Also, I worked in consulting for a few years and while Github was common, so was Azure devops (ironically also Microsoft, though) and one project used BitKeeper. Github might be dominant, but is very far from a "monopoly" in my opinion (I don't know if anyone used that word but if that's not the concern then what's the big deal?)
Worked at another company circa 2015 with around 80k employees (not sure how many engineers) that used git + cgit + Jenkins + JIRA.
Bitbucket has a lot of Jira-integration and a lot of large corps are in Jira for the long haul now and "might as well get the synergy of Bitbucket".
Plus, don't underestimate the sunk cost fallacy factor keeping a lot of large corporations locked into things like Perforce and ClearCase.
I made a filter list for uBlock Origin (and compatible) ad-blockers that can hide many of the social media aspects of Microsoft GitHub to make it a bit more tolerable when forced to interact with it due to project lock-in.
This is getting off-topic, but at a certain point Google's results started to feel less and less relevant. Strict keyword matching stopped becoming a feature since it started to "guess" the "hidden meaning" behind your searches, modifiers like "include this exact text" and -"exclude this" stopped being supported and the amount of crap it throws in the search results that are not links to websites became distracting and annoying much of the time rather than useful.
I was happy when Atlassian finally sunset Mercurial, forcing those repositories to move to git. I'd rather all major source code hubs standardize on git since it's harder to contribute needing to learn different incompatible workflows.
I guess what I'm getting at is that Mercurial supporters tend to view it as distinctly better than Git, so sticking to Git just because of its widespread usage does feel like a sort of defeat... One could say similar about Fossil, a much more ambitious project that has faltered probably mostly due to Git claiming all the mind share.
If I could spin up my own cell network like I could an email server and a domain, you bet that the big cell/telecom players would "blacklist" my domain for "spam" the moment they caught whiff and instantly make it almost useless, same as Gmail does.
Git does not have that problem. If four of my buddies want to get together and start a startup together, and I start out by simply setting up git on an SSH server, the four of us do not care that GitHub recently extended the Git protocol with a hot new feature that only works on Windows 13. We just use our git. I am not in the position of wanting to share arbitrary source code with everyone in the world. The relevant social circles are much, much smaller.
I don't know that I'd say git is immune to such takeover, but it is about as resistant as it is feasible to be.
The problem with GitHub isn't git. In fact, that is the least problematic aspect of all. The git repo, the branches, the tags, commit history, signatures, that's all the stuff that is not only easy to take with you, the git protocol all but stuffs them all down your throat with a "git clone". You have to ask git not to give you a fully independent copy of the whole repo.
The problem with GitHub is all the other stuff; the issues, the PRs, the conversations in there, the fact you gave out so many URLs pointing to github, etc. The thing that makes me most nervous about GitHub and my Go libraries is that if GitHub does turn evil, I've got "github.com" embedded in a lot of people's source code. I've decided it's not quite worth it to host on my own domains (plus I imagine people look askance at that anyhow) but it's a long-term concern of mine.
[1] There are some tweaks I would make, like using a standard Merkle tree hash function instead of sha1("blob "+size+"\0"+data), and a text-based encoding for commit and tree objects, but those are bike shed paint. I created a Git-like system based on these ideas that uses URNs and RDF/XML to make going 'under the hood' with a text editor a bit more pleasant than the experience with Git is, which is important because Git's tooling is far superior: https://github.com/TOGoS/ContentCouch/
So you're running your mailing list which people have triple opted into, with confirmations and everything. Someone marks your email as spam and now everyone with a google address is bouncing. You contact google to resolve this and get a canned reply saying they've reviewed it, it's correct, and they will take no further action, and won't discuss it further. Good luck.
Jumped through every hoop, got rDNS, SPF, DMARC, and DKIM all configured correctly...only to find out that the major email providers (at least Google and Microsoft, which account for a depressingly large fraction of email accounts) have decided that the VPS provider doesn't meet their standards, and either auto-block or mark as spam anything I send.
So I'm left with the choice of trying to migrate my entire VPS to a different provider—after figuring out which one, if any, Google and Microsoft actually deign to bless as being worthy of sending email to them—or not host my own email.
There is no provision for a fair shake, special categories for new entrants, or anything similar.
It nonetheless demonstrates how Pareto allows popular providers to make decisions that negatively impact those who want to exercise the freedom granted to them. It doesn’t run in isolation, after all.
> I don't get it, what exactly do you think is wrong with the email landscape? Its a standard protocol everyone can use, the only restrictions being when you start using other's services. Maybe I'm just not abreast enough on this.
I would say most readers would interpret that as in response to email, the standard.
> you're entirely dependent on how the other servers run by the recipients view you.
A protocol can be bulletproof. But spammers will spam, blockers will block, trusted platforms will control.
But Git is pretty good, so it will take something vastly better to replace it.
I don't believe any version control system achieved as deep industry penetration as git has.
