Guido van Rossum has never been a big fan of this idea, and he recently unloaded a pile of reasoning as to why. Much of this really boils down to the unsuitability of existing Python implementations as a platform for a capability version of the language, though clearly there are language features that must go, too. There’s more on this point from tav, but perhaps his idea of translating Capability Python into Cajita is a more fruitful course…
Anyway, what intrigued me more than the specifics was this statement from Guido
The only differences are at the library level: you cannot write to the filesystem, you cannot create sockets or pipes, you cannot create threads or processes, and certain built-in modules that would support backdoors have been disabled (in a few cases, only the insecure APIs of a module have been disabled, retaining some useful APIs that are deemed safe). All these are eminently reasonable constraints given the goal of App Engine. And yet almost every one of these restrictions has caused severe pain for some of our users.
Securing App Engine has required a significant use of internal resources, and yet the result is still quite limiting. Now consider that App Engine’s security model is much simpler than that preferred by capability enthusiasts: it’s an all-or-nothing model that pretty much only protects Google from being attacked by rogue developers (though it also helps to prevent developers from attacking each other). Extrapolating, I expect that a serious capability-based Python would require much more effort to secure, and yet would place many more constraints on developers. It would have to have a very attractive “killer feature” to make developers want to use it…
There are two important mistakes in this.
Firstly, capability enthusiasts don’t prefer a security model in the sense that Guido is suggesting; we prefer a way of enforcing a security model. App Engine does this enforcement through layers of sandboxing whereas capability languages do it by not providing the untrusted code with the undesirable capabilities. Of course, a side effect of this approach is that capabilities allow far more subtle security models (e.g. “you can only write this part of the file system” or “you can only write files a user has specifically designated” or “you can create sockets, but only for these destinations”) without much extra work and so capability enthusiasts have a tendency to talk about and think in terms of those subtler models. However, Guido’s all-or-nothing model can be implemented easily with capabilities – we don’t have to be subtle if he doesn’t want us to be!
This fallacy causes the second error – because the security model does not have to be subtler, there’s no particular reason to imagine it should take any longer to implement. Nor need it place many extra constraints on developers (I will concede that it must place some constraints because not all of Python is capability-safe). Developers are really only constrained by capability languages in the intended sense: they can’t do the things we don’t want them to do. If the security models are the same, the constraints will be the same, regardless of whether you use sandboxes or capabilities.
Incidentally, I tried to sell the idea of capabilities to the App Engine team several years ago. Given how far we’ve come with Caja in a year, working on a language that is definitely less suited to capabilities than Python is, I would be very surprised if we could not have done the same for Python by now.