2. Raspberry pi as a forge for a long time: also check, the git server shim is super lightweight, its just an XRPC layer over git repositories + an sqlite3 database. there are folks running it on a riscv board with 512 megs of RAM.
3. Actions are critical and they should be runnable on my local machine: IMO this ask is slightly misplaced. it is mostly your build-systems' job to be hermetic, run anywhere, handle cross-builds etc. it would be really cool to "promote" results of such builds to the forge itself.
>it is mostly your build-systems' job to be hermetic, run anywhere, handle cross-builds etc.
yes, and... the idea here is that it would be neat to extend the hermetic builds idea such that this can be run locally / anywhere where there's compute easily. The root problem that's being called out here is that idea of running something until the CI says it's green when there's a change, commit, network call, in the cycle is a pain in the ass. (The best way to avoid this churncycle is to just never write bugs! TFIC ;P)
Surprised to see raspberry pi for hosting data that's supposed to be integral to a workflow. I've been burned too many times by SD card corruption in the past. Do you use NVME drives nowadays, just curious.
A remote execution cluster and CAS for build artifacts is a good way to avoid duplicate work on local vs CI, and avoid the problem of needing to trust local builds.
> Stuff happens in the wrong order. You know the PR. Commit 1: 'Feature.' Commit 2: 'fix.' Commit 3: 'fix.' Commit 4: 'actually fix.' Commit 5: 'please.' Commit 6, made at 11:47 PM on a Thursday: 'asdfasdf'. This person has a family. This person has hobbies. This person is, at this moment, crying. You don't want the feedback loop after the commit you want it before. Let me do an enforced pre-commit hook to run the jobs remotely on the forge and provide the feedback to the user before they push.
How would a pre-commit hook help? Would the developer not be crying and working late if their work was rejected by the pre-commit hook instead of the PR? Also, if the tests are so fast they wouldn’t block the terminal running `git commit` for more than a minute or two, you can just run the tests on the local machine, and you should be running them as part of your workflow.
> PRs are too inflexible. I don't need 4 eyes on every change, especially in a universe where LLMs exist. The global GDP lost annually to senior engineers staring at a four-line PR waiting for someone — anyone — to type 'LGTM' could fund a moon mission. A nice one. With legroom. Let me customize and more easily control this. If the person is a maintainer and the LLM says its low risk/no risk just let them go.
You can do this with the existing forges, you can give trusted people the right to bypass the rules. Or you could build your own small PR auto-approval bot, which hands the diff to a LLM, and if the LLM approves, the bot approves the PR on the forge.
I really like Gerrit's workflow with diff reviews as opposed to pull requests, but unfortunately compared to something like gitea it lacks everything else we've come to expect from git hosting (issues, project planning, etc) which makes it seemingly a hard sell for many. I really wish there was a nice diff review platform kind of like phabricator but alas.
One of my issue with the PR process is the way they are focused on comments first, instead of the code. You also have the same issue that email threads have, if you comment something substantial, with details you pretty much always get a response that will take one small point, and sort of ignore the rest.
I also hate how noisy the discussion part is, and how GitHub handles that noise by just… hiding part of the discussion. For controversial discussion it is such a mess and horrible to track what is happening.
It’s just not a great discussion platform, while also putting that as the default tab in the PR view
Stuff happens in the wrong order. You know the PR. Commit 1: 'Feature.' Commit 2: 'fix.' Commit 3: 'fix.' Commit 4: 'actually fix.' Commit 5: 'please.' Commit 6, made at 11:47 PM on a Thursday: 'asdfasdf'. This person has a family. This person has hobbies. This person is, at this moment, crying. You don't want the feedback loop after the commit you want it before. Let me do an enforced pre-commit hook to run the jobs remotely on the forge and provide the feedback to the user before they push.
Isn't this already totally possible? Or am I thinking subversion?
I started Free.ai as a weekend project with the same mindset. And a month in the work hasn't stopped. So I second this. Just find a good name, it helps.
If you want a certain app with a feature and the app isn't open source, then you may as well just clone the app and add the feature.
Claude Code and Codex (and other tools) have computer use and are perfectly capable of navigating, experimenting, cloning functionality, writing tests...
If the app is open source it's probably easier to just fork and add your features though. And cheaper.
The alternative to GitHub is already here. It is called self-hosting and there are many alternatives.
The Linux kernel is not hosted on GitHub and uses cgit. Others use GitLab, or Gitea and there is also Forgejo (Which Codeberg uses) that people are using and can be self hosted.
This is why now everyone is realising why "centralising everything to GitHub" [0] was a terrible idea and now GitHub has been (unsurprisingly) run into the ground.
> Well Y Does Some Of That
yes but tangled.org really does do most of that!
