While neural sequence generation models achieve initial su c-cess for many NLP applications, the canonical decoding procedure with left-to-right generation order (i.e., autoreg res-sive) in one-pass can not reflect the true nature of human revising a sentence to obtain a refined result. In this work, we propose XL-Editor, a novel training framework that enables state-of-the-art generalized autoregressive pretrainin g methods, XLNet specifically, to revise a given sentence by the variable-length insertion probability. Concretely, XL-E ditor can (1) estimate the probability of inserting a variable-le ngth sequence into a specific position of a given sentence; (2) execute post-editing operations such as insertion, deletion, and replacement based on the estimated variable-length insert ion probability; (3) complement existing sequence-to-sequen ce models to refine the generated sequences. Empirically, we first demonstrate better post-editing capabilities of XL-E ditor over XLNet on the text insertion and deletion tasks, which validates the effectiveness of our proposed framework. Fur - thermore, we extend XL-Editor to the unpaired text style transfer task, where transferring the target style onto a gi ven sentence can be naturally viewed as post-editing the senten ce into the target style. XL-Editor achieves significant impro ve-ment in style transfer accuracy and also maintains coherent semantic of the original sentence, showing the broad applic ability of our method.