https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71965
Bug ID: 71965
Summary: [7 regression] [concepts] Substitution error *after*
failure to satisfy an earlier constraint
Product: gcc
Version: 7.0
Status: UNCONFIRMED
Severity: minor
Priority: P3
Component: c++
Assignee: unassigned at gcc dot gnu.org
Reporter: Casey at Carter dot net
Target Milestone: ---
Created attachment 38949
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=38949&action=edit
Repro
p121R0 [temp.constr.op]/2 states: "... The satisfaction of a conjunction’s
operands are evaluated left-to-right; if the left operand is not satisfied,
template arguments are not substituted into the right operand, and the
constraint is not satisfied. ..."
So this program:
template <class T>
concept bool Destructible() {
return false;
}
template <class T, class...Args>
concept bool ConstructibleObject =
Destructible<T>() && requires (Args&&...args) {
new T{ (Args&&)args... };
};
int main() {
using T = int[2][2];
static_assert(!ConstructibleObject<T, T>);
}
Should compile without diagnostics, as is the case with GCC 6. GCC 7 (r238631),
however, diagnoses substitution failure in the "new" expression despite that
the "Destructible<T>()" constraint is unsatisfied:
casey@Semiregular:~$ ~/gcc-r238631/bin/g++ -std=c++1z -fconcepts
dev/cmcstl2/repro.cpp
dev/cmcstl2/repro.cpp: In function ‘int main()’:
dev/cmcstl2/repro.cpp:9:9: sorry, unimplemented: cannot initialize
multi-dimensional array with initializer
new T{ (Args&&)args... };
^~~~~~~~~~~~~~~~~~~~~~~~
"Hiding" the requires clause inside another named concept is sufficient to
workaround the problem.
