Fossil
fossil-scm.org
fossil-scm.org
I don't like the monoculture either, but at this point some knowledge of git is an essential skill. In fact pretty much the only chance you have of avoiding git would be to build your own stuff on something else, hire/collaborate only with others who share the same view, and never interact with basically any other open source.
So given some git knowledge is necessary anyway, there are alternatives interfaces to git that solve the simplicity problem, and any other tool is going to have friction (small ecosystem, few integrations, smaller community), why use something else? (I mean this genuinely)
I just don't see the dominance of git changing for the next decade, especially if there's no answer to my first question (for whatever aims to replace it). It's not even like there's git lock-in: it could be replaced, just like SVN was. It's just it doesn't seem like there's compelling reason to, and thus no desire.
I have never used, but I’m interested in trying… maybe some personal project in the future :)
Also, as someone who really liked Monotone before git won, having your whole repo in a single file is just nice for the same ops-related reasons it’s refreshing to be able to download and run a self-contained binary rather than needing an installer or whatever.
(Yes I obviously still use Git as well)
I worked with a guy once who took a principled stand against git and (almost) exclusively interacted with the org's git repos through Hg-Git: https://hg-git.github.io/
A. He liked Mercurial more, although I forget all his reasonings. Mercurial's commands are definitely better named, no contest.
B. He hated that git was "winning" and so tried to help prolong Mercurial's lifespan by writing plugins and helping out in the ecosystem.
The big three I know of (github, gitlab, bitbucket) are all git-only.
(Unaffiliated, but a happy user.)
If you click on the link it lists a ton of additions to git. The first bullet lists "bug tracking, wiki, forum, email alerts, chat, and technotes" none of which are supported by git.
Then again, bullet 4 is "Self-host Friendly" (every git client is also a full git server -- after all it's fully distributed) and bullet 5 is "Simple Networking - Fossil uses ordinary HTTPS (or SSH if you prefer)" Pretty sure git supports https out of the box these days (though I don't use it -- ssh is so much simpler).
So yes, I would call those killer features for those who want them, and are pretty unique to any revision control system I've used, though some centralized sites have encrusted git with those features. Having them distributed sounds like a good idea.
Sure, for code. But fossil distributes issues and wiki the same way, which makes comprehensive self hosting much easier.
It’s open to objections: first, Github does this already, so it’s a solved problem. Second, lack of focus. Why would VCS experts be good at designing a compelling forum or ticketing system?
A more compelling selling point, which is buried in the docs, is the use of SQLite as storage. Apparently this makes it easy to traverse a commit’s descendants, which git struggles with.
I can buy the reasoning that it's a solved problem with additional software, but citing github specifically here is absurd.
Github is a SaaS, centralised, closed source and unfederated platform.
I can understand that you think they're a good service, but there's a huge difference between something like git (which is a tool and a flexible one) and SVN, vs a hosted service.
You can think of it like VLC media player vs youtube.
No, you're misunderstanding the feature (or they don't explain it well). GitHub solves the problem on their platform yes, but not in the protocol (git) itself.
Fossil embeds that data in the repository, all the data is local. Fossil works perfectly with the "local first" doctrine, something Git also does, but not for everything around Git, that Fossil embeds into the repository.
Basically, Fossil is Git + GitHub but self-hosted and offline first. You pull down a Fossil repository is like pulling down the Git repository + all related GitHub data.
> Why would VCS experts be good at designing a compelling forum or ticketing system?
They don't have to be experts about something to make it good enough. For a lot of us with poor internet connectivity, the offline first doctrine is essential in a lot of the work we do, as internet can disappear at any notice. Fossil works much better for this, as I can still read the wiki and tickets when my internet drops, while if I were using GitHub, I'd have to results to writing my own code/use 3rd party software to get the same experience.
The other argument, that this is needed for offline use, is more meaningful. I don't personally think this is a big enough deal to be the USP. But sure, for some people it may be.
As an interesting example, the very website in this submission (fossil-scm.org) is in fact just a standalone Fossil website, running as you'd run your own repository. Everything that is on that website, is something you get for free for each repository you use Fossil for.
Oh, so systemd of SCM world?
Also, making moving out of it extra hard by embracing everything.
