> you should prefer different functions for each case instead.

No I shouldn't. I use this frequently in c++ to deduplicate common functionality that only differs by a single line. The "flag" is a zero-size struct and checking the flag is done at compile time.

  struct do_bar_tag {}
  template<typename do_bar_t>
  int foo(int x, do_bar_t) {
    if constexpr (is_same<do_bar_t, do_bar_tag>::value) {
      bar(x);
    }
    ...
  }

The compiler will emit two implementations, as you suggest is proper, but my maintenance overhead is reduced. And the cost is zero, in time and space, at runtime.

Of course, this isn't always the right thing to do, and you can definitely overdo it. But in moderation, it's fine. "Code smell" is too often a needlessly strong opinion.