> K&R

I've read a lot of books about programming over the many decades I've been doing this, and the K&R book still stands out as one of the best examples of how to do a good job of documenting the tool you've just built. The K&R book gave me the heuristic of "read the book by the folks who wrote the language". That heuristic turned out to be mostly wrong. But the book has stood the test of time.