[Bug c/85872] New: False positive for -Wmaybe-unitialized

[jakub.leszczak at nokia dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85872

            Bug ID: 85872
           Summary: False positive for -Wmaybe-unitialized
           Product: gcc
           Version: 8.1.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
          Assignee: unassigned at gcc dot gnu.org
          Reporter: jakub.leszczak at nokia dot com
  Target Milestone: ---

Created attachment 44161
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=44161&action=edit
File contains informations about the bug and environent (uname, gcc --version,
*.i files etc.)

There is a false positive for maybe-unitialized when using -O1 (but not for
-O0).

Minimal case I got after stripping code:
```
int test1(int *t);

int test(int *t, int e)
{
        int i, f = 0;

        if (e != 0) {
                for(i = 0; t[i] != e; i++);
        } else {
                f = 1;
        }

        int j;
        if (test1(&j)) return 1;
        if (test1(&j)) return 1;
        if (test1(&j)) return 1;
        if (test1(&j)) return 1;
        if (test1(&j)) return 1;

#if TEST
        if (test1(&j)) return 1;
#endif

        if (!f) {
                return t[i];
        }

        return 0;
}
```

doesn't compile with `$ gcc -Wall -Werror -O1 -DTEST=1 -c bug.c`.  But it
compiles when either I turn off optimizations or set TEST=0.  For some reason
adding one additional `if (test1(&j)) return 1;` makes a difference.

Tested on arch linux, ubuntu 16.04 and gcc.godbolt.org, on multiple gcc
versions: 5.4.0, 7.3.1, 8.1.0, aarch64-linux-gnu-gcc6.4.1 and
arm-linux-gnueabihf-gcc4.7.3.