Linus Torvalds built Git in 10 days - and never imagined it would last 20 years
Linus Torvalds built Git in 10 days - and never imagined it would last 20 years
It could've been mercurial, but I'm glad that didn't happen. Being shouted at in a mailing-list for fixing a bug doesn't sound like fun. Also, the amount of CPU resources that would be wasted running a VCS in python would be phenomenal. And have fun trying to develop a project using a separate python version than supported by your python VCS.
If Mercurial were as popular as Git I would presume that it would be rewritten in C or Rust, but who can say.
I'm sure it could, but no one thinks it's worth their time when Git is right there and does the same job.
Care to explain your comment for a layman?
From my limited experience mercurial is way more intuitive than git. The big one is named branches are a thing instead of an abstraction.
What do you mean by "are a thing?" Git has branches.
Git branches are very different to Mercurial branches. In git they're similar to tags that move along with the head commit of that particular branch. In Mercurial every commit contains meta data indicating the branch it's on. It also has a query language that lets you do sone quite neat things with selecting groups of commits based on their metadata, which can be useful in code reviews and similar.
That just sounds like an implementation detail.
Can you provide an example of something that's possible in Mercurial, but not git?
I've never used Mercurial, but a simple one based on the explanations and my experience with Git:
Locating the branch a commit originated from. If a git branch has been merged into (or rebased on) main or another branch, there's no way to tell which commit came from which branch. But sometimes I'd really like that information to figure out what prompted a certain change. Without it, I need to use external tools like a ticketing system and hope the other developers added in the necessary information.
That seems to be the opposite of useful if a commit is initially pushed to a development branch, which is relatively standard practice; now you're polluting the tree with data that's purposefully ephemeral, and even potentially leaking internal information.
Also, I'd argue that such deep details do belong in another tool, rather than asking the source control tool perform triple duty by being a CR and issue tracker as well.
Most of the time you're right, it's little more than a detail, but sometimes I miss the querying that it allowed. You could ask for things like a list of all branches that branched from a specific parent branch and modified a specific file, Which can be handy when you want to understand the impact a change might have before you make it and try merging.
Having the branch name embedded in the commit means you can meaningfully ask this sort of question. In git's model you can't say a changeset is in a specific branch once there are child branches further downstream because the changeset is in all of those branches.
Rather than come up with lots of examples for other queries (I know it wasn't the focus of your question, but I think it's really neat), I found this page which seems like a reasonable description.
No, git has labels on heads of branches. Once the head moves you loose the information. It also makes for a more messy history, which I believe created the whole "rebase everything" philosophy to cope.
What information is "loosed" when another commit is made to the branch?
If I hand you a commit, you cannot tell which ‘branch’ it is on without searching the git history and hoping that you only get one answer. That’s a bummer if, for instance, you’re a github action and only get handed the commit. If it’s on the master branch, I want to do different things than if it’s a dev branch.
A commit all by itself doesn't mean as much without context.
Why would I not want to be able to apply a commit to any arbitrary branch?
Also, GitHub is not git - it's based on git. Any shortcomings it may have aren't necessarily due to a flaw in git.
A commit all by itself doesn’t mean as much without context.
Luckily a commit points to its parent, which means the context is inherently present. What’s your point?
Why would I not want to be able to apply a commit to any arbitrary branch?
Nobody said that.
Any shortcomings it may have aren’t necessarily due to a flaw in git.
True enough.
I think it's less user experience and more that mercurial is a lot more demanding hardware wise to do the same rough job?
Mercurial is written in Python, Git in C.
Given the number of git instances, had it been implemented in Python, more CPU cycles / electricity would have been used.
Blah blah Mercurial is responsible for global warming. (I’m being sarcastic by the way - I love Mercurial).
Python 10s of multiples more CPU cycles than git. It is an interpreted language: every instruction is read by another process, checked, and then run. Hit on the other hand is executed straight by the CPU. It has at least one layer of indirection less than python (the python interpreter may have multiple). That means it can be slower but it definitely uses more energy.
Since git is so popular, if it were instead mercurial, the energy requirements would be much higher for version control. Whether that will be noticeable on a bill is debatable. I haven't run the numbers.
Regarding the different python versions. As mentioned before, there is a python interpreter. That interpreter is versioned and so is the python language. Many things are backwards compatible meaning something written in a higher version of the python language can be interpreted by a lower version of the python interpreter. The reverse is also true, so python interpreter with a higher version can interpret a python file using a lower version of the language.
Notice that I put "can" in bold. That's because newer versions can deprecate certain features or parts of the language. So, if you're writing a project in a different version of python, mercurial may or may not run depending on your version. Resolving that may not be as intuitive as one thinks.
I think those were he points you were referring to when you asked your question?
Surely it could be rewritten in Fortran if performance is really a concern.
Assembly is the true language of the gods, sir.
Apart from the obvious lack of portability, compilers write better assembly than most humans.
I don't think it's hyperbole to say a significant percentage of Git activity happens on GitHub (and other "foundries") – which are themselves a far cry from efficient.
My ultimate takeaway on the topic is that we're stuck with Git's very counterintuitive porcelain, and only satisfactory plumbing, regardless of performance/efficiency; but if Mercurial had won out, we'd still have its better interface (and IMO workflow), and any performance problems could've been addressed by a rewrite in C (or the Rust one that is so very slowly happening).
Imagine if their VCS operations were 10s of times less efficient 😉
Most of the VCS ops in Hg are actually written in C.
GitHub is mostly written in Ruby, so that's not really a performance win.
Like I said, we're stuck with Git's UX, but we were never stuck with Hg's performance.
Being shouted at in a mailing-list for fixing a bug doesn’t sound like fun.
What's that a reference to?
Some instances of one of the mercurial devs being a bully. It was a long time ago and I can't remember the alias of the dev, but I do remember they had merge or commit rights - whatever those are called in mercurial. It felt like they had their own Linus but different. Whether they are still active or whether they still are like that, I dunno. Certainly hope they changed.
Dunno if the git mailing list was as toxic. Didn't investigate that.
