josh VS VFSForGit

Why use submodules when you can properly vendor the upstream git, and export/import commits without breaking hashes on either side?

https://github.com/josh-project/josh

We've been using josh at TVL for years and it's just amazing.

Scalar explicitly does not implement the virtualized filesystem the OP is referring to. The original Git VFS for Windows that Microsoft designed did in fact do this, but as your second link notes, Microsoft abandoned that in favor of Scalar's totally different design which explicitly was about scaling repositories without filesystem virtualization.

There's a bunch of related features they added to Git to achieve scalability without virtualization. Those are all useful and Scalar is a welcome addition. But the need for a virtual filesystem layer for large-scale repositories is still a very real one. There are also some limitations that aren't ideal; for example Git's partial clones IIRC can only be used as a "cone" applied to the original filesystem hierarchy. More generalized designs would allow mapping any arbitrary paths in the original repository to any other path in the virtual checkout. Tools like Josh can do this today with existing Git repositories[1]. That helps you get even sparser and smaller checkouts.

The Git for Windows that was referenced isn't even that big at 300GB, by the way. Game studios regularly have repositories that exist at multi-terabyte size, and they have also converged on similar virtualization solutions. For example, Destiny 2 uses a "virtual file synchronization" layer called VirtualSync[2] that reduced the working size of their checkouts by over 98%, multiple terabytes of savings per person. And in a twist of fate, VirtualSync was implemented thanks to a feature called "ProjFS" that Microsoft added to Windows... which was motivated originally by the Git VFS for Windows they abandoned!

[1] https://github.com/josh-project/josh

[2] https://www.gdcvault.com/play/1027699/Virtual-Sync-Terabytes...

Good overview, I know these sorts of pains well. Lots of hard questions and few definitive wins/right answers. How to organize a massive repository out in the open is still an open question. On that note, recently, I've been experimenting with this project called josh, which basically is like 'git subtree on extreme steroids, functioning as a git proxy':

https://josh-project.github.io/josh/

It basically lets you unify/view many repositories as a single one, or equivalent to split a mono-repo into smaller sized units of work for CI, specific teams, etc. It's bidirectional, so you push and pull from josh and everything goes into a single linear history in the mono repo. And because it's bidirectional, people in the mono-repo can still do things like make large-scale atomic changes across all sub-repositories, and those get reflected.

Josh currently isn't suitable for a lot of workloads due to various reasons (authentication is one that stands out), but it's actually the first tool I have seen that manages to offer BitKeeper-like "subtrees" that work really well, at scale, for large repos and teams. It requires some care to make sure "sub-trees" can be usable units of work, but it was one of the best features of BK in my opinion and really great for people doing one-off contributions, or isolating trees/changes to specific developers.

I'd be interested to know if there are other open alternatives to this. It's a nice point in the design space between solutions like "integrate with the filesystem layer to do sparse clones" or "just split up the repos."

With regard to repo composition, I have been following this project: https://github.com/josh-project/josh

It's not only Windows that uses Git at Microsoft, but Sharepoint and Office (which includes the on-prem version of SharePoint). In terms of repo size Windows and Office are similar. I was part of the team that migrated Sharepoint from a Perforce clone to Git and helped build the tooling to allow Office to move as well. VFS for Git [1] and Scalar [2] are really good pieces of software.

[1] - https://github.com/microsoft/VFSForGit

[2] - https://github.com/microsoft/scalar

Congratulations! That means you basically figured out how the clone procedure works and found a way to do so just in a partial way (also in an unsafe way). But it is a cool idea, nonetheless.

Also check out the Scalar [1] project and its predecessor, GVFS [2], both from Microsoft to manage their monorepo via a VFS layer.

[1]: https://github.com/microsoft/scalar

[2]: https://github.com/microsoft/VFSForGit

VFS for Git is still Open Source: https://github.com/microsoft/VFSForGit

Microsoft's blog posts have indicated a move to use something as close to off-the-shelf git as possible, though. They say they've stopped using VFS much and are instead more often relying on sparse checkouts. They've upstreamed a lot of patches into git itself, and maintain their own git fork but the fork distance is generally shrinking as those patches upstream.

https://github.com/microsoft/VFSForGit

better than it used to, with the caveat that git in particular is not and has never claimed to be good at versioning blobs.

Oh dear you're not going to like this.

It’s virtualised file system: https://github.com/microsoft/VFSForGit, only downloads what you actually use. Same thing in every large company, but different implementations.

This is where something like VFSForGit[0] helps out. Instead of cloning the entire repo, it creates a virtual file system and fetches objects on demand. MSFT uses it internally for the Windows source tree (which now exceeds 300GB).

[0]: https://github.com/microsoft/VFSForGit

There's also a performance consideration. Not excluding /bin/ or /obj/ folders means dependencies are being tracked as well, and sometimes dependencies themselves are bigger than the program's source code itself. This is commonly the case with node projects, as the node_modules folder can balloon to hundreds of megabytes. They should never be tracked in git due to the nature of how git's internal database works. For e.g. if you delete a dependency because it's no longer needed, you can never fully reclaim that disk space (at least for the master branch) as git will need to keep the binary data stored in its internal tracking database because a previous commit in the master branch has captured the data. As you make more branches, git needs to store the data required to reconstruct your repo to a different state when you switch branches. When a branch has changes measured in the kilobytes, check out is very manageable, but when the differences balloon to many MBs due to the presence of heavy binary files, then checkout between different branches/commits can get very slow. Though, this happens anyway when source code data eventually reaches a certain threshold, beyond the hundreds of megabytes, it's made unnecessarily worse by including any binary files. It's one of the reasons Microsoft created VFS for git: https://github.com/microsoft/VFSForGit.

You mean like VFSforGit? Or the successor for that called Scalar? This has been a solved problem. Microsoft moved their entire Windows codebase to git. There have been a ton of huge improvements to performance as a result of that. And the above two plugins are easily better ways to deal with what you're referring to without resulting to dead tech.