Reference section of the original "Boosting initialization time of each training sample given equal weight 1/n, and then use this algorithm to the training set training t wheel, every time after the training, the training of failure sample attached to the larger weight)"

Don't quite understand is
1. When the weight spanning tree on earth have what effect? The weight is assigned to which part of the training sample and training sample has many characteristics, the weight is given to all the features? Assume I'm a CART, the weights of how to affect the division of the tree?
2. The training samples are used to generate the tree, the training sample of failure "is what meaning, training good tree is not to the test of the test set to yao, how to use training set to test again?