Comment by xmcqdpt2 | Hacker News Reader

Comment by xmcqdpt2 | Hacker News Reader

While I agree that people generally think of version control (and git) in terms of diff and patch, I think it's important to note that git itself doesn't have diff and patches. The fact that git doesn't have diffs is one of its defining characteristics.

https://git-scm.com/book/en/v2/Getting-Started-What-is-Git%3...

It makes git significantly faster (because it doesn't apply sequence of patches when checking out) and much more robust than a patch and diff system.

I think it's an important point here because it does play a bit of a counterpoint to the OP's argument. In fact, the git storage system is precisely very simple because it's content addressed only and has no notion of history or difference,

https://git-scm.com/book/en/v2/Git-Internals-Git-Objects

In the article, the author mentions something about Alice and Bob sharing only differences in JSON has a way to simplify changes. If anything the lesson from git's success should be that the aurhor's approach (the diff/patch way of Subversion etc.) is exactly the wrong one. Communicating only with full, valid JSONs and showing diffs only as a convenience to the user would be the actual "git way".

But actually git does use diff and patch. It just doesn't store objects as diffs and patches. Whenever you do a cherry pick operation or rebase operation it synthesises patches. You can try yourself by cherry-picking a merge commit. What even does this mean? Well, git doesn't know either because it can't figure out how to synthesise the patch unless you tell it.

> If anything the lesson from git's success should be that the aurhor's approach (the diff/patch way of Subversion etc.) is exactly the wrong one.

The problem with subversion was not the mechanism of storage of objects. And the strength of git doesn't lie here either. It's rather the very flexible nature of its commit metadata and it's design which builds on the idea of multiple masters. This was what was stifling in subversion.

The method of storage whether of the deltas, or the individual states is trivially equivalent. You can inter-convert them if you'd like, so clearly this can't be a game changer.

But even if it were and git's approach was "The Right Thing", this would not be a lesson to learn from git when dealing with data storage because it scales poorly.

Every state transition would have to re-represent the entire state of the database. This will work fine for code, but for databases with tons of objects changing and being extended all the time it would be crazy. It makes a lot more sense to store the differences.

Reply on news.ycombinator.com