> It is not surprising to me that the 2nd most famous distributed source control tool - Mercurial - was also initially created by a Linux kernel developer.
Because Linux was in a crisis and hot off an excellent experience w bitkeeper? Or something more profound?
> Because Linux was in a crisis and hot off an excellent experience w bitkeeper? Or something more profound?
I'm sure the crisis was part of it.
But also, the way Linux development is organized is very weird compared to how code is developed in most places. "Distributed" is not quite right, but it's the best word I've got. A source control tool like Perforce just wouldn't fit linux development very well, whereas large companies at the time were perfectly happy with it.
I agree with you that Linux development is weird compared to most places. Most places have a designated central repo that acts as the source of truth, whereas the Linux developers don’t really have this.
This is possibly an unpopular opinion, but I think that when there is an official main repo, git is far more complicated than it needs to be. It’s confusing to a lot of developers because of all this unnecessary complexity. Most teams could probably be served better by a simpler system.
I share this opinion as well. For centralized-repo style, which is what most teams need, the Git model has three problems:
- It is hard to use. There's a reason most teams have a single or two Git "experts" to help everyone else resolve repository / merge / pull / whatever problems that happen regularly. The onboarding experience for developers unfamilir with Git is very bad, and many of them end up never able to use it. If you have non-developers (e.g. artists), it is even worse.
- Binary files. Yes, there is Git LFS, but there are many small integration problems when you try to use it, and the simple fact that Git needs something special to manage binary files is a problem. And no, storing binary files outside the repository is NOT a good solution. In games for example, the binary files and the engine usually evole together and if you want to checkout an old version, the code and the assets need to stay in sync. You want to version the game, not just the engine.
- Partial (subtree) clones and permissions. There's a reason monorepos are popular: they're much simpler to manage. Unfortunaly, because Git barely does partial clones and cannot manage permissions (inside a repository) at all, monorepos are much more limited than in SVN or P4. That's why many projects end up divided in multiple repos, but this brings all sorts of problems and a large administrative overhead for all developers.
The point of being hard to use is valid. But I would focus more on being hard to learn instead. Once you know it you forgot about it 99% of times, for other 1% it's good to have this "git expert" on a team ;)
For other points there seems to be ongoing process [0][1] and I think in one year or more having huge monorepos with binary files should not be such a big problem or require additional tools. It should be build in core git.
Permissions seems to be partially "solved" by git hosting sites like Github. But I agree it's a pity we don't have anything build in core git for this :(
Thanks for the thoughtful reply and kicking off a great discussion about gits strong points and pain points. I see too many discussions that are just “fans” who seem to think Linus/Linux can do no wrong, without acknowledging what Linux development is, and how probably 99% of us will never be there, but still live w the pain points. Like suffering through supercar gas mileage and repair costs for a commuter/grocery-getter, but “McLaren can’t be wrong, can they?”
Because Linux was in a crisis and hot off an excellent experience w bitkeeper? Or something more profound?