https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108266
Bug ID: 108266 Summary: [13 Regression] ICE during cxx_eval_constant_expression on IMPLICIT_CONV_EXPR since r13-4564 Product: gcc Version: 13.0 Status: UNCONFIRMED Severity: normal Priority: P3 Component: c++ Assignee: unassigned at gcc dot gnu.org Reporter: jakub at gcc dot gnu.org Target Milestone: --- #include <initializer_list> #include <vector> struct S { S (const char *); }; void bar (std::vector<S>); template <int N> void foo () { bar ({"", ""}); } void baz () { foo<0> (); } ICEs with -std=c++20 starting with r13-4564-gd081807d8d70e3e87eae41e1560e54d503f4d465 internal compiler error: unexpected expression '(S)""' of kind implicit_conv_expr 11 | bar ({"", ""}); | ~~~~^~~~~~~~~~ In Fedora test mass rebuilds I saw 33 ICEs like that. Instead of the includes one can use just namespace std { template <class T> struct initializer_list { const T *ptr; decltype (sizeof 0) len; }; template <typename T> struct vector { vector (initializer_list<T>); }; } to get the same ICE. The problem is that maybe_init_list_as_range -> maybe_init_list_as_array -> maybe_constant_init is called when processing_template_decl (from build_user_type_conversion_1). Wonder if the call shouldn't be skipped for processing_template_decl, or whether say maybe_init_list_as_range shouldn't always return NULL if processing_template_decl.