[Bug middle-end/86284] New: Insert trap instruction in place of missing return statement on dodgy code

[ktkachov at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86284

            Bug ID: 86284
           Summary: Insert trap instruction in place of missing return
                    statement on dodgy code
           Product: gcc
           Version: unknown
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: middle-end
          Assignee: unassigned at gcc dot gnu.org
          Reporter: ktkachov at gcc dot gnu.org
  Target Milestone: ---

Consider a non-void function that's missing a return statement:

int test()
{
  volatile int i = 0;
}

g++ at -O2 for aarch64 will generate:
_Z4testv:
.LFB0:
        .cfi_startproc
        sub     sp, sp, #16
        .cfi_def_cfa_offset 16
        str     wzr, [sp, 12]
        .cfi_endproc
.LFE0:
        .size   _Z4testv, .-_Z4testv

No return statement. That's fine (the code is undefined), and GCC even warns
about it.

We got a suggestion that GCC could insert a trap instruction at the end of such
a function so that at runtime we get a trap rather than the execution running
off into the abyss.

GCC already inserts traps in unreachable paths in gimple-ssa-isolate-paths.c,
so maybe we can also do this when we already detect that a function is broken
in this way?