Looking for more references about stacking models with linear regression!

This was brought up briefly but without any external links. This seems like an excellent approach to a large class of machine learning problems and I would like to learn more. What are the constraints / drawbacks to using this approach? How do you decide on the final “layer” of classification, like the number of tree leaves? what are you using to supervise the learning of the tree leaves?

This is such a fascinating topic and I want much more information.