Authors:
(1) Xu Zhou, College of Computer Science, Wuhan University, Wuhan, China ([email protected]);
(2) Ping Liang (author), College of Computer Science, Wuhan University, Wuhan, China ([email protected]);
(3) Becky Chang, College of Computer Science, Wuhan University, Wuhan, China ([email protected]);
(4) Zengyang Li, College of Computer Science, Central China University, Wuhan, China ([email protected]);
(5) Aakash Ahmed, College of Computing and Communications, University of Lancaster Leipzig, Leipzig, Germany ([email protected]);
(6) Mojtaba Shahin, College of Computing Technologies, RMIT University, Melbourne, Australia ([email protected]);
(7) Mohamed Wasim, College of Information Technology, University of Gifksel, Jevskil, Finland ([email protected]).
Links table
Abstract and 1. Introduction
2. Methodology and 2.1. Search questions
2.2. Collection
2.3. Place a sign of data
2.4. Data extraction
2.5. Data analysis
3. Results and interpretation and 3.1. Type of problems (RQ1)
3.2. Type of causes (RQ2)
3.3. Solutions type (RQ3)
4. The consequences
4.1. The effects of Copilot users
4.2. The effects of Copilot team
4.3. The effects of researchers
5. Justice threats
6. Related work
6.1. Evaluating the quality of code created by Copilot
6.2. Copilot effect on practical development and 6.3. Conclusive summary
7. Conclusions, data, approval, a statement of credit and references contribution
a summary
With the latest advances of artificial intelligence (AI) and LLMS models, AIBASED code generates a practical solution to software development. GitHub Copilot, a AI’s pair programmer, uses automated learning models trained on a large group of code scraps to create symbol suggestions using natural language processing. Despite its popularity in the development of software, there is limited experimental evidence for the actual experiences of practitioners working with Copilot. To this end, we conducted a experimental study to understand the problems faced by practitioners when using Copilot, as well as their basic reasons and possible solutions. We collected data from 476 GitHub issue, 706 GitHub discussions, and 142 publications in the flow. Our results reveal that (1) the operating problem and the issue of compatibility are the most common problems faced by COPILOT users, (2) The internal error of the tattoo, the network connection mixing, editing the problem of the editor/IDE compatibility as the most common causes, (3) errors that were repaired by COPILOT, the composition/preparation modification, and the use of the appropriate version are the prevailing solutions. Based on the results, we discuss the potential areas of COPILOT for reinforcement, and provide the effects of Copilot users, COPILOT team and researchers.
1. Introduction
In the development of software, developers seek to achieve automation and intelligence that most code can be created automatically with the minimum human coding efforts. Several studies (EG, Luan Et Al LLMS models are a type of natural language processing technique that depends on deep learning that is able to learn grammatical rules, connotations, and pragmatism of the language, and generate a wide range of contents. Recently, with the rapid development of LLMS, the tools that are trained in large quantities of code are increasingly increased in the spotlight (for example, AI-UAGMENted in Gartner Trends 2024 (GARTNER, 2023), which makes it possible for programmers to generate a code automat et al., 2021).
On June 29, 2021, GitHub and Openai jointly announced the launch of a new product called GitHub Copilot (GitHub, 2024C). This innovative tool is operated by an openai manuscript, which is a large -scale nervous network model that is trained in a huge database of the source code and the text of the natural language. The goal of GitHub Copilot is to provide automatic completions and advanced obstetrics of developers, and works effectively as “AI’s husband programmer” that can help in actual time coding tasks. Copilot is designed to work with a wide range of integrated development environments (IDES) and software instructions, such as VSCODE, Visual Studio, Neovim, and Jetbrains (GitHub, 2024C). By collecting contextual information such as job names and comments, Copilot is able to create code for code in a variety of programming languages (for example, Python, C ++, Java), which can improve developers productivity and help them complete more efficient coding tasks (IMAI, 2022).
Since its release, Copilot has gained great attention within the developer community, and a total of 1.3 million paid users was paid until February 2024 (Wilkinson, 2024). Several studies define effectiveness and interests on the potential impact of the security of code and intellectual property (Pearce et al Some previous research has investigated the quality of the code created by Copilot (Letissire Et Al
However, there is currently a lack of the systematic classification of problems that arise during the practical use of COPILOT from the perspective of developers, as well as the reasons behind them and their treatment solutions. To this end, we conducted a comprehensive analysis of the problems
In the face of software developers when coding with GitHub Copilot, as well as their causes and solutions, by collecting data from GitHub issues, GitHub discussions, excessive job functions, which would help understand Copilot restrictions in practical settings.
The results we find: (1) The issue of the operation and the issue of compatibility are the most common problems faced by developers(2) Copilot internal error, a network connection error, and the issue of the editor’s compatibility/IDE as the most common causesAnd (3) Copilot errors, composition/setting modification, and using the appropriate version are the prevailing solutions.
The contributions of this work are:
• We presented two levels of COPILOT problems in the practice of software development.
• We have developed a single -level classification for causes of problems and solutions to address problems.
• We attracted a maps fee of specific problems for their causes and solutions.
• We suggested practical instructions for COPILOT users, COPILOT and other researchers.
The rest of this paper is organized as follows: Section 2 displays research questions (RQS) and the research process. Section 3 provides and interpreting results. Section 4 discusses the effects of search results. Section 5 shows the potential threats to the validity of this study. Section 6 reviews the relevant work. Finally, Section 7 concludes this work along with future directions.
