Almost 20 years ago I helped our company choose between Git and Mercurial as the replacement for Subversion. Unfortunately, I helped them make the wrong choice, Mercurial.
I say wrong because clearly Git won the war and I haven't used Mercurial since then. However, I still think I made the right choice from a technical perspective; I thought Mercurial was way more user-friendly while providing all the features and performance needed. But I guess I couldn't read the future in terms of which one would win out!
Mercurial is one of the many sad stories of far better technology being forgotten by the popularity contest juggernaut of something else.
I still use mercurial for all my personal project where I don't need to care what anyone else thinks. It is pleasant to use good tools, just like I like to buy top quality rachets or such.
Can't comment on Mercurial, but "for all my personal project where I don't need to care what anyone else thinks" I am using Fossil. Ever since that decision, I've felt a bit, well, held back, or rather, I don't feel quite as comfortable as I do at home when I have to use Git.
Mercurial wasn't the better technology, though. The UX is almost the same as git, diverging in ways that are arguably worse, but the tools were written in much slower Python (initially, and for many years after).
How do you consider the UX "nearly identical" or "arguably worse"?
The Mercurial CLI has clear, well named commands that are predictable and easy to memorize. hg histedit is clean and easy to use and visually shows you what is going to happen - what the new order will be - nondestructively.
The Git CLI requires you to understand its internal data structures to understand the difference between a rebase and a merge, and most people still can't explain it.
I've worked with Mercurial for 5+ years and no one on my team has ever given up on a client and done rm -rf to start anew. Every single git user I've talked to has done that multiple times.
> How do you consider the UX "nearly identical" or "arguably worse"?
The core concept is similar -- history is a stream of content-addressed commits. Concepts map almost 1:1. git does some things arguably better.
> hg histedit is clean and easy to use and visually shows you what is going to happen - what the new order will be - nondestructively.
hg histedit is basically identical to git rebase -i. The names are different, but the operations end up being more or less the same. hg amend -> git commit --amend. Graft -> cherry-pick.
> I've worked with Mercurial for 5+ years and no one on my team has ever given up on a client and done rm -rf to start anew. Every single git user I've talked to has done that multiple times.
I don't know what to tell you. I've also worked with Mercurial for 5+ years, but I've never rm -rf'd a git repo.
> history is a stream of content-addressed commits
Not quite true for mercurial. You also get stable identifiers for commits that remain the same even after being manipulated such as after rebases or amends. It also enables tracking the evolution of a changeset which then enables `hg evolve`.
Being content addressable isn’t a desirable feature in a user-friendly version control system. Who cares about it? Giving stable identifiers to commits is a much more needed feature.
I thought it was enormously better because it helped you not to cut yourself with all the dangerous things in a way that git didn't. It also had an excellent GUI (thg).
It was a much less stressful tool to use and git hasn't really got much better since then - I've just converted a repo to git and the team using it have had about 4 unpleasant mistakes in the last week as they adapt.
As for speed.....I cannot say I ever noticed any problem. Waiting around for the version control system has never been an issue for me.......except a git repo with 70,000 commits and we worked out how to merge a lot of those to fix the problem.
I’ve never met a single person who can use git to move a commit and its descendants from one parent to another. This requires using the extremely unintuitive `git rebase --onto A B C` invocation. The only exception are magit users who are dealing with a much better interface and a better name (magit calls it rebase subset rather than onto).
In contrast every single mercurial user I know can intuitively use `hg rebase` with its `-s` and `-d` flags. That’s one giant difference in UX.
Writing your tooling in python is valid while starting out and prototyping.
One of the big criticisms I've seen levied against Rust is that refactoring is extremely difficult, so prototyping on ideas in the language itself is a poor experience.
I've personally had great success using python, then gradually rewriting the tool I have with py03 to "oxidise" the program iteratively.
Starting with C was great for performance of Git, but damn if it's not a terrible UX these days, I can believe that the choice of toolchain and language was a contributor to that fact.
The Mercurial project has been incrementally rewriting core operations in Rust for several years now. As Pierre-Yves says in the talk, you can do an hg status on a million-file repo in 100ms. I rewrote hg annotate (aka blame) in Rust last year.
It's kind of late, though, right? Git had core components ("plumbing") in C from 2005, with gradual rewriting of the "porcelain" layer from Perl to C in the late 2000s and early 2010s. People have been complaining about Mercurial performance for a long time. I'm sure the Python 2->3 headache did not help.
I've used git and mercurial for roughly the same amount of time.
Your statement is, frankly, something that makes me question your sanity. They're not remotely similar. Outside of something like Perforce, I've not used a VCS with a worse UI.
I'm trying a new strategy where, whenever I'm tempted to Read The Comments, I try to remind myself of all the confidently stated nonsense (e.g., "[Mercurial's] UX is almost the same as git") that I would probably accept if I didn't know it was nonsense, and that I'd almost certainly be better off by just ... not reading the comments.
It doesn't seem to be working all that well, since, y'know, I'm here.
I'm glad that I stuck to git for a similar reason; it won the war. And I understand the need for simpler tools.
But to offer a point I haven't heard from anyone before: at least I feel that I am done with it, I learned this tool sufficiently and I can move on with my life. From time to time I add something to my git toolbelt. I feel if Mercurial or anything else have won, I would maybe have to learn another tool in 5 years, whatever else got popular, and another in next 5 years. But now I have everything I need in git, and always have needed. I hold some hope in it that perhaps the learning curve was worth it.
I’m glad there is one clear winner and we’re not in the common position of having 2-3 relevant/semi-relevant choices that you’re frequently asked to switch between depending on which project you’re looking at at the moment.
Git is modern version control, whatever you think of it, and there’s a simplicity to that.
Despite not using anything else, I don't know all the git commands I need to get my job done so I use UIs and my agent. It's just not intuitive enough, some things are just not possible to do right, and I look forward to ditching it.
I have the feeling that Git winning the war hinges heavily on GitHub being the way to do open source projects, and that is changing given the sad state of GitHub.
Another contender is Jujutsu (jj) which allows you to use jj as frontend and use Git as the backend (with the potential to support any backend, e.g. Google's proprietary Piper), with the best ergonomic and the widest availability of hosting solutions.
