[Bug c++/92003] [8/9/10 Regression] constexpr-ness of char const* doesn't propagate

[jason at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92003

--- Comment #3 from Jason Merrill <jason at gcc dot gnu.org> ---
Created attachment 47787
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=47787&action=edit
proposed fix

The change in C++17 is that static constexpr data members are now implicitly
'inline', so no out-of-class definition is needed, and we use a weak symbol on
ELF.  From this the symtab code was concluding that we can't be sure that it
has a non-zero address.  Here's a proposed fix.