https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122609
Bug ID: 122609
Summary: modules and class type structured bindings
Product: gcc
Version: 16.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c++
Assignee: unassigned at gcc dot gnu.org
Reporter: nshead at gcc dot gnu.org
Blocks: 103524
Target Milestone: ---
It appears that the search for the tuple customisation points in a template
instantiation when processing a structured binding doesn't work nicely.
Consider the following set of files:
// a.cpp
module;
namespace std {
using size_t = decltype(sizeof(0));
template <typename T> struct tuple_size;
template <size_t N, typename T> struct tuple_element;
}
export module std;
namespace std {
export using std::size_t;
export using std::tuple_size;
export using std::tuple_element;
}
// b.cpp
export module X;
import std; // marking 'export import' fixed the issue
struct S {};
template <> struct std::tuple_size<S> {
static constexpr std::size_t value = 2;
};
template <std::size_t I> struct std::tuple_element<I, S> {
using type = int;
};
template <int N> int get(S) { return N; }
export template <typename T>
void foo() {
S s;
auto [a, b] = s;
}
template void foo<int>(); // works
// c.cpp
import X;
int main() {
foo<double>(); // errors
}
$ g++ -Wno-global-module -fmodules -S [abc].cpp
In module X, imported at c.cpp:1:
b.cpp: In instantiation of ‘void foo@X() [with T = double]’:
required from here
c.cpp:3:14:
3 | foo<double>(); // errors
| ~~~~~~~~~~~^~
b.cpp:16:8: error: cannot decompose class type ‘S@X’ without non-static data
members
16 | auto [a, b] = s;
| ^~~~~~
Unfortunately, by a (somewhat hostile) reading of the standard I think this
example can be considered invalid. In this scenario, from 'c.cpp' name lookup
for 'std::tuple_size' fails, and so [dcl.struct.bind] p7 does not apply. But I
think we should make sure to support this for QoI purposes. (Maybe there's a
wording issue here somewhere...)
Referenced Bugs:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103524
[Bug 103524] [meta-bug] modules issue
