https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107148
Joseph S. Myers <jsm28 at gcc dot gnu.org> changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |jsm28 at gcc dot gnu.org --- Comment #2 from Joseph S. Myers <jsm28 at gcc dot gnu.org> --- The following test produces the same ICE without any prior errors. I'm not sure if a statement expression should be considered valid in this context, or if that indicates that an earlier error is needed. Compile this test with -fimplicit-constexpr (thus, needs GCC 12 or later, so this test does not show a regression, unlike the ICE-on-invalid regression in the original bug report). int f(int); class A { public: A(int); }; class C { C() : m(f(({ int x = 1; x; }))) {} A m; }; Alternatively, not needing -fimplicit-constexpr (rejected by GCC 10 and before; GCC 11 and later ICE; so again not showing a regression): int f(int); class A { public: A(int); }; class C { constexpr C() : m(f(({ int x = 1; x; }))) {} A m; }; The failing assertion is gcc_assert (VAR_P (*p) && DECL_ARTIFICIAL (*p) && !TREE_STATIC (*p)); where *p is the variable x in the statement expression, so fails DECL_ARTIFICIAL. If this test is valid, I don't know if it should reach the code with this assertion (or what additional issues might arise if the variable in the statement expression were declared static).