Type variance

I spent a while fixing up the genetic algorithm I wrote a while ago; up until the version that is up now, I didn't really like how it looked and the reason was the type hints. It simply didn't click to me why the classes from the implementation weren't compatible with the protocols in the definition and why I had to create the type variables (which I find really ugly, and hopefully won't be required anymore in the near future). Reading up on how to improve it led me to the concept of variance, of which I only had a vague memory from college.

Writing this note to ensure it doesn't become a vague memory again:

• If something cannot be replaced whatsoever, either by subclass or superclass, it's deemed invariant in its context. An example is identity: It makes much more sense to compare a movie with another movie instead of with a song, even if both cinema and music are considered arts.
• If something can be replaced by a subclass, it's deemed covariant in its context. An example is possession: If you need something with a camera, then any standalone camera, smartphone or tablet will do. A telescope, however, might not.
• If something can be replaced by a superclass, it's deemed contravariant in its context. An example is action: If you can run, it follows that you can do something that only needs for you to walk. Conversely, just because I see you walking I can't assume you can run a marathon.

I spent a while fixing up the genetic algorithm I wrote a while ago; up until the version that is …