A C/C++ program defect automatic repair method based on deep learning and program synthesis

Abstract

The invention relates to a C / C++ program defect automatic repair method based on depth learning and program synthesis, which comprises the following steps of: entering a C / C++ program with defects, error positioning information, a protocol that the program needs to satisfy, and a correct program set that meets the same protocol, establishing a correct writing structure model by learning the writing structure in the correct program; according to the error positioning information, using the code structure before the error point as the input of the model, predicting the writing structure of the error point, expanding the writing structure into a repaired option set, and organizing the repaired option set in the form of a selection expression; using the program synthesis method to convert theprogram with a selection expression and the protocol which the program needs to meet into a logic expression, and using a constraint solver for solving in a set time; according to the option of each selection expression, outputting the final C / C++ code to obtain the repaired program. According to the method, the common defects of the C / C++ program are repaired in the time complexity of a polynomial, and the correctness of the repaired program is ensured.

Description

technical field [0001] The invention relates to a method for automatically repairing program defects, which uses program synthesis and deep learning methods to realize automatic repair of program defects, and belongs to the field of software engineering. Background technique [0002] In the field of computer science, program synthesis refers to the automatic construction of programs from a programming language so that they meet specific specifications. A common practice is to search the program space to generate programs that satisfy the constraints. Program synthesis is often used in code completion, software development, and algorithm design. [0003] In computer science, the concept of deep learning originated from the study of artificial neural networks. Deep learning combines low-level features to form more abstract high-level representation attribute categories or features to discover distributed feature representations of data. Deep learning is widely used in natur...

AI Technical Summary

Problems solved by technology

Through the verification method of the test case, the correctness of the repaired code cannot be guaranteed

Images