Why DocComment? And How It Helps You Read Code Faster And Innovate Faster?

As a senior developer/architect who has been in software development industry for more than ten years, I know there's a lot of code to read. Good code, bad code, normal code, and AI code. After I've understand these code, I can make modifications on it, apply it in other places, and come up with my own code. The question is, the progress to understand those code is not always happy. Good code might have good comments, but most code are not that good. There are functions with hundreds lines of code without any comment. There are code that comes with bad variable names. There are code that is hard to understand in the business view. And the percentage of code generated by AI keeps growing and growing, making it impossible to find the person who wrote it. All this clutter could mean understanding existing code takes a lot of time. And that is both a cost to pay and might also delay time-to-market for business.

And for most developers, we sometimes don't have the time for writing our own comment, is there a tool for us to help us understand others' code and the other way around? That is when I started this exploration.

When Copilot was out, I was happy to learn that Copilot provides a function for generating comments. However, after using it, it really disappoints me.

First of all, the comments are not accurate enough. The commenting mechanism is just a generic prompt to LLM so it does not really incites the LLM think, rather it is just a follow-up of tokens based on previously trained based model. So it does not cover for some cases.

Secondly, the comments are not suitable for reading. The comments generated by Copilot is either too broad or too fine for the level of details. Thus could lead to clutter of unnecessary information, which could slow the reading of code rather than make it fast.

Thirdly, the comments intrude our original code. While writing code with inline comments is the best practice, sometimes we just don't have that much time to audit all the things generated by AI. But the comments generated by Copilot is inline comments. That could take a lot of time to audit.

So that's when I decided to develop DocComment. It will make you understand code faster. It will help you generate comments that applies to regulations. These are the things that we have done:

Firstly, using code structural data and LLM, DocComment not only understand the code snippet, but also the whole code. Thus it can generate more accurate comments for code.

Secondly, the comments grain of detail is carefully controlled, so all the comments generated are coarse enough for human to provide most necessary information; but also detail enough to quickly understand nitty-gritty of a few lines. All of this leads to a final goal: READ CODE FASTER.

Thirdly, DocComment does not intrude your code; it just comes in tandem with your code. So the level of burden to audit does not lies on you. Besides, combining it with git repo will make you understand the whole code repo better.

This is DocComment, a code tandem reader to help you understand code faster. It helps me a lot when I was learning new code this year. And I hope it helps you, too. Try it out with free sign up and let me know what you think.