Rendered at 11:48:56 GMT+0000 (Coordinated Universal Time) with Cloudflare Workers.
ramon156 2 days ago [-]
Readme seems generated then minimized to the extreme, to a point where I cannot follow it anymore.
> The JavaScript init function initializes exactly `at`; Git-root selection and `--here` are CLI behavior.
What does this mean? Maybe I'm missing something
Also some of the stuff in this README seems like it should be in comments above/in their respected code blocks.
It also did not tell me why rift is a better alternative. Because it's fast? git worktrees are also fast.
rmunn 2 days ago [-]
The AI instructions in the `specs.md` file gave me a couple clues, e.g. `at` is the directory where the worktree should be created.
And it's clearly using btrfs subvolumes for managing a collection of related Git working trees; there's a concept of "parent" and "child" worktrees.
I don't yet understand why it's better than worktrees, other than being theoretically instant to create new ones (which could, I suppose, be a noticeable speedup if your repo is very very large).
But yeah, some more hand-written instructions in the README would definitely be helpful. I'd be particularly interested to learn whether some of the common "gotchas" one can run into with worktrees are solved by Rift or not. (E.g., I've never needed to move my "root" git repo, but apparently that causes problems because the worktrees then can't find the root repo; does Rift deal with that situation correctly?)
zjy71055 2 days ago [-]
git worktree (or any COW snapshot like this) still leaves you reinstalling node_modules per tree and fighting over a dev server port. That's the actual cost, and none of these tools touch it.
So I gave up on parallelizing inside one repo. I run agents across different projects — one repo each — and stay serial within a single project.
madarco 1 days ago [-]
you are right, I've experimented with cp -a on macOS as well for https://github.com/madarco/agentbox and in the end found it's actually faster to use worktrees inside docker containers while mounting your .git repo inside them.
Then after the node_modules (or apt packages) are installed, take a docker commit snapshot.
Now I have truly isolated parallel workspaces in <10s.
Also the system was easy to adapt to cloud environments as well so now I have Hetzner, Vercel, Daytona as well (using their native snapshotting systems for fast boot after the initial setup)
jarym 2 days ago [-]
This! A hundred times over. It's hard enough having to review one serial set of changes managing parallel changes into a single code base has been a nightmare load on my brain so I avoid that unless I'm trying to prototype something quick and dirty.
lanycrost 2 days ago [-]
Don't see the real reason to use this, as well as readme file is too short and give no actual info what is the better versus existing tools. Can be called Yet another :D ...
sbinnee 2 days ago [-]
Is it just an experimental tool by opencode team? If there is some article about this tool, I would love to read it. It’s not clear to me why I should use this instead of git worktree.
simonklee 2 days ago [-]
It's a ~1d old experiment. Not sure why posted on hackernews.
thdxr 2 days ago [-]
this is a 1 day old vibe coded experiment where i'm exploring some ideas
rippeltippel 2 days ago [-]
I find it hard to understand what it is about. Better in what way?
orf 2 days ago [-]
I wrote something similar with go, but MacOS only.
Creating a worktree became instant, but the bottleneck shifted from that to git needing to build its index. Claude code runs `git status` in the background, meaning any speed gains are instantly gone.
pure-orange 2 days ago [-]
Currently it just sounds like an alternative to work trees, but with no explanation on how it’s better. Seems early stages, use of btrfs is cool, but unsure why I’d use this right now
pure-orange 2 days ago [-]
Digging in I now see how btrfs will also copy over non git tracked files, e.g. target/ dir. this is super nice actually.
luckymate 2 days ago [-]
If that achieves quick COW copies of whole repo and works on Mac OS that's the solution I've been looking for last few weeks. Internets and Claude were insisting that such copies are possible only on Linux via OverlayFS. Seamless switching between unrelated features in the same repo – here I come!
mikroskeem 2 days ago [-]
No. XFS too - `cp --reflink=auto`. or LVM / ZFS snapshots if you will
pikdum 2 days ago [-]
Neat! Would a similar approach work with ZFS instead of btrfs?
gucci-on-fleek 2 days ago [-]
With btrfs, you can freely create subvolumes and snapshots anywhere (including nested inside of each other), you can have thousands of them without any noticeable performance impact, and you can easily convert a snapshot to a writable subvolume. I don't have much experience with ZFS, but from reading another post [0], my impression is that this isn't really doable with ZFS. And based off of rift's Readme, I think that these features are required for it to work. But I'm not an expert, so I may be mistaken about something here.
