[Bug target/117229] [15 regression] "libcpp, c, middle-end: Optimize initializers using #embed in C" vs. GCN '-march=gfx908'

[tschwinge at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=117229

--- Comment #2 from Thomas Schwinge <tschwinge at gcc dot gnu.org> ---
OK, that's going to be "fun":

(In reply to Jakub Jelinek from comment #1)
> sed -i -e 's|abort ()|__builtin_printf ("%d FAIL\\n", __LINE__)|' 
> c-c++-common/cpp/embed-1.c
> 
> and retry and see what FAILs?

Nothing FAILs -- it PASSes, without any 'FAIL' diagnostic!?  %-O

During "manual bisecting", I found that I can, for example, also make it PASS
this way:

    --- gcc/testsuite/c-c++-common/cpp/embed-1.c
    +++ gcc/testsuite/c-c++-common/cpp/embed-1.c
    @@ -258,9 +258,11 @@ main ()
         abort ();
       if (g.a != 'H' || g.b != 'e' || g.c != 'n' || g.d != 'r')
         abort ();
    +#if 1
       if (!strstr ((const char *) h, "imprisonétur")
           || !strstr ((const char *) h, "blah blah"))
         abort ();
    +#endif
       if (v.a != 42 || v.b.a != 'H' || v.b.b.a != 'e'
           || v.b.c != 'n' || v.c != 'r' || v.d != -42)
         abort ();

(Note always-true '#if 1'!)

Uh.