[Bug c/85563] New: -Wmaybe-uninitialized false alarm regression with __builtin_unreachable and GCC 8

[eggert at cs dot ucla.edu]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85563

            Bug ID: 85563
           Summary: -Wmaybe-uninitialized false alarm regression with
                    __builtin_unreachable and GCC 8
           Product: gcc
           Version: 8.0.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
          Assignee: unassigned at gcc dot gnu.org
          Reporter: eggert at cs dot ucla.edu
  Target Milestone: ---

When compiling Emacs with gcc (GCC) 8.0.1 20180324 (Red Hat 8.0.1-0.20) x86-64
I noticed a regression compared to previous GCC versions, a regression that
causes a false alarm. Briefly, Emacs uses a statement like this:

  ((CONSP (Vframe_list)) ? (void) 0 : __builtin_unreachable ());

to tell Emacs that Vframe_list is a cons (CONSP is inline and side effect
free), and then follows this with a loop:

  for ((tail) = Vframe_list;
       (CONSP (tail) && (frame1 = XCAR (tail), 1));
       tail = XCDR (tail))
     { loop body }

which obviously must initialize frame1. In a later use of frame1, though, GCC
8.0.1 incorrectly complains:

f.i: In function ‘delete_frame’:
f.i:69688:7: warning: ‘frame1’ may be used uninitialized in this function
[-Wmaybe-uninitialized]
       do_switch_frame (frame1, 0, 1, builtin_lisp_symbol (0));
       ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

where GCC 7 and earlier do not complain. To reproduce the problem, use the
command:

gcc -O2 -S -Wmaybe-uninitialized f.i

on the attached file f.i. Although strictly speaking this isn't a bug since the
"may be used uninitialized" is deliberately wishy-washy, still, it's an
annoyance that I'll have to fiddle with Emacs source to pacify the misguided
compiler here.