As a long-time TextMate user I hate to say it, but the TextMate 2 saga is a modern-day poster child for why the "throw it all out and start again, it'll be done when it's done" model of refactoring is rarely a good idea.

Users want those "incremental features" that Allan mentioned in this post, and don't care too much what the underlying architecture looks like. It's hard to believe that it wouldn't have been possible to deliver those features whilst iteratively refactoring the underlying architecture.