I'm still torn about the need for a __file__ attribute. Assuming it's more likely that people write code that *reads* stdlib source for some reason (maybe printing more context for error messages) than *writing* it (which would be truly strange), the amount of 3rd party code that might break due to the lack of __file__ can be expected to be larger than the amount of code (or number of users) that would be confused to having a __file__ pointing to a file that isn't actually read by the interpreter.

Okay, so I'm no longer torn. We should set __file__ for frozen modules based on the pre-computed location of the stdlib.