For more technical precision in Rust (not because it actually changes anything), += will use AddAssign rather than Add, if it’s implemented, which mutates in-place, whereas `a = a + b` would move twice (though in a way that will always optimise to the same thing). This means you’re actually invoking

  impl AddAssign<&str> for String {
      #[inline]
      fn add_assign(&mut self, other: &str) {
          self.push_str(other);
      }
  }

which the doc comment notes “has the same behavior as the `push_str` method”.

And this shows yet another option: add a separate overload for +=. Python does actually have that in the form of the __iadd__ magic method, and I presume CPython’s optimisation could be implemented that way, though it doesn’t look to be (and this might not have quite the same semantics, I’m not sure):

  class str:
      def __iadd__(self, other):
          if sys.getrefcount(self) == probably 2, it’s fiddly:
              mutate self
          raise NotImplemented