1. JJ as the VCS: tangled supports stacked PRs using jj change-ids. https://blog.tangled.org/stacking , we use it a lot to build tangled itself: https://tangled.org/tangled.org/core/pulls
2. Raspberry pi as a forge for a long time: also check, the git server shim is super lightweight, its just an XRPC layer over git repositories + an sqlite3 database. there are folks running it on a riscv board with 512 megs of RAM.
3. Actions are critical and they should be runnable on my local machine: IMO this ask is slightly misplaced. it is mostly your build-systems' job to be hermetic, run anywhere, handle cross-builds etc. it would be really cool to "promote" results of such builds to the forge itself.
>it is mostly your build-systems' job to be hermetic, run anywhere, handle cross-builds etc.
yes, and... the idea here is that it would be neat to extend the hermetic builds idea such that this can be run locally / anywhere where there's compute easily. The root problem that's being called out here is that idea of running something until the CI says it's green when there's a change, commit, network call, in the cycle is a pain in the ass. (The best way to avoid this churncycle is to just never write bugs! TFIC ;P)
Surprised to see raspberry pi for hosting data that's supposed to be integral to a workflow. I've been burned too many times by SD card corruption in the past. Do you use NVME drives nowadays, just curious.
A remote execution cluster and CAS for build artifacts is a good way to avoid duplicate work on local vs CI, and avoid the problem of needing to trust local builds.
Honestly, something like Radical seems a lot more seamless and complete. Or at the very least, I'm wary of Tangled's alpha tag.
> Stuff happens in the wrong order. You know the PR. Commit 1: 'Feature.' Commit 2: 'fix.' Commit 3: 'fix.' Commit 4: 'actually fix.' Commit 5: 'please.' Commit 6, made at 11:47 PM on a Thursday: 'asdfasdf'. This person has a family. This person has hobbies. This person is, at this moment, crying. You don't want the feedback loop after the commit you want it before. Let me do an enforced pre-commit hook to run the jobs remotely on the forge and provide the feedback to the user before they push.
How would a pre-commit hook help? Would the developer not be crying and working late if their work was rejected by the pre-commit hook instead of the PR? Also, if the tests are so fast they wouldn’t block the terminal running `git commit` for more than a minute or two, you can just run the tests on the local machine, and you should be running them as part of your workflow.
> PRs are too inflexible. I don't need 4 eyes on every change, especially in a universe where LLMs exist. The global GDP lost annually to senior engineers staring at a four-line PR waiting for someone — anyone — to type 'LGTM' could fund a moon mission. A nice one. With legroom. Let me customize and more easily control this. If the person is a maintainer and the LLM says its low risk/no risk just let them go.
You can do this with the existing forges, you can give trusted people the right to bypass the rules. Or you could build your own small PR auto-approval bot, which hands the diff to a LLM, and if the LLM approves, the bot approves the PR on the forge.
I really like Gerrit's workflow with diff reviews as opposed to pull requests, but unfortunately compared to something like gitea it lacks everything else we've come to expect from git hosting (issues, project planning, etc) which makes it seemingly a hard sell for many. I really wish there was a nice diff review platform kind of like phabricator but alas.
I recognise myself in this post without having realised it previously.
The PR review process is flawed, it adds something, but maybe not what it intends.
One of my issue with the PR process is the way they are focused on comments first, instead of the code. You also have the same issue that email threads have, if you comment something substantial, with details you pretty much always get a response that will take one small point, and sort of ignore the rest. I also hate how noisy the discussion part is, and how GitHub handles that noise by just… hiding part of the discussion. For controversial discussion it is such a mess and horrible to track what is happening.
It’s just not a great discussion platform, while also putting that as the default tab in the PR view
Doesn't sound too hard.
Why don't you see how far you get in a weekend with Claude.
I started Free.ai as a weekend project with the same mindset. And a month in the work hasn't stopped. So I second this. Just find a good name, it helps.
gitwheel.com
This can be the mindset now with a lot of things.
If you want a certain app with a feature and the app isn't open source, then you may as well just clone the app and add the feature.
Claude Code and Codex (and other tools) have computer use and are perfectly capable of navigating, experimenting, cloning functionality, writing tests...
If the app is open source it's probably easier to just fork and add your features though. And cheaper.
The alternative to GitHub is already here. It is called self-hosting and there are many alternatives.
The Linux kernel is not hosted on GitHub and uses cgit. Others use GitLab, or Gitea and there is also Forgejo (Which Codeberg uses) that people are using and can be self hosted.
This is why now everyone is realising why "centralising everything to GitHub" [0] was a terrible idea and now GitHub has been (unsurprisingly) run into the ground.
[0] https://news.ycombinator.com/item?id=22867803