It's kind of true with git-lfs.

I wonder if anyone has built a system with the philosophy of the opposite extreme.

"Oh yeah, space is cheap, so I just let my automated system make 10,000 commits per second. It'll be fine."

I have been thinking about this. Why no record all keystrokes. Then checkpoint where you are whenever you desire. These days AI could describe any arbitrary span of code changes.

>Why not record all keystrokes?

That was Google Wave, it stored everything as an "operational transform". The presentation they used when showing it off to the world was brain damaged, they insisted as showing it off as a ton of separate blocks, instead of one continuous document, which gave people the wrong impression of it, and tanked it.

In a similar way, the impression that Git stores deltas, got stuck in my head somehow at the beginning and made it an opaque mystery for me.

It was only years later, when I somehow I managed to learn the truth, that Git stores full snapshots, and except for compression in extreme cases to save storage, it doesn't do anything with deltas (except fake them to show diffs)

It's not quite "all keystrokes", but git-wip is meant to add a commit every time you save a file: https://github.com/bartman/git-wip I've been using this (via magit-wip) for a while now and it stays out of the way. I've never needed to use it, though. I'm probably too careful now having been trained in a world without git-wip.

Magit stopped using the git-wip script ten years ago.