https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121411
--- Comment #2 from David Faust <david.faust at oracle dot com> --- Created attachment 62061 --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=62061&action=edit proposed patch Use unsigned HOST_WIDE_INT (which should always be 64 bits) when calculating the array bounds so we can at least know that we overflowed uint32. The value stored in DW_AT_upper_bound can be >32-bits. Seems like falling back to CTF_K_UNKNOWN is the only safe option since anything else misrepresents the array size with no way for a CTF consumer to know. If we want a useful warning though I think we'll have to pass down source location info from the variable/member/whatever decl. is using the problematic type. Otherwise we don't have a location nor name to use, and the diagnostic from warning () points to the last source line.
