Am I the only one who think Light Table won't work for Python?
I mean, look at all those getattr() setattr() stuff. LightTable doesn't stand a chance. Even dealing with some very primitive Django models.
I mean, look at all those getattr() setattr() stuff. LightTable doesn't stand a chance. Even dealing with some very primitive Django models.
I programmed in Smalltalk for years. Smalltalk has similar issues: dynamic typing, dynamic method generation/handling via #doesNotUnderstand:, and so on. Yet quite a few Smalltalk environments have accurate and useful autocompletion, on-the-fly documentation, and more.
How? You get three freebies that end up covering most situations:
1. Globals (which includes classes) are trivial to detect and centrally stored,
making it easy to pick up when a class is used in source code. So we've
now got autocomplete for all static methods and class documentation.
2. Most of the time, a variable is only actually assigned to once--or,
worst case, from one location. After that, it's merely manipulated.
See what class is getting instantiated there (see above), and you've got
a great guess on what type you're dealing with. (And, sadly, it's a
guess; "Foo new" *probably* gives you a new Foo, but it doesn't have to.)
3. Most methods *either* only get implemented once, *or* specialize a
method whose parent calls "self subclassResponsibility". So either
you've only got a single place to look-up, or you can hedge your bets
and show documentation for the version of the method that calls
"self subclassResponsibility" (i.e., the abstract one) and figure it's
applicable to the concrete instance.
Past that? Things get really hard. Because Smalltalk's development and execution cycle is blurry, you can actually go further than the above rules by checking what types variables are in instantiated classes and methods, and turning your source-based guesses into statistics-based ones. There's no equivalent mechanism in Python. You could probably do something with an idempotent test suite and some hooks into the Python VM, but the problem gets much more complicated. Doable? Yes. Likely to come from Light Table? I'm not honestly sure, but I am sure that they'd have to be willing to get pretty dirty with Python to get there.
You can get a long way by being slightly smart about things, instead of just relying on a naive static analysis. (I mean "naive" in the sense of ... well, not in the insulting sense :P)