[Bug c++/126811] New: [14/15/16/17 Regression] c++: ICE in tsubst when a constexpr init contains an NSDMI

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=126811

            Bug ID: 126811
           Summary: [14/15/16/17 Regression] c++: ICE in tsubst when a
                    constexpr init contains an NSDMI
           Product: gcc
           Version: 17.0
            Status: UNCONFIRMED
          Keywords: rejects-valid
          Severity: normal
          Priority: P3
         Component: c++
          Assignee: unassigned at gcc dot gnu.org
          Reporter: iains at gcc dot gnu.org
  Target Milestone: ---

struct S { double m = 1; };
template<class T> void g() { constexpr S s = {}; (void) s; }

(shows on some release branches with -fchecking, however, that might be
misleading some interaction between the branch build conditions)

-std=c++20 -fchecking -fsyntax-only
(probably -std=c++14 is sufficient to make the class init an aggregate).

Looking at the code, it seems that the substitution was never expecting this -
the NSDMI legitimately contains a FLOAT_EXPR and tsubst has no case for it.  

The question is whether tsubst should grow handling for that (and other
similar) cases - or whether the callers need adjustment... 

====

<source>:2:47: internal compiler error: in tsubst_expr, at cp/pt.cc:23634
    2 | template<class T> void g() { constexpr S s = {}; (void) s; }
      |                                               ^
0x2bae358 diagnostics::context::diagnostic_impl(rich_location*,
diagnostics::metadata const*, diagnostics::option_id, char const*,
__va_list_tag (*) [1], diagnostics::kind)
        ???:0
0x2ba2f9b internal_error(char const*, ...)
        ???:0
0xb427ee fancy_abort(char const*, int, char const*)
        ???:0
0xea8d87 store_init_value(tree_node*, tree_node*, vec<tree_node*, va_gc,
vl_embed>**, int)
        ???:0
0xc541c2 cp_finish_decl(tree_node*, tree_node*, bool, tree_node*, int,
cp_decomp*)
        ???:0
0xd98493 c_parse_file()
        ???:0
0xf2b969 c_common_parse_file()
        ???:0