};
}

where the pack is actually still in scope, but the initializer still depends on the contents of the pack.

[jensmaurer]

CWG3218

[Halalaluyafail3]

There is also an adjacent case of

struct x { int a, b, c; };
template<int> void foo() {
  auto [...xs] = x{xs...};
}

where the pack is actually still in scope, but the initializer still depends on the contents of the pack.

I believe this is ill-formed because of "If the initializer refers to one of the names introduced by the structured binding declaration, the program is ill-formed." [dcl.struct.bind]/5.