In that sense, I don't believe Git has a knowledge lock in, because it is easy for anyone to switch or to simultaneously use available alternatives. It's all about the ecosystem, and really, Github.
Can you elaborate on this please?
Fossil is basically GitHub itself stuffed inside the DVCS. Issues and the wiki are stored alongside the code.
But that doesn’t make git or fossil right for every project.
Fortunately you can easily switch between them.
See https://www.fossil-scm.org/home/doc/trunk/www/fossil-v-git.w...
If I was to have to build a VCS from source decades from now to get to my repo, I believe that fossil will be much simpler than git (which has a fair amount of build dependencies).
Other commenters are talking about GitHub/Discord/Jira etc but I think that's missing the point a bit - in 20 years time are those closed sass services still going to exist? If they are, is your project still going to be there? And will you remember which flavour of the month sass you used for tickets?
This is a strange thing to say. It's like saying the only chance you have of avoiding python is to use something else.
> never interact with basically any other open source.
You don't need git knowledge to download an open source library from GitHub (there's tarballs on the web page), or to use your languages package manager, even if it speaks git under the hood (go get/npm)
> I just don't see the dominance of git changing for the next decade,
The dominance of git is real particularly in certain industries, but even now git still has terrible support for huge repositories and binary files, partial checkouts, submodules. The problem is that most projects are suited just fine to the limitations of git, (and SVN and p4 and hg), so the value comes in the supporting tooling and infrastructure. Why would I choose fossil over git when for free I can get everything that GitHub/gitlab provides for me?
In my somewhat simplified view, Git's 'killer-feature' by now is simply its ubiquity.
By the same token, for Fossil's this could be claimed as a single binary that does it all.
Just download the binary, and you've got the ubiquitous set of project-related facilities: VCS/tickets/docs/wiki/team/forum/chat with simple aporoach to sync and self-host.
So I guess the killer feature for Fossil would be Fossil in that case, as Fossil is basically Git + GitHub but in one package, that works offline and local first.
Sadly, it's not git and nobody else is using it, and I'm not a fan of using lots of nonstandard tech.
It seems to be better than Git, but there's no free hosts that are as big as Github/Gitlab, and convincing other people to switch technology is very hard.
Plus, the people obsessed with clean commit history won't like the lack of easy squash and rebase.
You can import an existing Git repository into it really quickly.
The app itself is built by the SQLite team, and takes full advantage of SQLite, in particular Many Small Queries Are Efficient In SQLite: https://www.sqlite.org/np1queryprob.html
Why should one use that simple forum instead of Discourse? Why should one use the simple wiki than say Wiki.js?
And of course the toolings are already way behind as in there's not a single GUI client?
And it doesn't seem to have any social login, so people have to create an account instead of using GitHub logins.
Who uses it outside SQLite team?
> Who uses it outside SQLite team?
Tcl/Tk is the only big user of Fossil, as far as I'm aware.
Also, fossil feels like someone built it for almost NIH reasons, which I freely admit is just my outsiders impression, but it feels to me like the tcl community tends to dogfood themselves a lot for cultural reasons.
But fossil is already more complete and battle tested than anything this lowly amateur will accomplish in his career.
The author of Fossil is drh, who is a lot more famous for being the author of sqlite (who also happens to be a major player in the TCL community). If I remember correctly[0], Fossil was written a) to dogfood sqlite (fossil uses sqlite) and also b) to get something able to not just do the VC part for sqlite, but tightly integrated issue tracking, docs, rel-eng, etc in one single system, as well as some VC-related features not present in git/mercurial and other competitors of the time (and in many cases not available in git or mercurial of today).
[0] I looked into all this new-fangled DVCS stuff back then, because it looked really interesting compared to subversion that I was using at the time, and was a pain.
Considering the rest of the SQLite development toolchain, it would be more surprising if they _didn't_ make something like fossil.
Although shunning mainstream tools like git is probably rooted in a "not invented here" attitude, I love that they bother to do so. Of course it's interesting to see what bespoke solutions they come up with, but it's amazing to see such a highly complex piece of software come with such a trivial set of dependencies. Indeed, even if it's antisocial for the SQLite devs to play in their own sandbox, the end result is nothing but increased portability for the rest of us.
