Git was invented by the Linux project in order to move beyond the crappy solutions they had in place before, and since then it has established itself as the de facto standard for version control and software collaboration.
But I'm wondering if, now that we've lived with git for a while, are there any clear alternate approaches we could be considering for what we do in software development. Git is not exactly the easiest thing to not mess up.
But it can be hard to even imagine an alternative to such an ingrained practice, but I'd love to hear any thoughts on the subject at all.
Top comments (50)
Git is the classic example of something that was invented to solve one problem, but ended up solving a different, more important problem along the way. For that reason, I do think that Git could (nay, will) eventually be supplanted by something better, something that directly addresses the more important problem that Git solved by accident.
To be more specific: when Linus created Git, his goal was to create a version control system that would be fully distributed. You have to understand, he was targeting the Linux kernel, a project which is extremely widely distributed and for which there are many "sources of truth". In other words, the goal was that the Linux Foundation could host a Git server with history, branches, and tags for the "official" Linux kernel, but RedHat or Canonical or Debian might also host Git servers with their own branches, history, and tags for kernels that are just as complete as the "official" version.
Most teams, however, don't have more than one "source of truth". Every time you see people complain that "GitHub is down", what they're really saying is "even though Git is fully distributed and capable of supporting multiple 'authoritative' nodes, we've agreed on using GitHub as the de facto central repository we all depend on".
Why is it, then, that so many people use a version control system when they're not interested in its raison d'être? Because along the way Linus solved a smaller, but much more valuable, issue: local versioning.
Young coders may not remember, but back in the bad ol' days of Subversion, if you wanted to save something you were working on so that you could move on to something else, your choices were: a.) save the file locally, then try and walk back the changes when it came time to
svn commit
, or b.) make sure you are connected to the Subversion server,svn commit
, and pray there are no conflicts with anything anyone else was working on.Actually, there was a 3rd option known as "SVK". On the surface,
svk
worked much likegit
does today: you would keep a copy of the Subversion repository locally, to which you could commit as needed, and periodically you wouldsvk sync
with the main Subversion repository.So, I honestly believe that Git "won" not because it is completetly distributed, but merely because it allows developers to work locally at will, and synchronize centrally only as needed. Really, this is a natural progression from the earliest version control systems where you would have to actually lock a file on the central server before you could edit it, to Subversion which did away with locks but still required access to the central server in order to version changes. Some new system that focused more on the local developer experience, and simplified the "central repository" situation (since the vast majority of teams will only ever have one central repository), could easily eat Git's lunch.
Great reply
Subversion... I used to dream of Subversion. I had to use CVS.
But anyway - Subversion did something truly amazing. People had used CVS - which is itself based around RCS - for so long that it'd somehow become a fact of life that version control systems were too difficult to write.
Even BitKeeper was itself based on RCS. And of course BitMover forbid anyone using BitKeeper - paid or not - if they worked on any other version control system.
Once Subversion proved that version control systems were, in fact, things which ordinary mortals could write, then suddenly everyone was writing them.
The distinction between a different branches and repositories is what makes git so genius. There is none. Just because you call your local branch master and I call my local branch master doesn't mean we are both on the same branch. Rebases and merges work the same when performed against local branches and remote branches. The use of a cryptographic content addressed object store is what makes this work. Syncing is as simple as exchanging read only objects.
So, even though teams don't necessarily use many remotes, they do use loads of branches, which is why they need Git. There are at least n+1 branches where n is the number of people plus whatever feature, production, developmnent, etc. branches people create. On my team that means well over a thousand in the past six months. Some branches are long lived, some are short lived.
Github is just a convenient way to synchronize change. Most teams use it incorrectly by allowing multiple individuals to write to the same repository instead of cloning and creating pull requests.
There is still plenty of room for improvement. Git is not very usable for non experts. A lot of the complexity has to do with dealing with the many ways merges and rebases can fail to work as expected. One limitation here is that a merge commit typically has only two ancestors. Git actually allows for more than two but the porcelain just does not use this. There have been some alternatives where this is used to track changes from multiple branches.
Another area where Git is struggling is larger code bases. The linux kernel is not small but there are examples of big companies where git just isn't good enough for tracking everything they have in one repository. E.g. Google does this (not using git for that). So scale is an issue. Then even though git is distributed, it is not sharded. This imposes upper limits on what can be in a single git repository. The model has always been that you need the entire repository locally to be able to work with it. Wouldn't it be nice if you'd only need a subset of that to be able to work.IPFS copies some of the design for git and stores content addressed objects in a distributed FS. Wouldn't it be nice if somebody rebuilt git on top of that?
So, there's plenty of room for improvement in the Git world.
Unless I'm misunderstanding, Microsoft actually built an implementation to allow working with a subset of files in a repo. They called it Git Virtual File System in case you want to check it out!
Here is an article about changes MS had to do to Git, interesting read
arstechnica.com/information-techno...
Well i know it's not Sharding. but if you need a Subset of a git repo you can do a shallow clone.
I find that useful for Repo's that are very large.
This is a good response, though I think it is missing one fairly important point in the why git was created portion. Bitkeeper, the original VCS for linux, originally had extremely restrictive licensing, but was free (money) for open source projects. Then in 2005, changed the licensing to start charging open source projects too. Most developers in the project would be able to purchase the software, but due to a separate project being worked on by the OSDL (Linux foundation now) that infringed on the restrictive licensing made it so that anyone working for the OSDL (Linus Torvalds) couldn't use the software.
