It's hard to get users to migrate to FileContext because they (rightly!) want their software to work against Hadoop 1.x which only has FileSystem. Backporting FileContext to branch-1 would allow people to migrate to FileContext and still work against both Hadoop 1.x and 2.x (or whatever we call it). It probably isn't that much work since FileContext is mostly net new code with lots of tests. It's a pain to support the same code in two places, but we already have to do that with FileSystem, and that's the cost of introducing a new API instead of improving FileSystem in place.