[feegeea] Reviewing exercise submission. #19
Description
Link to the PR in the PyHeat Repository
What did the AI tool do nicely?
- the README.md in my opinion is a bit lengthy, but overall quite beautifully written
- as far as I could tell, the generated code is correct, even though I myself don't know much about heat equation. I found it impressive, that the code was correct in the first attempt
- the pyproject.toml was also correct
Where did it fail?
- the LLM was very inconsistent about the style of the docstrings and I had to ask for correct, restructuredText-style ones multiple times
- the LLM was inconsistent regarding the type hints
- I dont like the format of the code, but this could easily be fixed by adding a linter and formatter like ruff, and defining a Workflow for pull requests, where it gets executed
Comparing the two reviews, what did Copilot focus on, what did you focus on?
- Given that I was completely unfamiliar with heat equations, I focused more on the overall structure and style of the code
- Copilot focused a lot on error handling and correct functionality, which was helpful in this case
- a few suggestions were nonsensical, e.g. the part about the dummy information in the LICENSE part of the README
- Overall, I am quite impressed with the review Copilot gave
Examples from your Challenge project
Note: by far most of the PRs came from the main maintainer, who merged without approval / a review
Examples for a PR with review/approval and another one
- I find the reviews quite shallow to be honest. Judging from the reviews, I get the feeling, that the contributors and reviewers work together at Bosch Research and talk to each other anyways, so not many words are needed
- I like, that the reviewers also focus on documentation / readability and code quality aspects like method names, docstrings, correct links to the documentation, et cetera