(arxanas shared a link to my project elsewhere in this thread :))
Are you aware that Git was not the first version control system? That doesn't mean Git was built for NIH reasons.
https://news.ycombinator.com/item?id=28168632
https://news.ycombinator.com/item?id=27736980
https://news.ycombinator.com/item?id=26578581
https://news.ycombinator.com/item?id=24643200
https://news.ycombinator.com/item?id=21974942
https://news.ycombinator.com/item?id=19006036
https://news.ycombinator.com/item?id=17230766
https://news.ycombinator.com/item?id=15752725
https://news.ycombinator.com/item?id=13668952
https://news.ycombinator.com/item?id=12673229
https://news.ycombinator.com/item?id=10737131
https://news.ycombinator.com/item?id=8697028
https://news.ycombinator.com/item?id=3036124
https://news.ycombinator.com/item?id=27236708
The reply I really wanted to write was 'the point is to confuse passers by', though.
https://hn.algolia.com/?dateRange=all&page=0&prefix=true&que...
I find the standardization that git has brought to the industry to be wonderfully simplifying for the lives of people developing software. If only we could all agree on things like a programming language, or a database - how much more productive would we be? How many life years do we spend trying out or learning new technologies that don’t go anywhere, or which only offer marginal benefits in exchange for real sacrifices? At least for now, source control doesn’t have that level of constant churn which is so commonplace in software.
To me, the benefits of standardization on git far outweighs any benefits that might be achieved by trying something new.
If we could all agree to a synchronised hype cycle that may help :)
* Does not track cherry-picks
* Does not track branch names
* Tags must be unique. (You cannot add a "release" tag to every release,
for example.)
* Not easily extensible - witness the pain and years of effort trying
to move from SHA1 to SHA2.I can remember where a place I worked used SVN, TFS, CVS and some commercial system (Perforce, I think), and it was really difficult switching between projects - I made quite a few mistakes.
Git isn't perfect, but it's more than good enough.
that sounds like reason to try to evolve instead kf saying "yeah well, we are done"
with some luck alternative tools come up with ideas, which become part of the default tool, maybe the default will change over time.
On the complexity front I have a couple perspectives. First, the fundamental concepts you need to contribute to a code base aren't that hard. Which I think is why it survives and thrives. But the CLI tools are NOT very friendly. So secondly, you might think we just need better front-ends. And maybe there are by now? But I've hated every alternative front-end to git I've ever used, but maybe that's just me. I also don't use an IDE, so I'm obviously a bit of a dinosaur by nature.
Why must it be so sufficient, with that narrow view on sufficiency. Fundamentally: If people annoy it, it is efficient in bringing joy, which in my view life should be about.
But secondly if I take the same view on efficiency: I see so much time wasted with git, so many times a repository is broken due to a bad merge or somebody doing something wrong. This can't be peak tooling.
Only git for version control. Perfect? No. Good for everybody and every project? No. Good enough? Yes. Only UNIX for any non-GUI application, and now even worse, almost only Linux. Perfect? No. Good for everybody and every project? No. Good enough? Yes. Only C for system programming. Perfect? No. Good for everybody and every project? No. Good enough? Yes. Only x86(_64) for almost anything non-portable. Perfect? No. Good for everybody and every project? No. Good enough? Yes. (same for ARM in portable world).
We use mediocre tools which share best property: ubiquity.
World of Good Enough, world of "multi-tools" which are not perfect for any application, but good enough for almost all of them.
We lost a tons of bright systems, ideas, approaches.
Now it is common to say, that diversity and inclusivity in teams are very well for corporate culture, as each person brings his/her own point of view, and multiple point of views are better for final product or service.
But in world of modern IT tools I see opposite. Monocultures everywhere.
Sometimes I read about different IT (mostly hardware, or integraged hardwae & software) of 1985-2005 and it is very sad reading: there was a tons of different systems, approaches, architectures, languages, ways to build systems. Each and every of them were more suitable for niche tasks. And what we can see now? Linux, Docker, git, x86, C.
Yes, I know, that there is Rust and there is ARM (at least!), but it is, what, 1% of landscape? And no new or specialized OSes (even IoT, which should be hyper-effective and realtime often is built on Linux!), no new platforms which shine in one or other niche, nothing
Yes, again, I know about POWER and z/Architecture (or how it is called this month), but, again, it is drop in the ocean of same boxes with same OS.
