写点什么

ChatGPT 与软件架构 (1) - 快速原型

作者:俞凡
  • 2023-06-01
    上海
  • 本文字数:3141 字

    阅读完需:约 10 分钟

通过 ChatGPT 生成设计和原型代码,可以帮助团队快速启动项目,验证想法,提高效率。原文: ChatGPT and Software Architecture


Surfing Croyde Bay @Unsplash


OpenAI 的 ChatGPT 现在越来越火,出现了各种有趣用例。


从许多方面来看,ChatGPT 都可以看作是 AI 赋能的架构师白板,除了画画线条和框框,还可以有许多用途。我在本文中将演示如何基于 ChatGPT 启动软件架构流程。


就像在白板上画画一样,过程会有点混乱,各种因素相互作用会导致不得不通过不断修改来找到最佳答案。本文旨在演示成功使用 ChatGPT 的一些提问技巧,这些技巧产生了真正有价值的结果。


需要注意,ChatGPT 仍在学习中,所以不要对一致性有太大期望,多次问同样的问题可能会得到不同的结果。


设置 ChatGPT 帐户很简单,可以按照官方文档完成: https://openai.com/blog/chatgpt/

用户故事

我们从要求 ChatGPT 写用户故事开始。



写一个用户故事,测量员登录到系统,输入地理空间坐标、建筑、建筑类型和特征信息,这是他们一天中要进行的许多调查中的一个


如你所见,这是个特定行业的用户故事。ChatGPT 拥有庞大的知识库,需要通过实验来了解 ChatGPT 的知识深度,即它知道些什么?


第一个问题的关键是要对关键细节非常明确,以获得最佳结果。这样,ChatGPT 可以响应一组体面的结果,如下所示:



如果获得的结果与上述类似,那就意味着 ChatGPT 已经找到了线索,并将根据进一步提出的问题进行构建。要记住的关键是,以下所有结果都将基于这个初始上下文,因此值得花时间定义第一个交互。


现在可以做些非常了不起的事情,构建数据库!

构建数据库


在 markdown 中以表格格式详细生成数据模型


ChatGPT 是这样回应的:



不止如此,我们继续…



生成 SQL


CREATE TABLE Surveyors (    id INT NOT NULL AUTO_INCREMENT,    name VARCHAR(255) NOT NULL,    email VARCHAR(255) NOT NULL,    password VARCHAR(255) NOT NULL,    PRIMARY KEY (id));
CREATE TABLE Buildings ( id INT NOT NULL AUTO_INCREMENT, type_id INT NOT NULL, coordinates VARCHAR(255) NOT NULL, notes TEXT, surveyor_id INT NOT NULL, PRIMARY KEY (id), FOREIGN KEY (type_id) REFERENCES BuildingTypes(id), FOREIGN KEY (surveyor_id) REFERENCES Surveyors(id));
CREATE TABLE BuildingTypes ( id INT NOT NULL AUTO_INCREMENT, type VARCHAR(255) NOT NULL, PRIMARY KEY (id));
复制代码


ChatGPT 输出了完全可用的 SQL,可以顺利创建 SQL 数据库。不止如此……

创建类图


<script src="https://cdn.jsdelivr.net/npm/mermaid@8.6.0/dist/mermaid.min.js"></script><script>mermaid.initialize({startOnLoad:true});</script>
<div class="mermaid"> classDiagram class Surveyors { +id: int +name: string +email: string +password: string } class Buildings { +id: int +type_id: int +coordinates: string +notes: string +surveyor_id: int } class BuildingTypes { +id: int +type: string } Surveyors --> Buildings Buildings --> BuildingTypes</div>
复制代码


ChatGPT 生成的类图如下所示…



ChatGPT 真是让人惊喜:




ChatGPT 没法显示图表,但可以生成能够创建图表的 Mermaid 代码。ChatGPT 非常熟悉 Mermaid 脚本,可以生成这种脚本语言支持的图表。


由此可见,ChatGPT 已经学会了编码,可以继续要求 ChatGPT 编写 JavaScript、CSS 和 HTML,通过简单的请求将原型组合在一起。ChatGPT 经过训练,可以用许多流行的编程语言生成代码。