Linus, in his typical Linus-y way, flipped Bitmover a bird and built his own, git, "the stupid content tracker". Bitkeeper was distributed and had most of the features necessary, but was too restrictive.
I used to just keep a copy of the entire Subversion checkout in another directory, then diff -qr them and rsync them as needed if I wanted to work on a local branch. I'm pretty sure that was common practice and it's basically just a manual way of doing what git does.
Git made it easier by reducing the friction and fragility of local history management, and Github made it "win" because the interface was so much prettier than WebSvn and because it was entirely hosted and mostly gratis.
I don't know anyone working in a company who has a second "source of truth" beyond GitHub, Bitbucket, or their local machine these days.
Sorry sir, what do 'authoritative' nodes mean?
I think there's some definite value to something like this if you get the right buy-in and do it the right way. I think that's a pretty herculean task, but I wouldn't be surprised if something like that could get some legs if executed on right.
Could there be another solution that optimizes in the other direction: Like, not caring as much about scalability, but focusing on productivity and safety at the early stages? Maybe with a built-in path to git when the time comes?
Here's an article I read awhile ago: hackernoon.com/where-do-we-go-afte...
I would be interested to see some Dropbox + Git combination.
I think Git is simple enough to create tools over it (like Github) in order to extend it.
Example: Pull Requests aren't a feature of Git, but nobody would use a version control platform (Github, Gitlab, etc.) without that feature which was built upon Git branches.
So instead of doing Git again from the ground I think is better to just improve tools built upon Git, maybe simplify the flow to avoid messing it up.
That's reasonable. Git could be a bit like the assembly language for version control. Regardless of abstractions you build, it compiles to git for compatibility and leverage the immense work put into it.
Git Flow, anyone?
Oh nice, I remember this but had forgotten about where I read it.
Fossil looks interesting. It was created by Richard Hipp who also runs the sqlite project.
Here's a talk about the problems he sees with GIT:
I love Fossil. It has its quirks, as do the alternatives (Git, Mercurial, et cetera), but it provides what I need much more effectively and with far less hassle than those alternatives. In fact, I've been doing more and more of my work in Fossil, and I host my open source Fossil repositories on the web. I'm not sure I'll ever create another repository in Git that isn't a clone of a Fossil repository again, except when I'm somehow required by someone else (e.g. an employer) to do so.
Speaking of clones, I wrote a fairly simple tool in Ruby, called FossGit (
gem install fossgit
), to mirror my work in Fossil repositories on GitHub. The GitHub network effect is too valuable to ignore, at this point.That's awesome, Chad. Thanks for sharing.
Git is, in my opinion, one of the most brilliant pieces of software ever written (I'd go as far as saying it's Linus' biggest achievement, rather than Linux). It's an example of software that's so well designed that it keeps surprising you (well me, at least) by its versatility and performance.
By now there aren't any serious competitors left and maybe that's what's triggering you to ask "aren't there alternatives". In other areas (backend frameworks, frontend frameworks, operating systems, etc etc etc) there are dozens of alternatives with none of them clearly superior, but Git is indeed dominant, but I'm saying deservedly so.
For me it's a breath of fresh air to have at least one area where I'm not bogged down by too many choices with no clear added value.
Add to that the huge installed based (just Github alone but also in enterprises) and the huge boost which Git (and Github) are for OSS and "social coding" (sharing, contributing) - then the conclusion for me is that any alternative has to be clearly superior, not like 50% or 100% "better" but an order of magnitude better, otherwise it won't stand a chance.
That's a very good take.
would just like to mention Mercurial (HG), it's still alive and well and is often much easier to use. Pair it with BitBucket (again has some nice things github doesn't) with it's unlimited repos and a bunch of nice clients (if cli isn't your thing) makes Mercurial a great git alternative.
Both came around after Subversion, but git won the popularity contest for variety of reasons.
HG all the way for me!!! I used HG for my team for about 4+ years and only in the last year moved to GIT as my team merged with a larger group and GIT was the main used by all teams merged. I still love HG's simplicity and paired with BitBucket was awesome.
Personally, I prefer either Mercurial (hg) or Bazaar (bzr). However, since GitHub is the place to do open-source development, if you're going to play in that arena, you've got to be at least familiar.
(There's probably an interesting thought there, something about superior tooling around acceptable-quality software being a killer combo, but my employer would probably rather me philosophize when I'm not on their dime.)
It's the GitHub pull request. That's the killer feature of Git, even though it's not a feature of Git itself.
I really hope not. It has great features but a terrible interface. The command-line is a chaotic mess of inconsistent names and patterns. There are numerous ways of doing certain things, yet there is a lack of orthogonality, meaning things like syntax referring to a remote differs depending on the command.
Using git feels like navigating planks laid across a swamp. One wrong move and you've screwed something up. This is not exactly inspiring for version control.
I prefer the consistency (and UI tools for that matter) of bzr. It's much harder to screw things up, and I don't find myself constantly looking for references online. Granted, I tend not to do as much in bzr than git, but still, it just feels easier.