https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104487
Bug ID: 104487 Summary: The substitution in the dependent name in a trailing return type should cause recursive instantiation Product: gcc Version: 11.1.0 Status: UNCONFIRMED Severity: normal Priority: P3 Component: c++ Assignee: unassigned at gcc dot gnu.org Reporter: xmh970252187 at gmail dot com Target Milestone: --- #include <iostream> struct C{}; template<class T> struct anotherType{ using type = C; }; template<class T> auto fun(T)->decltype(fun(typename anotherType<T>::type{})){ //#1 return 0; } int fun(C){ // #2 return 0; } int main(){ fun(0); // #3 } GCC incorrectly accepts this case and uses `fun<int>(int)` for the function call at `#3`. The dependent name `fun` in the trailing return type will be looked up in the instantiation context, since the type of the argument is class type `C`, ADL would find `#1` and `#2` for that dependent name. Overload resolution will perform again, this causes the substitution for `#1` again, which will repeatedly perform the aforementioned step. The instantiation should have exceeded the limits of the implementation. The weird thing is that the following variant example can cause that error: #include <iostream> struct C{}; template<class T> struct anotherType{ using type = C; }; template<class T> auto fun(T)->decltype(fun(typename anotherType<T>::type{})){ return 0; } int fun(int){ // in the above example, parameter type is `C` return 0; } int main(){ fun(C{}); // in the above example, the argument is of type int } I do not find any explicit difference for looking up the dependent name since the argument in the dependent name is always type `C` regardless of the argument in the call at `#3`.