I’ve recently switched to jj and it is truly amazing. It too about a week for me to “get it”. The tool is amazing but I think there’s way too much emphasis on what it does/allows rather than what benefits it brings to your workflow. If they get that marketing right I could see it growing. If not, I’ll keep using it
> I have the feeling that Git winning the war hinges heavily on GitHub being the way to do open source projects
Nah. At the time BitBucket was the better way to do open source projects, and they were Mercurial-first. But eventually they had to add Git support because there was so much demand.
But I agree: Mercurial was definitely friendlier for people who didn't have time time to go through the technicalities of git. To use git smoothly you pretty much need to learn how it works internally.
- Its cross-platform support was superior to Git's. (Git didn't get good Windows support until some years later.)
- Its ergonomics were superior to Git's, which was an important factor on its own, and especially important when trying to get a whole organization to retrain and retool around a distributed model.
- (It had a third major advantage over Git that I unfortunately cannot recall at the moment.)
So you weren't wrong back then...
...but Git improved over time, tipping the scale closer to a balanced state. It also had unbeatable author recognition, making it the obvious choice for anyone unaware of Mercurial's advantages, and eventually leading it to benefit from the network effect. And GitHub appeared, greatly improving Git's ecosystem with no support for Mercurial.
That's a matter of taste. I used both for serious work, at the time, and found Git much more usable. My experience with Mercurial was "welcome to Mercurial, how can we help you merge and push your work in progress even though that's not what you want?" My experience with Git was one where I felt in control at all times, had a clear workflow for when I did and didn't want to publish my changes (and for when I wanted to edit them first), and allowed me to quickly make and switch branches within a single working copy.
Thank you. I vaguely remember being put off by Mercurial at the time and feeling much more comfortable with git, but couldn't have put it into words as well as you did.
People always say you have to know git internals in order to use it, but that's just not true. Git has the right data model and has always been about empowering users to edit their data. Which makes the data model be "in your face" compared to the alternatives (and I think that's what people latch on to when they talk about "internals"), but it ultimately makes for a better tool.
I don't know what you mean by ergonomics, but I remember trying both Mercurial and Git back in the days after using Subversion before. I didn't like how Mercurial didn't easily let me rewrite history and do stuff like `git commit --ammend` or `git rebase`. Mercurial users kept telling me using an extension to manage patches on top of Mercurial (I think it was quilt).
I agree about the Windows support. hg serve was also nice. Plus TortoiseHg.
I also did this. Both in hindsight and at the time, I thought Mercurial had far better tooling. But it was not all amazing: Mercurial’s branching model was very poor, and its sequentially numbered revision system was and remains a very bad design.
I think I liked the Mercurial branching model better than git, due to the branches being a first class record of events. What I did not know is how common the git rebase/clean linear history would become or a desire to change history on merge.
Mercurial had bookmarks that were roughly the same as git branches.
The linear version numbers were quite useful to reason about and use in places that call for a "number" version number, and were useful relative to your "master" clone. That was not the primary way though, it had hashes like git too, that were the same from clone to clone.
The clean linear history thing is something I never really got, despite using git for 12 years now. I worked with some smart developers whose rule was "rebase if you want, but if too complicated, just merge", and it didn't hurt the delivery or maintainability of the code they wrote.
Yes - whenever I'm in a team and I hear someone who insists on a linear history, I always wonder why they have trouble with merge when lots of folks like me have no problem with it.
Finally, in one team, I more or less forced a senior engineer use merge (or rather, I was in control of the project and did not force other developers to use rebase). After a year, he admitted that he no longer really saw a benefit in rebase and switched to just using merges in his own projects. He also noticed fewer merge conflicts this way.
Rebase makes sense when you realize git doesn't have branches. Git has tags that move but no branches. That means when you merge you have no clue which branch was the mainline and which was the fork. This is a question I often ask 10 years after switching to git. Sadly git has better tooling so it is worth using despite the issues.
For a complicated long running feature branch I can see it. Instead of repeatedly merging the root in during development it can be cleaner. Tools aren’t always good at figuring out in a PR what was written and what was caused by those merges from root. And history looks better at the end.
For a short branch that can merge cleanly or perhaps very close to it, I’d kind of rather have the ‘true’ history. I don’t think it’s worth it.
I’ve never understood the “everything must be rebased before every merge” desire.
The branching model being keeping a record is fine, but needing to make up a name for your branch before committing (unless you like rewriting) was not awesome, and the names being a global namespace was unpleasant.
And then, when you pull someone else’s in-progress work to inspect it, you end up with their branches showing up with their names and you ended up with revisions 13564-13592 belonging to someone else and showing up in your history graph even when you continue on your own work at revisions 13563 and 13593. I ended up using temporary clones and strip a lot.
git branches, in contrast, are delightfully unobtrusive.
Sequentially numbered versions is still used at main Google monorepo (at least did a few years ago), named "changelist number", from perforce. Up to the point that people define extension field numbers in protobufs using their changelist number, to ensure it will never intersect with anyone else.
I always hear it has far better “tooling” but then the comments say that branching sucks, revisions suck and there is no good got stash equivalent - this is like a third of what I use daily with git.
What does “far better tooling” mean exactly, could you give an example of what amazing tools I’m missing out on (never have used anything else but git, when I came to the industry it was already the standard)
Branching in Mercurial is as good as or better than in git, but it takes a few minutes of additional reading until the full flexibility of Mercurial's model actually reveals itself. Revisions can be safely ignored in favor of commit hashes. I haven't used revision numbers in hg for at least a decade now. And hg has a stash implementation that is on par with every other VCS that I've used so far.
TortoiseHG is a very good client that covers all common Mercurial operations and then some. It's on par with a couple of commercial git clients that I've used. On the server side, there's e.g. heptapod as a GitLab fork that has a deep Mercurial implementation.
When I used it a decade ago, virtually everything in mercurial was slightly better-designed, more user-friendly, and more polished. Much shorter learning curve.
It means that git invented a bunch of new jargon and ideas that confuses people exposed to it for the first time, where hg's usage metaphor hews closer to the received wisdom of people coming from stuff like subversion and perforce.
It's true that git's ad hoc command line UI isn't exactly it's greatest strength. But given the complexity of the design space here that's a pretty weak argument IMHO. The two weeks it takes to get the basic git workflow into your muscle memory pale in comparison to the years it'll take you to be good at bisection and tree maintainership.
It's also sort of a wrong argument in the modern world. People new to git have extensive assistive technologies available. There is, after all, no HgHub out there.
> I also did this. Both in hindsight and at the time, I thought Mercurial had far better tooling.
