Skip to content

reviewing_exercise.md

Latest commit

 

History

History
94 lines (61 loc) · 5.86 KB

reviewing_exercise.md

File metadata and controls

94 lines (61 loc) · 5.86 KB
Raw

Reviewing code

Let us get familiar with code review tools and practices.

Deadline: December 17, 2025, 9:00

Note: In this exercise, we use GenAI tools, which evolve rapidly. Feedback and suggestions are very welcome - open an issue.

Prerequisites

  • Login to the Responsible AI
  • Get access to GitHub Education. After verification, according to GitHub, it might take up to 72h for the features to be available. If for any reason you cannot or do not want sign up for it, you can try any other code reviewing tool you might have available (e.g., Copilot in VSCode). For the grading, we do not need to see the output of the tool, but only your comments on your experience with it.

Overview

In this exercise, you will work responsibly with AI tools to write code, review it as it if was a colleague's code, and use more AI tools for an additional review. More specifically, you will:

  1. use Responsible AI to write a simulation program,
  2. create a repository and publish the first implementation as a pull request to your main branch,
  3. review the code as if it was written by another human,
  4. use GitHub copilot to conduct an automated review,
  5. compare the two reviews to each other and to examples from your challenge project.

In the end, you will need to open an issue in the exercise repository with name [username] Reviewing exercise submission.

Generate code

Generating code can be very useful to kick-start a project or to fill rather clear code parts (such as basic tests or documentation). With the right prompts and some iterations, one can create functional programs.

Important note: Always check for mistakes and check everything you do not understand with the respective documentation. You are entering dangerous territory: AI tools have particular strengths, but also weaknesses, and they will very often confidently give wrong answers (and code with bugs), which might not be immediately visibly wrong. You remain in control and responsible for your projects.

Now that we have set some guidelines, let's see the strengths in action:

  1. Login to the Responsible AI provided by the University of Stuttgart.

    • If the first prompt does not work, start by saying "Hi" before proceeding with actual prompts. Apparently, bots also have psychological needs. :)

  2. In your own words, ask the tool to generate code for solving a heat equation, in a programming language you understand (e.g., Python). Ask it to take parameters from a configuration file and write results to an output file. Ask for some documentation, tests, or anything else you would like to see in such a code.

  3. Create a repository on GitHub (under your namespace) and initialize it with a README.md file.

  4. Add the files that RAI created to the repository, in a branch called feature. In your commit message, clearly disclose the tool. For example:

     (... usual commit message ...)

 I acknowledge the use of RAI (https://rai.uni-stuttgart.de/) to implement the entirety of this code.

 Prompts:
 1. I would like a Python script that solves the heat equation.
 2. ...

    The more details you add, the better.

  5. Open a merge request from the branch feature to the main branch. Make sure to give a good PR title and description, as if you were contributing this code, and focusing on what this PR adds; not only the AI generation.

Review the code

Assume that this code was written by another human being. Be kind and thorough, and leave some comments in the pull request. Make sure to:

  • Describe what you understand this PR does.
  • Does the code run? Does it do the right thing?
  • Mention what you like about it
  • Add inline comments and at least one inline code suggestion
  • Question parts that do not make sense to you - ask for clarifications. Are the "why" and "how" clear?
  • Could parts of the code be made simpler?
  • Comment on individual aspects such as documentation, testing, design, corner cases, and potential bugs. See the Google Reviewer Guidelines for some ideas.

Your goal with any code review is to help the author improve the contribution to a state that can be accepted into the project.

Start an issue in the exercise repository and add a link to your repository and this PR. Give a summary: what did the AI tool do nicely, where did it fail?

Get an automated review

GitHub Copilot can automatically review a pull request. Add it as a reviewer and wait for it to submit a review.

Carefully go through the review. Comment on the suggestions, apply suggestions you agree with, comment on the points you don't agree with.

Optional: Give custom instructions to Copilot (such as code reviewer). You could directly ask Copilot (agent) to apply specific suggestions in the same branch or in separate PRs.

In the issue you just opened, edit the description and comment:

  • Comparing the two reviews, what did Copilot focus on, what did you focus on?
  • What did you learn from the suggestions?

Examples from your Challenge project

In the issue you just opened, edit the description and:

  • Add 1-3 examples of pull/merge requests with extensive review.
  • Comment: What do reviewers seem to typically focus on?