The code that generates the completions [1] and displays them [2] is fairly simple. It uses a mix of scanning for a function's docstring, Python's built-in rlcompleter module [3] to generate the possible identifiers and keywords, its own module scanner for imports [4].
Hi. It's done using ncurses for the display (if you've ever worked with curses you can imagine how much of a headache the whole thing was). The introspection is done using the `inspect` module for the arg spec/doc string, and either inspect.getmembers or just dir() to get the attributes (it's been a while since I wrote it). We had to do some pretty weird stuff to get the attribute inspection to not trigger __getattribute__ for objects that implement it (i.e. to introspect without side-effects) but it seems to work pretty well.