I recall checking Mercurial back in the day and being puzzled by the lack of basic features such as the ability to stash changes. I also recalled that the community was dismissive of the lack of such a basic feature, with comments such as users could always create local branches, of even we could perhaps install a module such as shelve.
That was the image that Mercurial left with me with regards to git: missing critical features and not bothering to bridge the gap.
> I recall checking Mercurial back in the day and being puzzled by the lack of basic features such as the ability to stash changes. I also recalled that the community was dismissive of the lack of such a basic feature, with comments such as users could always create local branches,
I started with Mercurial, eventually got forced into git, and now use jujutsu.
Totally agree with the Mercurial developers: Just use a branch/bookmark. When I encountered it in git, it seemed neat, but became yet another concept/thing to clean up that you don't need to.
And lo and behold, after switching to jujutsu, everyone shows how you can do a stash using an (anonymous) branch.
Even though I used stash a lot in my git days, I don't miss it at all while using jujutsu. The benefit of jj is the ease with which one makes branches (without needing to name them). That's why you may not have liked the advice in mercurial - it wasn't the solution that was problematic, but that mercurial didn't make it as easy as it should have been.
(Same goes for index - no one misses it once they switch to jujutsu).
It did have a kind of equivalent to stashes in the mq extension, but its interface was a bit esoteric compared to the rest of Hg, from what I remember.
A lot of features that git had by default had to be enabled as plugins in mercurial.
The plugins were usually shipped with mercurial so you didn't have to install them separately, but you needed to know that you had to enable them in a config. And I beleive this turned a lot of people off.
I think some of the extensions were very basic stuff like graph logging and colorized output -- and mq like you said. So it was kind of unfortunate that people got a bad impression of hg from that and bounced off.
git rebase, for all its warts, was always better than mq. A failed mq-driven rewrite was destructive! (And it kind of had to be — if you were trying to edit revision 17, there was no number available for the original revision 17 because the schema didn’t allow two revisions numbered 17, so the original had be excised.)
I'd made the exact same choice around the same time at the company where I was working. Last I heard, they're still using it. My rationale was that Mercurial was a lot safer and user-friendly compared to git. Needless to say, git has improved by leaps and bounds since then.
I don't think Git winning a popularity contest is a reason for choosing Mercurial to have been wrong, or unfortunate. Was there some negative consequence from the decision -- either directly from Mercurial itself, or just because over time everyone expected Git, perhaps?
(Hopefully this comes across as curious, which it is, and not antagonistic, which its not)
Not GP but there is a consequence. These days if someone has used version control they almost certainly know Git.
That means pretty much everyone who comes in the door needs Mercurial training, whether formal or informal. You’d get the same effect from still using CVS, SVN, or other things.
That may not be a big issue. If someone understands version control I’d hope they could adapt to another minder pretty fast.
I used Mercurial back in the day too. I agree, it was better. That being said, GitHub was better than other similar services which no doubt helped git win and now, git is ubiquitous.
I hate that social factors like popularity are a thing in technical decisions.
I have not used mercurial (though heard good things about it) but I saw a similar thing play out in Rust gamedev. There are 2 competing game engines, one better technically, the other much more popular. Now, if everyone made a purely technical decision, they'd pick the first one and eventually it would become the more popular one. Unfortunately, whenever I asked people why they chose the second one, they said because it was more popular. Tragedy of the commons.
If it's any consolation, maybe jj will take over. I haven't tried it yet (I use the staging area a lot in my workflow), but AFAIK they made the choice to be git-compatible which means it's not a choice between one or the other but lets people and teams migrate gradually.
FWIW you can still have a staging-area-like workflow with JJ - it's just that while git has "commits", "the staging area", "the working directory", and "stashes" as four separate concepts with four separate toolkits, in JJ all of those things are "commits" and a single toolkit works with all of them :)
Mercurial was safer and better. I still use it and it's still safer.
The bookmarks feature which is supposed to be the solution for short-lived branches is hard to understand though. I'm probably dumb but I can't work it out and hence the overall tool is that much less useful.
It needs a github-like website and Heptapod would be great if I could use it - I've set up a project, been unable to do anything and then had it all closed down. OTOH self-hosting is a lot more feasible nowadays with today's fibre connections.
I hated so much how Mercurial dealt with short-lived branches, that after seeing how Git did it, I've never looked back. I also remember how some people told me to use the quilt or something extension to manage patches, but it was too complicated for me.
Jujutsu took the best parts about mercurial and enabled these to be used with git and git forges. Once I had gotten used to jujutsu I didn’t really miss mercurial.
I haven't used hg in years now, bitbucket discontinuing support is what killed it for me and I had no interest in self hosting.
My git usage is very basic, my gitconfig has been pretty much untouched for years but on those occasions where I get stuck or hit something I end up searching and usually get through a bunch of posts/comments/sites and wish I was using hg.
At my previous big tech employer we used to have a mercurial layer on top of our legacy version control system. I loved it, much simpler and more clear than the typical workflows I have to deal with in git. I get it that with git you have more power, but do really most teams need that?
Mercurial wasn't as simple as Subversion. But with hg I still felt like understanding 80% of what the tool had to offer and actually being able to mold the timeline the way I wanted.
Git has so many gotchas, bells and whistles that whenever I'm doing something out of the ordinary I'm wondering if there isn't an easier / canonical / smarter way I should be doing it.
There was a quote somewhere about Mercurial having a mental model small enough you can fit in your head - and that was the big win for me.
It was also fast and had very clean, easy to contribute to code. I remember submitting a patch and getting a bit of Python education from Matt, which was very useful.
Git is fine but it's inconsistent enough in the interface department, even after all this time, that I still get regularly frustrated. On the other hand, you can't just break a workflow that already exists and I very much appreciate it scales to work far beyond mine.
I do like that the git people are doing the difficult work of improving the UI over time - it's hard to change the engines while the plane is flying!
git's mental model is very, very small, if you care to learn it. Then all the commands and their "inconsistencies" start to make sense - they operate on the model almost without any magic, and not on whatever is user's intent (it can vary a lot)
What? Subversion is by far the most complex versioning software I've ever used.
> Git has so many gotchas, bells and whistles
The Git UI leaves a little to be desired. But inside, Git is basically just blobs, trees, commits, and refs. It'd be hard (or impossible?) to find a conceptually simpler versioning system.
I found that doing anything even slightly unusual in Git was pretty incomprehensible until I learned its internals, and pretty easy once I did. Fortunately as you say the internals are conceptually pretty simple to learn.