如果有足够的时间,每个有经验的人都可以做到,只不过速度要慢一些。但这不是生产代码,要使代码达到生产质量和可用性水平,还需要做更多工作。不过 ChatGPT 已经很让人惊艳了。


我们尝试用 ChatGPT 来提高代码质量:



生成遵循 C#实践原则的类设计,包含错误处理和数据库访问层


ChatGPT 的响应:


public class Surveyor{    private readonly string _name;    private readonly string _email;    private readonly List<ISurvey> _surveys;    private readonly IDatabaseAccess _databaseAccess;
public Surveyor(string name, string email, IDatabaseAccess databaseAccess) { _name = name; _email = email; _databaseAccess = databaseAccess; _surveys = new List<ISurvey>(); }
public void AddSurvey(ISurvey survey) { try { _surveys.Add(survey); _databaseAccess.SaveSurvey(survey); } catch (Exception ex) { // Log the error } }
public void RemoveSurvey(ISurvey survey) { try { _surveys.Remove(survey); _databaseAccess.DeleteSurvey(survey); } catch (Exception ex) { // Log the error } }
public List<ISurvey> GetSurveys() { try { return _databaseAccess.GetSurveys(); } catch (Exception ex) { // Log the error return new List<ISurvey>(); } }}
public interface IDatabaseAccess{ void SaveSurvey(ISurvey survey); void DeleteSurvey(ISurvey survey); List<ISurvey> GetSurveys();}
public class DatabaseAccess : IDatabaseAccess{ public void SaveSurvey(ISurvey survey) { // Implementation for saving the survey to a database }
public void DeleteSurvey(ISurvey survey) { // Implementation for deleting the survey from the database }
public List<ISurvey> GetSurveys() { // Implementation for retrieving a list of surveys from the database }}
public interface ISurvey{ decimal Latitude { get; set; } decimal Longitude { get; set; } string BuildingType { get; set; } Dictionary<string, string> BuildingCharacteristics { get; set; }}
public class Survey : ISurvey{ public decimal Latitude { get; set; } public decimal Longitude { get; set; } public string BuildingType { get; set; } public Dictionary<string, string> BuildingCharacteristics { get; set; }
public Survey(decimal latitude, decimal longitude, string buildingType, Dictionary<string, string> buildingCharacteristics) { Latitude = latitude; Longitude = longitude; BuildingType = buildingType; BuildingCharacteristics = buildingCharacteristics; }}
复制代码


作为一组脚手架类,非常不错了。


接下来让 ChatGPT 为 RESP API 创建 Swagger,ChatGPT 再次返回了足以启动开发的不错成果。


结论

ChatGPT 功能广泛,可以帮助启动开发,但还无法实现比较深度的工作。每个回合都必须靠人来激发 ChatGPT 完成任务,缺乏主动性,需要有人发起并推动接下来的步骤。


不过可以开发一组标准问题,促使 ChatGPT 提供好的结果,也许足以帮助团队以多种方式开始开发。




你好,我是俞凡,在 Motorola 做过研发,现在在 Mavenir 做技术工作,对通信、网络、后端架构、云原生、DevOps、CICD、区块链、AI 等技术始终保持着浓厚的兴趣,平时喜欢阅读、思考,相信持续学习、终身成长,欢迎一起交流学习。微信公众号:DeepNoMind

发布于: 2023-06-01阅读数: 53
用户头像

俞凡

关注

公众号:DeepNoMind 2017-10-18 加入

俞凡,Mavenir Systems研发总监,关注高可用架构、高性能服务、5G、人工智能、区块链、DevOps、Agile等。公众号:DeepNoMind

评论 (1 条评论)

发布
用户头像
厉害啊,还可以这样用。
2023-06-01 23:16 · 广东
回复
没有更多了
ChatGPT与软件架构(1) - 快速原型_人工智能_俞凡_InfoQ写作社区