Using ChatGPT - Genie AI in VS Code to Explain, Refactor and Test Rails Code

I have used GitHub Copilot for a long time and found it saves me time. In this April 2023 tutorial, I explore a VS Code extension while waiting for access to the newly announced GitHub Copilot X. The extension I chose, ChatGPT - Genie AI, worked immediately with my Ruby on Rails project, whereas another extension I tried did not.

At the time, Genie required an OpenAI API key and appeared to support GPT-3.5; I expected GPT-4 support to follow, but that was a prediction, not a feature I demonstrated. After installing it, open the Genie icon in the VS Code sidebar. Its interface offers a prompt box, temperature choices, saved shortcuts, and six actions for selected code.

The buttons add tests, find potential problems, optimize, explain, add comments and complete code. I have found test generation particularly useful. For optimization, I have previously given it code I knew was unnecessarily complex and received a shorter, clearer version that passed my tests; that experience does not guarantee every generated change preserves behavior.

Explain can help when entering an unfamiliar codebase: select a method or file and ask for an explanation. I have not extensively tested the comment generator, and I wonder how much the ability to ask for explanations changes the need for comments. Code completion resembles a function GitHub Copilot already offered.

First I select a Rails webhook endpoint model. It uses the AASM gem to manage states such as active, paused, archived and disabled. Genie explains the state machine and its events, and I find the description accurate. The example suggests a way to navigate unfamiliar Ruby code, including code that uses metaprogramming.

Next I examine an AssociatedPublicIds file, which creates public IDs for associated models so the API does not expose sequential database IDs. I select the code and ask Genie to explain it. Its response identifies methods returning public IDs from associated records, which matches my intent.

I ask Genie to optimize the file. The resulting version consolidates similar has-many, belongs-to and has-one logic, reducing the displayed code from 45 lines to 21. It looks cleaner to me, but shorter code alone does not establish equivalence: generated changes still need review and tests.

The Find Problems action flags possible issues: dynamic constant lookup could be risky if a class name comes from user input; dynamically defined methods can make maintenance harder; and the safe-navigation operator requires Ruby 2.3 or later. I agree that the metaprogramming may be too fancy, and I learned about the version constraint. Genie suggested Module.const_get as a safer alternative, but changing the receiver alone does not validate untrusted input; its suggestions need scrutiny.

I then select the code and ask Genie to add tests. One generated example asserts that a customer’s company public ID equals the associated company’s public ID, which is the relationship I expect. The generated test is a useful starting point; the demonstration shows the code rather than a full test run.

The interface provides Precise, Balanced and Creative temperature options. Temperature affects the variability of responses; I guess that the presets correspond to values such as zero and 0.5, but I do not verify their exact settings. The star button lets you save custom prompts as shortcuts, such as an instruction to implement tests.

My major concern is context. In the interface I can select the code passed to a prompt, but I have not figured out how to ask questions with the whole codebase as context. I speculate that Genie might have more project context, while acknowledging I could be wrong; the video does not demonstrate that it sees the full repository.

I would like a tool that can answer questions across the entire project rather than only the open file and highlighted code. Even with that limitation, I find this extension useful, and I hope the forthcoming Copilot X will address some of these rough edges.