I loved Mercurial - still do I guess as I just installed it on the Linux Mint VM I keep around for messing with Linux. The thing I really liked about it was TortoiseHg, which provides integration between Mercurial and Windows Explorer. There is a similar TortoiseGit but, at least back when I was doing serious development, it had quite a few problems.
Glad it is still around. hg was the better experience as it had fewer command surface. But slower, and sometimes you need the extra detail commands to analyze the history.
Anyone know what Matt Mackall is up to these days? He started Mercurial and got people involved early on with a lot of enthusiasm, you could tell he cared about what he created and the people who joined him ("hg crew"). I learned a lot from him on how to think like an engineer and saw him manage different personalities in the project in a kind and sincere way (I think this was around ~2010).
FYI: Matt Mackall is now Olivia Mackall [1], so that can make searching for things harder. Looks like they work at Valve, now? Agreed that they were a really stabilizing and healthy personality in the project, and made a lot of good early decisions, which Mercurial has continually benefited from. Eg: the commitment to backwards-compatibility [2].
Olivia uses she/her, and I believe she is now retired :)
I owe her and the Mercurial community a great debt. The community taught me how to think like an engineer building infrastructure and the importance of backwards compatibility, something I've tried to carry forward in tools like cargo-nextest.
+1 I learned so much about software engineering (Software design, network protocol incl how to handle backward compat in sane ways, ...) thanks to the mercurial community.
My first full time job after university was using hg, and particularly https://tortoisehg.bitbucket.io/ made it really pleasant.
Prior and post that I'd always used git but I'll always have a bit of a soft spot for mercurial, especially as our forge usage at the time predated strict guardrails and controls - we did code review, but it was your responsibility to tag the appropriate people and wait for them to respond, if you felt it was necessary to merge prior to that you could - but better be ready to defend that decision.
Git likely won because of GitHub; I abandoned Mercurial quite early when it was clear that it had lost. It was good, but I cannot imagine using it today.
I just installed it on a Raspberry Pi (with an otherwise too-small-for-any-other purpose SSD) for use at home. I wanted something with low power consumption, and I didn't want to have a single point of failure by running it in a container alongside everything else.
The only hiccup was forgetting that when pushing via the SSH connection, it will have paths relative to the home directory of my hg user.
IIRC Facebook switched to HG from SVN in the 2010s, one (main?) motivator being that the single repo was getting too big and svn’s only way was to start splitting it up. Which was against the philosophy of openness of the single repo. No idea what’s Meta doing now.
Last I checked (2 years ago) Meta was using Sapling, a very heavily customised open source Mercurial frontend with proprietary backend.
FWIW the Sapling frontend can also be connected to a Git backend, and I've been using that for all my open source projects to get the best of Mercurial's user experience niceness while collaborating via GitHub <3
I liked too. It was `hg serve` [1]. Instant web interface and an easy way to share the repository with other people (assuming your computer was accessible from the LAN or Internet).
Just a bit off topic. But is anyone still using git or whatever versioning in the traditional way? I hardly see myself using the git command these days. Everything I handled by Claude. From pulling to pushing to solve merge conflicts and more. In a way it doesn't matter anymore whether it's git or something else.
Mercurial was the only Python application that worked, so van Rossum and crew did not like it and had to move to git and then GitHub (Ruby).
Python developers, especially core are used to perpetually broken software and don't like stuff that just works. As long as the software is in Python - C (git) and Ruby (websites that used to work at that time) are fine.
Honestly hg shot themselves in the foot by not releasing any stable API and making developers use their CLI interface. Between that and the performance and dynamic typing issues of Python, it was almost sure to lose the race to Git.
Git's filesystem layout has always been stable and Git encouraged developers to target it. Mercurial strongly cautioned developers away from its filesystem or source code APIs and told them to subprocess out to the CLI. To be clear, I dislike Git, but it was _designed_ with developers in mind (which isn't to say that it has a nice API, only that the Git people intended for Git to be extended). I prefer Mercurial's concepts and interface, but Mercurial was somewhere between indifferent and hostile toward developers.
Almost 20 years ago I helped our company choose between Git and Mercurial as the replacement for Subversion. Unfortunately, I helped them make the wrong choice, Mercurial.
I say wrong because clearly Git won the war and I haven't used Mercurial since then. However, I still think I made the right choice from a technical perspective; I thought Mercurial was way more user-friendly while providing all the features and performance needed. But I guess I couldn't read the future in terms of which one would win out!
Mercurial is one of the many sad stories of far better technology being forgotten by the popularity contest juggernaut of something else.
I still use mercurial for all my personal project where I don't need to care what anyone else thinks. It is pleasant to use good tools, just like I like to buy top quality rachets or such.
Can't comment on Mercurial, but "for all my personal project where I don't need to care what anyone else thinks" I am using Fossil. Ever since that decision, I've felt a bit, well, held back, or rather, I don't feel quite as comfortable as I do at home when I have to use Git.
I've always been interested in Fossil, especially how they handle all the things in a project that aren't strictly code but still need to be tracked.
And in a similar vein, Darcs. It unfortunately couldn’t compete with Git on performance, but the user experience is on a whole another level.
Mercurial wasn't the better technology, though. The UX is almost the same as git, diverging in ways that are arguably worse, but the tools were written in much slower Python (initially, and for many years after).
How do you consider the UX "nearly identical" or "arguably worse"?
The Mercurial CLI has clear, well named commands that are predictable and easy to memorize. hg histedit is clean and easy to use and visually shows you what is going to happen - what the new order will be - nondestructively.
The Git CLI requires you to understand its internal data structures to understand the difference between a rebase and a merge, and most people still can't explain it.
I've worked with Mercurial for 5+ years and no one on my team has ever given up on a client and done rm -rf to start anew. Every single git user I've talked to has done that multiple times.
> How do you consider the UX "nearly identical" or "arguably worse"?
The core concept is similar -- history is a stream of content-addressed commits. Concepts map almost 1:1. git does some things arguably better.
> hg histedit is clean and easy to use and visually shows you what is going to happen - what the new order will be - nondestructively.
hg histedit is basically identical to git rebase -i. The names are different, but the operations end up being more or less the same. hg amend -> git commit --amend. Graft -> cherry-pick.
> I've worked with Mercurial for 5+ years and no one on my team has ever given up on a client and done rm -rf to start anew. Every single git user I've talked to has done that multiple times.