There were of course places that were late to switch. I remember one of my old work places used P4 due to legacy in the CI/CD pipelines, but many teams kept a "shadow" git repository in Bitbucket with a bridge to P4. After a while, the company finally switched to Github Enterprise
It definitely wasn't everywhere 10 years ago (but github was). Do you mean 15-20?
Regardless, having completed a CS degree during the period SVN was supposedly everywhere and only encountered it once because someone in my capstone team knew how to use it, I'm inclined to think it didn't permeate everything the way git does now.
If I could complete a bachelors in CS without encountering it in any of my courses, I think it's safe to say git is much more ingrained in the field now (pretty much every syllabus I come across online for CS courses from the last 5 years has had some github links)
Perhaps, but I've never seen that done. The version control system that is in fashion changes over time, but for my entire multidecade career there has always been one.
Personally, a guy I dealt with circa 2010 stood up in a meeting on building better site-wide development processes, and asked what the point of version control was.
So, you've been lucky.
Now, if you go back to the mid '80s and earlier, version control was rare.
yes ... how did you know we used visual source safe?
Yes, we had MS support, and they looked into it, and eventually basically shrugged (I'm told. I wasn't the one working with them).
I'd say each predecessor had about the same level of penetration, which was 'nearly all'. CVS->SVN-GIT. Git will be replaced. It's ripe for that - its interface is terrible, storage is now cheap, and many companies have no actual need for decentralized version control and the complexity that comes with it.
Easiest to use of the bunch is still svn. You can teach a developer everything they need to know in 5 minutes and there's 3-5 commands total you'll need to use ever. I never once saw anyone get into some nightmare state they needed help to dig out of like I've seen routinely with git.
I preferred just a normal folder, rsync (or some nt utils equivalent) and backups.
As I recall, the main problem was that VSS didn't have a proper server! Instead the client modified files in the repository directly using Window's file sharing, relying on locks to maintain consistency. This would have been a bad idea even in Window's file shares weren't slow and buggy.
Edit: And Visual Source Safe
As for dealing with merge conflicts, it's called
- keep up to date with the branch your tracking
- merge frequently
- get on a call when you do have merge conflicts
- avoid having multiple people working in the same files simultaneously if possible
The problem with merge conflicts is you have two people changing the same lines of code at the same time. So I would imagine if you wanted to improve the situation you would need an AI powered version control system that understood the intent of the previous developers. But that's not going to fix bad habits.That sounds like a lot of words for "try really hard to avoid having merge conflicts, because they suck to deal with"
I've seen other systems essentially say, "Oops both versions modified this file... Good luck!" It didn't matter that the modifications were to very different parts of the file. I've never seen that in Git. Any merge conflict I've gotten in Git has been a true potential conflict, even if the resolution is straight-forward from my perspective.
A 3 way merge would have "original", "version a", "version b", and "version c", where "original" is the common ancestor of a, b and c. Mercurial disallows merges like this, but git supports them. I have no idea why git allows it.
But yes, older VCSs ignored the "original" concept, which meant diff tools could just pop up "a" and "b". Which is just a way worse idea than including the original as well.
If everyone develops from master branch and makes small incremental changes, no problem even if it's git.
If everyone develops from their own branches and goes off and does their own thing for a month, but those things are all in totally separate areas of the codebase, not as much of a problem either.
If everyone is developing in their own branches for a month, and they're working on overlapping things, and they aren't communicating? Then you get merge problems
But it's not really a merge problem, it's a communication problem. (And it's not git-only, you see merge problems even in perforce when people go dark for a long time and come back with a huge change that's got conflicts on 50% of the lines now)
I was using ClearCase for a few years and they had a really nice visual merge tool, but the complexity of the merge was great because they were so infrequent. I made a mistake and let someone find out I could merge and ended up getting sucked into these quarterly multi-day merge sessions that made me feel like a guild navigator stressing about breaking one of thousands of changes from 50 people over 90 days.
The patch-centric model of Darcs and Pijul is insufficient. You also need to remember where patches came from. If rewriting a patch does not keep its provenance, you're also in for a world of hurt.
But even that is not sufficient. You need to have more context than even the patch-centric model gives you.
Git is fine, but honestly, my favorite of the several I've used over the years is still svn.
Correction: Git is super mediocre, but sufficiently wide spread it will take something vastly better to replace it.
Don’t confuse inertia with goodness.
Are you saying that the CLI is bad, or that Git’s underlying concepts and data model are bad?
I think the plumbing is probably in-between poor and bad for most projects. It’s maybe great for Linux. However almost all projects are effectively centralized so the D in DVCS is a waste. (Offline/local support and fully distributed are different). I think source control should make it impossible for 99% of users to shoot themselves in the foot. And Git is still terrible at large binary assets, even with Git-LFS.
I use Git + GitHub for my personal projects. It’s mediocre but tolerable. Git is insufficient and not used for every professional project I have ever worked on. YMMV.
What VCS would you use if it was all up to you?
Almost 100% of game devs use Perforce. Git is a non-option.
BigTech all use highly custom source control systems. They have some nice properties.
My stance isn’t that we should use something other than Git per se. It’s that Git is a local maxima, and it kinda sucks. And I think the industry would be healthier if people agreed Git sucks.
