Comment by zahlman
I don't know why people look at paragraphs of documentation that explain the exact results in strange edge cases (which have to exist because of the underlying complexity; in this example, Python can't change how IEEE-754 works, nor the original C printf specifier they're emulating) and conclude that this proves some strange and unexpected complexity was introduced.
When documentation isn't this thorough, people complain about that, too.