I don't know what to tell you. I've also worked with Mercurial for 5+ years, but I've never rm -rf'd a git repo.
> history is a stream of content-addressed commits
Not quite true for mercurial. You also get stable identifiers for commits that remain the same even after being manipulated such as after rebases or amends. It also enables tracking the evolution of a changeset which then enables `hg evolve`.
Being content addressable isn’t a desirable feature in a user-friendly version control system. Who cares about it? Giving stable identifiers to commits is a much more needed feature.
I thought it was enormously better because it helped you not to cut yourself with all the dangerous things in a way that git didn't. It also had an excellent GUI (thg).
It was a much less stressful tool to use and git hasn't really got much better since then - I've just converted a repo to git and the team using it have had about 4 unpleasant mistakes in the last week as they adapt.
As for speed.....I cannot say I ever noticed any problem. Waiting around for the version control system has never been an issue for me.......except a git repo with 70,000 commits and we worked out how to merge a lot of those to fix the problem.
The dangerous parts of git exist to make it trivial to undo mistakes. You don’t have to use those for your regular workflows.
With any other system, your only option is usually checking out a fresh copy from a server or backup.
I’ve never met a single person who can use git to move a commit and its descendants from one parent to another. This requires using the extremely unintuitive `git rebase --onto A B C` invocation. The only exception are magit users who are dealing with a much better interface and a better name (magit calls it rebase subset rather than onto).
In contrast every single mercurial user I know can intuitively use `hg rebase` with its `-s` and `-d` flags. That’s one giant difference in UX.
[delayed]
Writing your tooling in python is valid while starting out and prototyping.
One of the big criticisms I've seen levied against Rust is that refactoring is extremely difficult, so prototyping on ideas in the language itself is a poor experience.
I've personally had great success using python, then gradually rewriting the tool I have with py03 to "oxidise" the program iteratively.
Starting with C was great for performance of Git, but damn if it's not a terrible UX these days, I can believe that the choice of toolchain and language was a contributor to that fact.
> Starting with C was great for performance of Git
Isn't the entire git rebase logic written in Bash scripts? Or was originally?
> Writing your tooling in python is valid while starting out and prototyping.
This fallacy again. Tell me, when did Mercurial decide "ok the prototype is done, we'll rewrite it in a proper language"?
They didn't, of course. Because nobody ever does. Your "prototype" gradually becomes a 100k line product that you can't afford to rewrite.
(I guess you can YOLO it with AI these days but that wasn't an option for Mercurial.)
> Starting with C was great for performance of Git, but damn if it's not a terrible UX these days
Git's terrible UX doesn't have anything to do with C. C doesn't make you always pick a different flag for "delete".
The Mercurial project has been incrementally rewriting core operations in Rust for several years now. As Pierre-Yves says in the talk, you can do an hg status on a million-file repo in 100ms. I rewrote hg annotate (aka blame) in Rust last year.
It's kind of late, though, right? Git had core components ("plumbing") in C from 2005, with gradual rewriting of the "porcelain" layer from Perl to C in the late 2000s and early 2010s. People have been complaining about Mercurial performance for a long time. I'm sure the Python 2->3 headache did not help.
There was no headache. The migration was extremely smooth.
> The UX is almost the same as git,
I've used git and mercurial for roughly the same amount of time.
Your statement is, frankly, something that makes me question your sanity. They're not remotely similar. Outside of something like Perforce, I've not used a VCS with a worse UI.
I'm trying a new strategy where, whenever I'm tempted to Read The Comments, I try to remind myself of all the confidently stated nonsense (e.g., "[Mercurial's] UX is almost the same as git") that I would probably accept if I didn't know it was nonsense, and that I'd almost certainly be better off by just ... not reading the comments.
It doesn't seem to be working all that well, since, y'know, I'm here.
> I've not used a VCS with a worse UI
...than git? than hg?
I'm glad that I stuck to git for a similar reason; it won the war. And I understand the need for simpler tools.
But to offer a point I haven't heard from anyone before: at least I feel that I am done with it, I learned this tool sufficiently and I can move on with my life. From time to time I add something to my git toolbelt. I feel if Mercurial or anything else have won, I would maybe have to learn another tool in 5 years, whatever else got popular, and another in next 5 years. But now I have everything I need in git, and always have needed. I hold some hope in it that perhaps the learning curve was worth it.
That’s a great point I hadn’t thought of before.
I’m glad there is one clear winner and we’re not in the common position of having 2-3 relevant/semi-relevant choices that you’re frequently asked to switch between depending on which project you’re looking at at the moment.
Git is modern version control, whatever you think of it, and there’s a simplicity to that.
Despite not using anything else, I don't know all the git commands I need to get my job done so I use UIs and my agent. It's just not intuitive enough, some things are just not possible to do right, and I look forward to ditching it.
I have the feeling that Git winning the war hinges heavily on GitHub being the way to do open source projects, and that is changing given the sad state of GitHub.
Another contender is Jujutsu (jj) which allows you to use jj as frontend and use Git as the backend (with the potential to support any backend, e.g. Google's proprietary Piper), with the best ergonomic and the widest availability of hosting solutions.
I’ve recently switched to jj and it is truly amazing. It too about a week for me to “get it”. The tool is amazing but I think there’s way too much emphasis on what it does/allows rather than what benefits it brings to your workflow. If they get that marketing right I could see it growing. If not, I’ll keep using it
> I have the feeling that Git winning the war hinges heavily on GitHub being the way to do open source projects
Nah. At the time BitBucket was the better way to do open source projects, and they were Mercurial-first. But eventually they had to add Git support because there was so much demand.
> But I guess I couldn't read the future in terms of which one would win out!
After Linus Torvalds gave this talk at Google in 2007, it was clear he would win. (Is there a better quality video somewhere?)
https://www.youtube.com/watch?v=idLyobOhtO4
But I agree: Mercurial was definitely friendlier for people who didn't have time time to go through the technicalities of git. To use git smoothly you pretty much need to learn how it works internally.
> I helped them make the wrong choice, Mercurial.
20 years ago, Mercurial was not the wrong choice.
- Its internal design was very similar to Git's.
- Its cross-platform support was superior to Git's. (Git didn't get good Windows support until some years later.)
- Its ergonomics were superior to Git's, which was an important factor on its own, and especially important when trying to get a whole organization to retrain and retool around a distributed model.
- (It had a third major advantage over Git that I unfortunately cannot recall at the moment.)
So you weren't wrong back then...