How about cp --reflink? Supported by btrfs, bcachefs and zfs. It's quite not as fast as subvolumes in btrfs, but it should be plenty fast.
This should actually be a feature for git itself, if it's not already.
gucci-on-fleek 2 days ago [-]
That reflinks the files, which should get you the space savings, but I'm pretty sure that that still has to recursively copy every file in the directory, which can be fairly slow if you have tens of thousands of files, whereas btrfs snapshots can reflink the directory itself, so it should be faster.
Buy yeah, this should be equivalent in most cases, since I can't imagine that many Git repos have enough files for the difference to be noticeable.
_flux 2 days ago [-]
I don't have a reflink-able filesystem in this host, but I just tested that
echo 2 | sudo tee /proc/sys/vm/drop_caches
time cp -rl foo bar
took 2.6 seconds for 273000 files, so I think it's highly manageable. Reflinking might be a bit slower than hardlinking, though.
andoma 2 days ago [-]
Also you can create and destroy (assuming the 'user_subvol_rm_allowed' mount option) BTRFS subvolumes without being root.
apex_sloth 2 days ago [-]
Could be good option for rust projects with huge compile artifacts. Saving space and rebuild time.
lordforever7 17 hours ago [-]
finalyy the claude blocking edits on wrktree might be solved
singiamtel 2 days ago [-]
Will this replace /warp in Opencode? Seeing as it's made by the same team
throwwwll 2 days ago [-]
Brought to you by the infamous author of yet another llm harness - will exfiltrate all your data, then feign ignorance:
Search issues for "privacy" and some very worrying discussions are there.
Include closed issues, as some have been closed without resolution.
rubnogueira 2 days ago [-]
Does it work well when we have gitsubmodules?
I had some issues regarding that.
jauntywundrkind 2 days ago [-]
Love to see btrfs subvolumes getting used!!
I spent some time & tokens starting to work on jujutsu support for opencode. Whose workspace support is so so good. I wonder if JJ does reflink-- maybe gonna go add that, as low hanging fruit.
> The JavaScript init function initializes exactly `at`; Git-root selection and `--here` are CLI behavior.
What does this mean? Maybe I'm missing something
Also some of the stuff in this README seems like it should be in comments above/in their respected code blocks.
It also did not tell me why rift is a better alternative. Because it's fast? git worktrees are also fast.
And it's clearly using btrfs subvolumes for managing a collection of related Git working trees; there's a concept of "parent" and "child" worktrees.
I don't yet understand why it's better than worktrees, other than being theoretically instant to create new ones (which could, I suppose, be a noticeable speedup if your repo is very very large).
But yeah, some more hand-written instructions in the README would definitely be helpful. I'd be particularly interested to learn whether some of the common "gotchas" one can run into with worktrees are solved by Rift or not. (E.g., I've never needed to move my "root" git repo, but apparently that causes problems because the worktrees then can't find the root repo; does Rift deal with that situation correctly?)
Then after the node_modules (or apt packages) are installed, take a docker commit snapshot.
Now I have truly isolated parallel workspaces in <10s.
Also the system was easy to adapt to cloud environments as well so now I have Hetzner, Vercel, Daytona as well (using their native snapshotting systems for fast boot after the initial setup)
Creating a worktree became instant, but the bottleneck shifted from that to git needing to build its index. Claude code runs `git status` in the background, meaning any speed gains are instantly gone.
[0]: https://news.ycombinator.com/item?id=45077119
This should actually be a feature for git itself, if it's not already.
Buy yeah, this should be equivalent in most cases, since I can't imagine that many Git repos have enough files for the difference to be noticeable.
https://github.com/anomalyco/opencode/issues/10416
Include closed issues, as some have been closed without resolution.
I had some issues regarding that.
I spent some time & tokens starting to work on jujutsu support for opencode. Whose workspace support is so so good. I wonder if JJ does reflink-- maybe gonna go add that, as low hanging fruit.