Just imagine a highly trained AI with super coding powers is trained to write an app just from reading your commit messages. How would your apps look like?
Would it be
- a stable running app, running according to the specs
- empty page (with thousand times
updatedoutput to the console) - a complete mess of unrelated files that won't even compile?
- core parts are working and milestone-critical features are either blank (
addedorupdated) or just don't work (got it done in time) - what else?
Top comments (15)
It would be nothing because my commit messages suck.
This is funny to think about.
Realistically, commit messages probably are a really useful labelling tool of some kind for AI, whether or not this task is accomplishable. Whether good or bad, they say something about an app.
I wonder what the correlation between commit message quality and project success is. They are probably correlated, but I could imagine scenarios where maybe some form of sloppiness might have personality trait crossover which helps get a project to market faster? Just spitballing.
Also question is, where concise commit messages are really important (and not just a formality): CI/CD? Onboarding? Revision/Audits? Refactoring?
The idea behind the thought experiment is to me that, if we have enough context and a hypothetical ai that can perfectly transform such context it should be able to recreate the code from the commit messages (a little bit like with event sourcing).
However the gap between this and reality is that we don't even know when commit messages are relevant and what/how much context is relevant.
Maybe this leads to a few creative ideas in the process.
Commit messages are good just when there's someone using
to extract the changelog plus there is a commit template for the team😂
Otherwise I get crappy cryptic messages that you can't understand without reading the issue and sometimes even a 'No messages for this commit' while doing CR and evaluating the commit history of the branch😒
Not much if it's from company code, probably quite a bit if from my private git.
Either way it wouldn't run, nobody describes their git commit messages THAT well. It would be missing context and all changes not worth mentioning
Let's assume a hypothetical scenario where our ai is super powered. How much verbosity and context is required to make this work?
It would result in a very angry program.
Can't wait for it to figure out the "TODO: solve bug" commit.
Our super AI could do that, the question is how it would do it :-D
My app:
while(true)
Print(“wip”)
Kind of opposite use case, but this stuff creates documentation from source code, and it is creepy. marketplace.visualstudio.com/items...
This is very interesting since this is where my thoughts went when thinking about verbosity in commit messages. How much context is required to make commits reproducible by reading their messages.
Imagine an AI can build an entire app/website by only reading your thoughts. That would be scary. Elon who?
They would probably break the app 😂