...but Git improved over time, tipping the scale closer to a balanced state. It also had unbeatable author recognition, making it the obvious choice for anyone unaware of Mercurial's advantages, and eventually leading it to benefit from the network effect. And GitHub appeared, greatly improving Git's ecosystem with no support for Mercurial.
> Its ergonomics were superior to Git's
That's a matter of taste. I used both for serious work, at the time, and found Git much more usable. My experience with Mercurial was "welcome to Mercurial, how can we help you merge and push your work in progress even though that's not what you want?" My experience with Git was one where I felt in control at all times, had a clear workflow for when I did and didn't want to publish my changes (and for when I wanted to edit them first), and allowed me to quickly make and switch branches within a single working copy.
Thank you. I vaguely remember being put off by Mercurial at the time and feeling much more comfortable with git, but couldn't have put it into words as well as you did.
People always say you have to know git internals in order to use it, but that's just not true. Git has the right data model and has always been about empowering users to edit their data. Which makes the data model be "in your face" compared to the alternatives (and I think that's what people latch on to when they talk about "internals"), but it ultimately makes for a better tool.
I don't know what you mean by ergonomics, but I remember trying both Mercurial and Git back in the days after using Subversion before. I didn't like how Mercurial didn't easily let me rewrite history and do stuff like `git commit --ammend` or `git rebase`. Mercurial users kept telling me using an extension to manage patches on top of Mercurial (I think it was quilt).
I agree about the Windows support. hg serve was also nice. Plus TortoiseHg.
Mercurial does let you amend commits like git does, histedit is essentially identical to rebase -i, and evolve is better than anything git has.
Around 20 years ago Facebook made the same choice, so you're in good company in terms of technically sophisticated shops.
Facebook has been using their own in-house Sapling/Eden for years and years now. I'm not sure how much similarity remains with open source Mercurial.
Our company also made this choice.
One of the first things I did was switch us to git.
Mercurial was way easier to use and fit our use case but all the tooling was built for git.
I also did this. Both in hindsight and at the time, I thought Mercurial had far better tooling. But it was not all amazing: Mercurial’s branching model was very poor, and its sequentially numbered revision system was and remains a very bad design.
I think I liked the Mercurial branching model better than git, due to the branches being a first class record of events. What I did not know is how common the git rebase/clean linear history would become or a desire to change history on merge.
Mercurial had bookmarks that were roughly the same as git branches.
The linear version numbers were quite useful to reason about and use in places that call for a "number" version number, and were useful relative to your "master" clone. That was not the primary way though, it had hashes like git too, that were the same from clone to clone.
The clean linear history thing is something I never really got, despite using git for 12 years now. I worked with some smart developers whose rule was "rebase if you want, but if too complicated, just merge", and it didn't hurt the delivery or maintainability of the code they wrote.
Yes - whenever I'm in a team and I hear someone who insists on a linear history, I always wonder why they have trouble with merge when lots of folks like me have no problem with it.
Finally, in one team, I more or less forced a senior engineer use merge (or rather, I was in control of the project and did not force other developers to use rebase). After a year, he admitted that he no longer really saw a benefit in rebase and switched to just using merges in his own projects. He also noticed fewer merge conflicts this way.
Rebase makes sense when you realize git doesn't have branches. Git has tags that move but no branches. That means when you merge you have no clue which branch was the mainline and which was the fork. This is a question I often ask 10 years after switching to git. Sadly git has better tooling so it is worth using despite the issues.
Agreed.
For a complicated long running feature branch I can see it. Instead of repeatedly merging the root in during development it can be cleaner. Tools aren’t always good at figuring out in a PR what was written and what was caused by those merges from root. And history looks better at the end.
For a short branch that can merge cleanly or perhaps very close to it, I’d kind of rather have the ‘true’ history. I don’t think it’s worth it.
I’ve never understood the “everything must be rebased before every merge” desire.
exactly the opposite:
"For a complicated long running feature branch" always simpler to repeatedly merge main into dev, easier conflicts solving etc
For simpler cases squash+rebase as default merge strategy trumps leaves a nice clean history.
The branching model being keeping a record is fine, but needing to make up a name for your branch before committing (unless you like rewriting) was not awesome, and the names being a global namespace was unpleasant.
And then, when you pull someone else’s in-progress work to inspect it, you end up with their branches showing up with their names and you ended up with revisions 13564-13592 belonging to someone else and showing up in your history graph even when you continue on your own work at revisions 13563 and 13593. I ended up using temporary clones and strip a lot.
git branches, in contrast, are delightfully unobtrusive.
Sequentially numbered versions is still used at main Google monorepo (at least did a few years ago), named "changelist number", from perforce. Up to the point that people define extension field numbers in protobufs using their changelist number, to ensure it will never intersect with anyone else.
I always hear it has far better “tooling” but then the comments say that branching sucks, revisions suck and there is no good got stash equivalent - this is like a third of what I use daily with git.
What does “far better tooling” mean exactly, could you give an example of what amazing tools I’m missing out on (never have used anything else but git, when I came to the industry it was already the standard)
Branching in Mercurial is as good as or better than in git, but it takes a few minutes of additional reading until the full flexibility of Mercurial's model actually reveals itself. Revisions can be safely ignored in favor of commit hashes. I haven't used revision numbers in hg for at least a decade now. And hg has a stash implementation that is on par with every other VCS that I've used so far.
TortoiseHG is a very good client that covers all common Mercurial operations and then some. It's on par with a couple of commercial git clients that I've used. On the server side, there's e.g. heptapod as a GitLab fork that has a deep Mercurial implementation.
“Far better tooling” means that you don’t need to do git help reset and try to remember each of the nonsensical choices, for example.
When I used it a decade ago, virtually everything in mercurial was slightly better-designed, more user-friendly, and more polished. Much shorter learning curve.
It had a much better GUI in 2009 (THG) and I think today the GUIs for git aren't really better - probably worse.
> What does “far better tooling” mean exactly
It means that git invented a bunch of new jargon and ideas that confuses people exposed to it for the first time, where hg's usage metaphor hews closer to the received wisdom of people coming from stuff like subversion and perforce.
It's true that git's ad hoc command line UI isn't exactly it's greatest strength. But given the complexity of the design space here that's a pretty weak argument IMHO. The two weeks it takes to get the basic git workflow into your muscle memory pale in comparison to the years it'll take you to be good at bisection and tree maintainership.
It's also sort of a wrong argument in the modern world. People new to git have extensive assistive technologies available. There is, after all, no HgHub out there.