It's pretty much Gitea in a single SQLite file (Wiki, tickets, version control, etc.). It's weird, but you can edit all of these things off-line and then sync everything back. `fossil ui` hosting the server for my docs locally, and from my home server is super useful and easy.
Some of the commands are super weird though, but I put them my homepage of my fossil docs :) The auto-sync feature helps since I work on multiple computers to help ensure that I don't get too far out of sync.
It currently is built on php/mysql with laravel with a web ui in bootstrap. This has become super annoying for a few years now because once a year or so I have to port it to a new version of laravel and cope with the new javascript framework du jour that it now promotes.
This would be fine, of course, if you run an agency which extracts money from clients for continuous upgrades, but not for a personal project.
Fossil, on the other hand, is built for longevity. It is a single executable and comes with all tools built in.
https://fossil-scm.org/home/doc/trunk/www/fileformat.wiki
This is from the same person who made SQLite.
I think its brilliant having issues tracked in fossil. Part of the promise of github is that the user is in control, since your computer is a first-class citizen of a repository. You have all the code, and all the history on your computer. You can use github, but you don't need github, since its just another node in git's distributed network.
But that promise falls apart with issues and pull requests. Issues and pull requests don't get replicated by git. If github goes down, you can't interact with issues. If github ever turns evil, or you decide you want to self host git over ssh or something, you lose the history of all your issues and conversations.
Git is a distributed, replicated data format. Why are issues fully centralized? Its bizarre - You can have a project on both github and gitlab. And you can replicate commits to both. But you can't replicate issues using the same mechanism.
Fossil is far from perfect, but I think putting the issues and stuff into the repository itself is brilliant.
Here's git's official website, which is very good, and was started by a GitHub cofounder, but is run by the git project rather than GitHub: https://git-scm.com/
The desire to move off github (and therefore any additional services besides hosting you were using, which issues are just one -- new ones keep getting added to try and cement your dependence) is something more worthwhile of thought than the idea of github turning evil. Fortunately that desire is an actual thing that's very common, or at least the desire to not be wholly dependent, and is why many people don't even use github issues to begin with even if they use github itself for hosting (or some other non-issues features). So there's not a big problem, and even if you start with using github issues, there are various migration tools to move github issues out to [alternative]. (And of course github issues have their own merits, people frequently want to switch to them! So similar migration tools exist to move from [alternative] to github issues, I wrote one for Jira years ago.)
It is brilliant to integrate things with the decentralized source control itself, you get free backups and deciding to migrate to something different in the future is easy, I think it's an overlooked approach for a lot of people. (It seems less overlooked when it comes to documentation in various forms like developer-focused .md files, or broader full static html websites which github can conveniently host for you.) Fossil is well-worth investigating for this free integration to see if it meets one's needs. But of course nothing stops you from doing it with git yourself in various ways. For personal projects, I'm pretty satisfied with being as minimal as having an issues.md file and moving things to an issues-closed.md file when I close one. I've also used the git-issue extension (https://github.com/dspinellis/git-issue -- see also its bottom section of Related Work).
But despite its brilliance it's not always the right approach. It's very easy and reasonable to want more than what is realistic for something deeply integrated with the source code itself to provide, if only for inherent conflicts of desire, let alone any question of manpower. There are very good reasons to have entirely separate (and even multiple-of-kind partially overlapping/integrating/cross-referencing) systems for source code management, issue tracking, code review, forums, IM communication, wikis, public websites, docs (of whatever kinds and types for various audiences and authors, or public or not, or team-level spikes or plannings or retrospectives)... One aspect of Fossil I found weak was its user capabilities (https://fossil-scm.org/home/doc/trunk/www/caps/ -- no custom user categories alone is a deal breaker for so many things) but flaws in the execution of a fully integrated thing isn't really my point, my point here again is just that full integration despite its overlooked benefits and brilliance when applied to certain things is still not necessarily the right choice for something.