Features I care about:
* Terabyte scale sparse checkouts * Ten terabyte scale full checkout * Petabyte scale history * Lossless history. * Automatic cloud/remote backups of every commit * Trivial for low-tech people to use (Perforce achieves this) * Cross-branch file locking for non-mergeable assets
Things I don’t care about:
* Distributed. Defacto centralized with forking is all I need. * Rewriting history. Linux needs it, I don’t.
There are of course lots of things I do like about Git. It’d be good if Git and Perforce had some competition. There’s a market opportunity for it imho.
I do want the open-source world to improve and if we merely build on top of git's abstraction, how can we have any progress then?
Overall, at least when it comes to features, I've actually had the opposite experience. That said, I'm one of those people such that you can pry `git rebase` out of my cold, dead hands, so it's that feature specifically that I really miss when I have to work with Perforce.
When I left, they were investing a ton of time and developer hours into trying to duct-tape git onto their workflow by having smaller groups of developers share cached copies of their huge Perforce repo, and shunting the inevitable merge conflicts onto someone else and hoping for the best.
I spent a considerable amount of time training folks at my acquisition how to work in this environment, and roughly 70% of that time was "yes, Perforce _is_ weird and confusing and annoying, and yes, it _would_ be better if we had lightweight branches, but instead we're stuck in the version-control stone age of Perforce, so here's the complex process you can try to go through to un-break your own local codebase temporarily until someone else has fixed the broken crap they checked in."
What's the standard programming language? The standard editor? The standard database?
> launching a competing standard at this point isn't going to make things better, the best case scenario is that it fragments the industry.
People keep coming out with new programming languages, new editors, and new databases.
Likewise the standard DB may be Postgres.
Others exist but they’re certainly not the recommended defaults you find people giving out nowadays.
VSCode was not the standard seven years ago. ("In the 2016 Developers Survey of Stack Overflow, Visual Studio Code ranked No. 13 among the top popular development tools, with only 7% of the 47,000 respondents using it" says Wikipedia).
If lolinder's thesis is true, then that means it can take over 50 years before the industry decides on a standard. Which seems rather a long time for that statement to be meaningful. Because within 10 years maybe it will all change again.
FWIW, I've mostly used SQLite as my DB.
https://survey.stackoverflow.co/2022/#most-popular-technolog...
VSCode is overwhelming at a staggering 74%. ;P
lolinder wrote "At some point every industry grows up and standardizes on something, anything, because any standard is better than no standard."
Either there was a standard editor before VSCode or there was not.
If there was a standard editor before VSCode, then it's clear that standards can change, which challenges lolinder's proposal that "launching a competing standard at this point isn't going to make things better". The VSCode developers likely thought it would make things better.
If it was no standard editor, then it took about 50 years for the industry to decide on one (dating from the late 1960s to 2016, when VSCode was definitely not the standard).
Leading to my observation that the "at some point" can take decades, and implies the urge to have a standard is not all that strong.
My memories of the 1990s was the CVS was the de facto standard, like how VSCode is now for text editors. Yet people fragmented the industry with SVN, BZR, Git, BitKeeper, Git, and more - because because they thought they could make things better. And at least one of them was right.
I'm therefore not so confident about rejecting launching an alternative to git. My limited understanding of Fossil, for example, tells me there are better approaches than git. (Then again, I'm a Mercurial user, so I already have my own preferences. ;)
Humans are such a weird species. The juxtaposition of the two arguments ("not ideal" and "the standard") is crazy to me. If it's not ideal, let's move away from it, not standardized on it. Is that too much to expect? :)
ok but that's not the argument of the article, so you aren't disagreeing with TFA. Maybe you're disagreeing with another top level comment.
TFA is arguing that the dominance of git hub is or is becoming a problem.
> git itself, while not owned by GitHub, is even more popular. This is largely because of GitHub’s popularity and in spite of git’s poor UX. It would be nice to move beyond git one day and have a better experience for managing complex codebases, and not on GitHub’s timeline.
> ...
> I want to underscore this because I think it’s important: git and GitHub’s symbiosis are holding us back as an industry.
> ...
> Other version control software exists that doesn’t need a separate company to make the code accessible from the web or makes dealing with conflicts easier. We as an industry cannot benefit from these alternatives if we continue to use GitHub this heavily.
like you, and i guess contrary to the article (again, like you), i think git itself as lingua franca is fine and maybe even great. certainly it has warts, some of them substantial, but what doesn't. we've come a long way since CVS and (shudder) SCCS.
Git is far from an ideal solution, but it has emerged as the standard way to manage version control in a software project, and we benefit from there being a standard. Version control is the first barrier to entry in starting on a new project, so having an industry standard for version control benefits everyone: open source projects get more contributors, companies don't have to train new hires on their system, and employees have one fewer barrier blocking them from specific jobs.
Could we have standardized on a better system? Certainly. But launching a competing standard at this point isn't going to make things better, the best case scenario is that it fragments the industry.