> I also did this. Both in hindsight and at the time, I thought Mercurial had far better tooling.
I recall checking Mercurial back in the day and being puzzled by the lack of basic features such as the ability to stash changes. I also recalled that the community was dismissive of the lack of such a basic feature, with comments such as users could always create local branches, of even we could perhaps install a module such as shelve.
That was the image that Mercurial left with me with regards to git: missing critical features and not bothering to bridge the gap.
> I recall checking Mercurial back in the day and being puzzled by the lack of basic features such as the ability to stash changes. I also recalled that the community was dismissive of the lack of such a basic feature, with comments such as users could always create local branches,
I started with Mercurial, eventually got forced into git, and now use jujutsu.
Totally agree with the Mercurial developers: Just use a branch/bookmark. When I encountered it in git, it seemed neat, but became yet another concept/thing to clean up that you don't need to.
And lo and behold, after switching to jujutsu, everyone shows how you can do a stash using an (anonymous) branch.
Even though I used stash a lot in my git days, I don't miss it at all while using jujutsu. The benefit of jj is the ease with which one makes branches (without needing to name them). That's why you may not have liked the advice in mercurial - it wasn't the solution that was problematic, but that mercurial didn't make it as easy as it should have been.
(Same goes for index - no one misses it once they switch to jujutsu).
It did have a kind of equivalent to stashes in the mq extension, but its interface was a bit esoteric compared to the rest of Hg, from what I remember.
A lot of features that git had by default had to be enabled as plugins in mercurial.
The plugins were usually shipped with mercurial so you didn't have to install them separately, but you needed to know that you had to enable them in a config. And I beleive this turned a lot of people off.
I think some of the extensions were very basic stuff like graph logging and colorized output -- and mq like you said. So it was kind of unfortunate that people got a bad impression of hg from that and bounced off.
git rebase, for all its warts, was always better than mq. A failed mq-driven rewrite was destructive! (And it kind of had to be — if you were trying to edit revision 17, there was no number available for the original revision 17 because the schema didn’t allow two revisions numbered 17, so the original had be excised.)
I'd made the exact same choice around the same time at the company where I was working. Last I heard, they're still using it. My rationale was that Mercurial was a lot safer and user-friendly compared to git. Needless to say, git has improved by leaps and bounds since then.
I don't think Git winning a popularity contest is a reason for choosing Mercurial to have been wrong, or unfortunate. Was there some negative consequence from the decision -- either directly from Mercurial itself, or just because over time everyone expected Git, perhaps?
(Hopefully this comes across as curious, which it is, and not antagonistic, which its not)
Not GP but there is a consequence. These days if someone has used version control they almost certainly know Git.
That means pretty much everyone who comes in the door needs Mercurial training, whether formal or informal. You’d get the same effect from still using CVS, SVN, or other things.
That may not be a big issue. If someone understands version control I’d hope they could adapt to another minder pretty fast.
It’s still an issue. There is technically a cost.
There is an initial cost to learning hg, but it is usually offset by the less destructive nature of hg's behaviors and defaults in the long term.
I used Mercurial back in the day too. I agree, it was better. That being said, GitHub was better than other similar services which no doubt helped git win and now, git is ubiquitous.
You did the right at the right time, you wasn't some prescient being. Why are you
meanwhile my org still uses svn for a lot of repos at least my teams are full git...ugh
I hate that social factors like popularity are a thing in technical decisions.
I have not used mercurial (though heard good things about it) but I saw a similar thing play out in Rust gamedev. There are 2 competing game engines, one better technically, the other much more popular. Now, if everyone made a purely technical decision, they'd pick the first one and eventually it would become the more popular one. Unfortunately, whenever I asked people why they chose the second one, they said because it was more popular. Tragedy of the commons.
If it's any consolation, maybe jj will take over. I haven't tried it yet (I use the staging area a lot in my workflow), but AFAIK they made the choice to be git-compatible which means it's not a choice between one or the other but lets people and teams migrate gradually.
FWIW you can still have a staging-area-like workflow with JJ - it's just that while git has "commits", "the staging area", "the working directory", and "stashes" as four separate concepts with four separate toolkits, in JJ all of those things are "commits" and a single toolkit works with all of them :)
I hoped that would be possible but wasn't sure, thanks.
Mercurial was safer and better. I still use it and it's still safer.
The bookmarks feature which is supposed to be the solution for short-lived branches is hard to understand though. I'm probably dumb but I can't work it out and hence the overall tool is that much less useful.
It needs a github-like website and Heptapod would be great if I could use it - I've set up a project, been unable to do anything and then had it all closed down. OTOH self-hosting is a lot more feasible nowadays with today's fibre connections.
I hated so much how Mercurial dealt with short-lived branches, that after seeing how Git did it, I've never looked back. I also remember how some people told me to use the quilt or something extension to manage patches, but it was too complicated for me.
I always preferred Mercurial to Git, but now that Jujutsu is out there, I think Mercurial's demise is all but guaranteed.
If there were a reliable way to use Mercurial on a Git repository, it could live. But why bother when one can use Jujutsu?
Jujutsu took the best parts about mercurial and enabled these to be used with git and git forges. Once I had gotten used to jujutsu I didn’t really miss mercurial.
I haven't used hg in years now, bitbucket discontinuing support is what killed it for me and I had no interest in self hosting.
My git usage is very basic, my gitconfig has been pretty much untouched for years but on those occasions where I get stuck or hit something I end up searching and usually get through a bunch of posts/comments/sites and wish I was using hg.
At my previous big tech employer we used to have a mercurial layer on top of our legacy version control system. I loved it, much simpler and more clear than the typical workflows I have to deal with in git. I get it that with git you have more power, but do really most teams need that?
Mercurial wasn't as simple as Subversion. But with hg I still felt like understanding 80% of what the tool had to offer and actually being able to mold the timeline the way I wanted.
Git has so many gotchas, bells and whistles that whenever I'm doing something out of the ordinary I'm wondering if there isn't an easier / canonical / smarter way I should be doing it.
There was a quote somewhere about Mercurial having a mental model small enough you can fit in your head - and that was the big win for me.
It was also fast and had very clean, easy to contribute to code. I remember submitting a patch and getting a bit of Python education from Matt, which was very useful.
Git is fine but it's inconsistent enough in the interface department, even after all this time, that I still get regularly frustrated. On the other hand, you can't just break a workflow that already exists and I very much appreciate it scales to work far beyond mine.
