Abstract. Modern web browsers implement a private browsing mode that is intended to leave behind no traces of a user's browsing activity on their computer. This feature is in direct tension with support for extensions, which can silently void this guarantee. We create a static type system to analyze JavaScript extensions for observation of private browsing mode. Using this type system, extension authors and app stores can convince themselves of an extension's safety for private browsing mode. In addition, some extensions intentionally violate the private browsing guarantee; our type system accommodates this with a small annotation overhead, proportional to the degree of violation. These annotations let code auditors narrow their focus to a small fraction of the extension's codebase. We have retrofitted type annotations to Firefox's a p is and to a sample of actively used Firefox extensions. We used the type system to verify several extensions as safe, find actual bugs in several others (most of which have been confirmed by their authors), and find dubious behavior in the rest. Firefox 20, released April 2, 2013, implements a finer-grained private browsing mode; we sketch both the new challenges in this implementation and how our approach can handle them.

Abstract. The jQuery library defines a powerful query language for web applications' scripts to interact with Web page content. This language is exposed as jQuery's a p i, which is implemented to fail silently so that incorrect queries will not cause the program to halt. Since the correctness of a query depends on the structure of a page, discrepancies between the page's actual structure and what the query expects will also result in failure, but with no error traces to indicate where the mismatch occurred. This work proposes a novel type system to statically detect jQuery errors. The type system extends Typed JavaScript with local structure about the page and with multiplicities about the structure of containers. Together, these two extensions allow us to track precisely which nodes are active in a jQuery object, with minimal programmer annotation effort. We evaluate this work by applying it to sample real-world jQuery programs.