Automatic test code generation from acceptance test cases for large-scale software products

Abstract

The process of evaluating whether or not a software system is conformed to the requirement specifications plays an important role in a large-scale software environment. The test engineers identify the test scenarios and implement the code according to the software standards which enables the verification of required criteria for the product delivery. In the development model adopted by several companies, including Ericsson, the requirements for a product are defined first on a high level. These requirements are analysed and transformed into test plans or test cases which are tested after the product development stage. But today, Ericsson has adopted a new practice of software development which is the test-driven development process where the test cases are defined before the software is fully developed. The test engineers then write executable code for the test cases and test the product for different scenarios. However, in the context of large-scale software development with multiple products and teams, it is not trivial to implement code for all the test cases. Today, the common way of performing acceptance testing is through manual labour by understanding the requirements, implementing the test scripts and execute them. The developers create tests to determine if the requirements are met with the contract. This serves as one of the challenges as it is a time consuming process. Thereby, it is important to automate the testing process to rely less on the test engineers and sufficiently serve the purpose. Today there are more advanced technologies and resources which could help in developing such a tool that can reduce the costs and time for the company. This project has as its goal to automate the process of translating the test description to test code. The approach involves the use of natural language processing techniques and other deep learning techniques for analyzing the the test case specification written in natural language and generate the corresponding executable test code. This test code follows the syntax of Robot Test Automation Framework. However, as every test scenario involves different parameters to consider, the aim is to generate the code with suitable functions and be user-friendly allowing human experts’ adjustments to add configurations and parameters.