I do like that the git people are doing the difficult work of improving the UI over time - it's hard to change the engines while the plane is flying!
git's mental model is very, very small, if you care to learn it. Then all the commands and their "inconsistencies" start to make sense - they operate on the model almost without any magic, and not on whatever is user's intent (it can vary a lot)
> Mercurial wasn't as simple as Subversion.
What? Subversion is by far the most complex versioning software I've ever used.
> Git has so many gotchas, bells and whistles
The Git UI leaves a little to be desired. But inside, Git is basically just blobs, trees, commits, and refs. It'd be hard (or impossible?) to find a conceptually simpler versioning system.
You're right about SVN's guts vs Git. I should have been clearer that I was referring to the client cli only.
Just an example that I recently noticed when I briefly had to use svn again:
Show a diff: svn diff / git diff
Show log with diffs: svn log --diff / git log --patch
Git calling the same or similar things different (or just terrible - tree-ish? ref?) names is one of the worst things about Git.
I found that doing anything even slightly unusual in Git was pretty incomprehensible until I learned its internals, and pretty easy once I did. Fortunately as you say the internals are conceptually pretty simple to learn.
Git doesn't give you more power, it just makes their internals your problem.
Do you actually even have more power with git?
No, it's the opposite
I loved Mercurial - still do I guess as I just installed it on the Linux Mint VM I keep around for messing with Linux. The thing I really liked about it was TortoiseHg, which provides integration between Mercurial and Windows Explorer. There is a similar TortoiseGit but, at least back when I was doing serious development, it had quite a few problems.
FWIW, as a recent user of TortoiseGit, it seems pretty decent nowadays.
That's what I expected - I was sure the Tortoise guys would get it right eventually.
Glad it is still around. hg was the better experience as it had fewer command surface. But slower, and sometimes you need the extra detail commands to analyze the history.
I sincerely hope jj gets the recognition it deserves in coming years.
Anyone know what Matt Mackall is up to these days? He started Mercurial and got people involved early on with a lot of enthusiasm, you could tell he cared about what he created and the people who joined him ("hg crew"). I learned a lot from him on how to think like an engineer and saw him manage different personalities in the project in a kind and sincere way (I think this was around ~2010).
FYI: Matt Mackall is now Olivia Mackall [1], so that can make searching for things harder. Looks like they work at Valve, now? Agreed that they were a really stabilizing and healthy personality in the project, and made a lot of good early decisions, which Mercurial has continually benefited from. Eg: the commitment to backwards-compatibility [2].
[1] https://repo.mercurial-scm.org/hg-stable/rev/d4ba4d51f85f
[2] https://wiki.mercurial-scm.org/CompatibilityRules
Olivia uses she/her, and I believe she is now retired :)
I owe her and the Mercurial community a great debt. The community taught me how to think like an engineer building infrastructure and the importance of backwards compatibility, something I've tried to carry forward in tools like cargo-nextest.
+1 I learned so much about software engineering (Software design, network protocol incl how to handle backward compat in sane ways, ...) thanks to the mercurial community.
My first full time job after university was using hg, and particularly https://tortoisehg.bitbucket.io/ made it really pleasant.
Prior and post that I'd always used git but I'll always have a bit of a soft spot for mercurial, especially as our forge usage at the time predated strict guardrails and controls - we did code review, but it was your responsibility to tag the appropriate people and wait for them to respond, if you felt it was necessary to merge prior to that you could - but better be ready to defend that decision.
Git likely won because of GitHub; I abandoned Mercurial quite early when it was clear that it had lost. It was good, but I cannot imagine using it today.
I am glad the Hginit is back https://hginit.github.io/index.html
Not sure why it has to disappear in the first place.
What's going on with hginit.com now ?
I just installed it on a Raspberry Pi (with an otherwise too-small-for-any-other purpose SSD) for use at home. I wanted something with low power consumption, and I didn't want to have a single point of failure by running it in a container alongside everything else.
The only hiccup was forgetting that when pushing via the SSH connection, it will have paths relative to the home directory of my hg user.
Mercurial is just better than Git. But GitHub won and so Git won.
IIRC Facebook switched to HG from SVN in the 2010s, one (main?) motivator being that the single repo was getting too big and svn’s only way was to start splitting it up. Which was against the philosophy of openness of the single repo. No idea what’s Meta doing now.
Last I checked (2 years ago) Meta was using Sapling, a very heavily customised open source Mercurial frontend with proprietary backend.
FWIW the Sapling frontend can also be connected to a Git backend, and I've been using that for all my open source projects to get the best of Mercurial's user experience niceness while collaborating via GitHub <3
Still Sapling + Eden to make local work on a sparse checkout viable.
Related: https://github.com/cloudstreet-dev/What-we-lost-when-Git-won
In my time with it, about ~20 years ago, it had a lot of nice features for instance hg came with a web server/interface out of the box.
I liked too. It was `hg serve` [1]. Instant web interface and an easy way to share the repository with other people (assuming your computer was accessible from the LAN or Internet).
[1]: https://mercurial-scm.org/help/commands/serve
Just a bit off topic. But is anyone still using git or whatever versioning in the traditional way? I hardly see myself using the git command these days. Everything I handled by Claude. From pulling to pushing to solve merge conflicts and more. In a way it doesn't matter anymore whether it's git or something else.
hg (fig) was definitely my favorite frontend for source control at google.
Our shop uses Mercurial, and I hop we never move to Git, but I do see us eventually moving to whatever is next.
Mercurial was the only Python application that worked, so van Rossum and crew did not like it and had to move to git and then GitHub (Ruby).
Python developers, especially core are used to perpetually broken software and don't like stuff that just works. As long as the software is in Python - C (git) and Ruby (websites that used to work at that time) are fine.
Honestly hg shot themselves in the foot by not releasing any stable API and making developers use their CLI interface. Between that and the performance and dynamic typing issues of Python, it was almost sure to lose the race to Git.
Git doesn't have any sort of API either, libgit2 is unofficial.
Git's filesystem layout has always been stable and Git encouraged developers to target it. Mercurial strongly cautioned developers away from its filesystem or source code APIs and told them to subprocess out to the CLI. To be clear, I dislike Git, but it was _designed_ with developers in mind (which isn't to say that it has a nice API, only that the Git people intended for Git to be extended). I prefer Mercurial's concepts and interface, but Mercurial was somewhere between indifferent and hostile toward developers.