Github has a nice, documented API to export issues, including issue history. Many open source tools, including Gitlab, are happy to import that list into their own database:
https://docs.gitlab.com/ee/user/project/import/github.html
Granted, this has a problem if Github suddenly, without any warning, decides to turn off that API. Or if it goes down and loses data permanently. I find both of those scenarios pretty unlikely, but it you are worried about them, it should be pretty straightforward to have some sort of export script in cronjob, to keep a local copy of issue database.
Centralized systems aren't bad per se, but the combination is incoherent. Github's architecture has some of the worst parts of decentralization, combined with some of the worst parts of centralization. Eg, Git's decentralized nature makes it extremely hard to evolve. (See: the desire to move from SHA1 to something more modern.) And we don't really get the benefits of a decentralized system, because I can't interact with issues while I'm offline, or while github is offline. And we're all locked in to github's ecosystem.
We can only fully interact with our opensource projects on github with permission, ironically, from Microsoft. And I don't know if I trust any big tech company enough to assume they'll remain a good stewardship over the world's opensource code.
The decentralized data store is right there in the middle of every github project. I wish github thought to use it.
But it seems that very few people care. How many people mirror wikis of their favorite projects, or backup their issues? I know only one such person, myself.
I think that for some types of data, distributed model with operations like "clone with full history", "pull", "merge", etc... makes very little sense. For example access control -- why would you ever want to clone this? If I set up a mirror of repo, I don't want to give their authors access by default. Same goes for discussions -- for a good discussion you want all members to see their own replies. Having original post appear on sites A, B, C; and reply 1 on sites A, B while reply 2 appears on sites B, C will be super confusing to users, and will likely just cause platform to be abandoned. And a lot of times, ticket list is better off shared, so people can coordinate the work they do even if they have their own fork.
Once you realize a lot of supporting services need to be centralized, all you can use "git" for is for data distribution. Is git better than SMTP for mailing list discussions? Is git better that SMTP / REST API for ticket list? Maybe, maybe not. But ultimately it does not matter, because most people do not care about mirroring stuff locally, and those who do care, can already achieve it.
All of these features cooperate and serve the same goal: coordinate the work product of people on a project, in a distributed fashion. One path to that is the nearly fully centralized model of GitHub. Another is the VCS + mailing list + bug tracker + wiki path, which requires considerable admin resources to manage, and at the end of the day is a pile of barely-cooperating services. Fossil's path is to put them all into one place so they all work properly together.
You can reference ticket IDs from a forum post.
You can point to a section of the timeline from a wiki article.
You can create diagrams in Pikchr format that live as version-controlled text in the repo and reference them from commit messages.
You can generate HTML diffs and include them into the body of a Markdown chat posting for discussion of a proposed change before committing it.
Etc., etc. It's all communication, which you need when you have multiple people working on a project, especially across time zones.
Edit: I even have receipts. Me, 11 years ago: https://news.ycombinator.com/item?id=2524574
> Am I the only one who's actively suspicious about this kind of thing? With Git, I can use whatever wiki, ticketing, documentation, and blog features I want. I don't want my VCS to be a Lotus Notes for software development.
The forum, chat, and Pikchr features have been added since then. Indeed, everything in the Fossil ChangeLog has been added since then, because it cuts off a few months after your post: https://fossil-scm.org/home/doc/trunk/www/changes.wiki
In 2011, I wanted the flexibility to use other wiki, ticketing, documentation, and blogging systems. In 2022, I also want the ability to use other forum and chat systems. In 2033, I will probably want the ability to choose my own text editor. I have never complained that Fossil is lacking in features, especially features that are not directly part of version control.
I did eventually try Fossil somewhat seriously for a personal project last year. I gave up and moved back to git, and don't think I'll try Fossil again, either personally or for a broader group or company project. I still find the idea of full integration interesting, just not Fossil's execution of it, not to mention some disagreements with the SCM philosophy itself. (For some it's about rebase, for me I realized I rather like the concept of staged/unstaged files or even Perforce-style pending changelists.) Meanwhile a collection of services approach actually works and cross-integrates pretty well especially when you don't have the requirement to try and self-host everything and give yourself that administrative overhead. And you'll get non-ghetto (for lack of a better descriptor) versions of those services.
Still, I don't recommend against Fossil, it's clearly good enough and aligns very well with certain values, people should evaluate it for themselves.
Considering this is the largest point between fossil and git that I can see, maybe we should just make it so git has the capability to expose an extension API for plugging additional capabilities like this in.
Then we could probably all use git and just be happy. You could theoretically use fossil's issues and wikis with git.
I can imagine someone could feel reluctant to “Fossil” only because of the name.
Also this reminds me about the Gimp – somehow not so catchy nomenclature.
Also, a single-file repos are akin to a historical artifacts of developers work and ideas surviving through the times.
Don't we all have tons of projects sitting in their various state of abandon under the work directory? Well, Fossil lets us sweep the shame and organize these 'fossils' in a neat set of individual Fossil repo files in a grand /fossils or such directory. All repos 'readily' waiting to be respawned into a fresh work directory checkouts to perhaps offer some chance of completion.
By the way, looks like the Fossil repos can be queried directly without a check-out into a separate directory. It simplifies the review of the history and contents.
Honestly all I want is a git command that runs
git commit . -m timestamp git push
I probably do something like that every 20 minutes or so.
Since everyone just stepped on to use git without thinking, "Because Linux uses it", we have to endure the pain of unfriendliness of git and it's already like S3 where if you want to start any competing project, it has to be git compatible to even get started.
I suppose there are SubGit, hg-git, bzr-git trying to keep their respective version control while facing git on the other end.
Somehow people miss these messages and projects like Bazaar got ran over by git for no good reason.
> We take great pride in making Bazaar easy to learn, easy to use and suitable for everyone, not just elite hackers.
> Ease of use is a core value for Bazaar and there are many places where our focus on usability shines though.
http://doc.bazaar.canonical.com/migration/en/why-switch-to-b...
I thought people were looking for tools with this philosophy.
Today, all the VCSs offer more-or-less the same set of features (Fossil being somewhat an exception in that regard). But in the past, there were quite a lot of differences. A lot of arguments for git were very technical.
- Patch-based VCS (darcs, pijul)
- CRDT (live collaboration)
- Image-based systems (Smalltalk)
- Features that require the server to do stuff that Git protocol won’t do, like push notifications
In other words, I can’t imagine leaving Git unless it’s as big of an upgrade as Git was to SVN. And you can’t do that while being wire compatible.
> so it’s extremely to just back it up or keep the repo in cloud storage
A single big file seems like it would cause more issues for a generic sync utility. Though I guess I can imagine sync algorithms that work better with one style vs the other.
I'm not a fan of that decision. I don't want to live with my mistakes forever, especially on local branches.
[1] - https://www.fossil-scm.org/home/doc/trunk/www/rebaseharm.md
They ruin history, get in the way of bisect, may not be reproducible, etc, etc.
Sometimes they may be used but more likely as a superuser operation
[0]: https://docs.github.com/en/pull-requests/collaborating-with-...
Without the ability to rewrite local history, you're either not commiting often (risky) or leaving long streams of poorly documented "work-in-progress" commits in the main project history.
For simple squash of WIP, the private branch works quite naturally, and simply merges onto the normal branch. For more involved reordering, it would be too cumbersome to be practical.
However, Fossil philosophy advocates a factual approach to commit history, unlike a 'maintained' approach that's so widely expected by Git-based workflows, and which calls for the use of rebase.
Not to mention a stream of WIP commits that mean at any point I can diff the mess I've made in my editor to when it was last working (or any other time). Often that's enough to reveal the problem. If you don't commit often enough, you can't do that, but if you can't rebase, the history is full of junk that should be collapsed before going public.
In case anyone is confused, git amend can amend the changes, not just the commit message. I didn't notice this for a while, and finally got around to reading about how it works. https://git-scm.com/book/en/v2/Git-Tools-Rewriting-History
Very clever way of letting user fix some attribution mistakes/typos, yet not mangle the original commit.
However, for amending the actual file changes, like it could be done with git --amend, Fossil requires another proper commit; for historical amends, the new commit has to be put onto a new branch.
Fossil-scm website has a detailed help section for all commands, including the amend.
https://fossil-scm.org/home/doc/trunk/www/fossil-v-git.wiki#...
Git is just a "me too" of the BitKeeper source control system that Linus used before he wrote git. Linus isn't even trying to hide the fact--he explicitly says he wrote git because he had to move off BitKeeper after a license conflict and he needed something that was functionally similar for kernel development.