Hacker News Reader

Comment by noasaservice | Hacker News Reader

So, we either go to Github, where our licenses are abused for their shitty ML.

Or we pay $20/month to Gitlab. And I can't figure out how the quotas will intersect with "professional", if at all.

For us Open Source devs, neither is a good option. Although I have heard good things about sr.ht / sourcehut. And for the service, it appears to be fair https://sourcehut.org/pricing/

View on HN

Why do you need more than 5gb for a git repo? IMHO, any repo above 100mb should have an exceptional reason for being so big, and if it's just code a normal repo is more like 1 to 10 mb in size, max.

You don't even need to pay gitlab, free tier can do for most stuff and there's a generous sponsorship for Open Source.

I don't get all the hate Gitlab is receiving these days.

I do hardware hacking, presentations, code, pictures, and full media to reproduce what I've shared.

FreeCAD files get big. And to accommodate easy printing, STLs are also needed.

KiCAD for board layout and schematics can also get larger. And remember, these also have 3d board components too.

Presentations are naturally larger.

Full high-rez pictures eat storage like you wouldn't believe.

Code is small, thankfully.

One such device I have created is hovering around 4.5GB for a full reproduction for the current snapshot. And if you do the command to pull the whole history (and not current), its around 10GB. And, I'm not sure if GL is counting the whole history, or the current? And are they pruning old after a certain date?

> And, I'm not sure if GL is counting the whole history, or the current? And are they pruning old after a certain date?

GitLab stores the Git history with all data revisions as commits. To reduce the repository storage [0], older Git commits can be pruned but this would rewrite the Git history as an invasive change [1]. To store larger files, it is recommended to use Git LFS [2]. Alternatively, you can use the generic package registry to store data. [3]

[0] https://docs.gitlab.com/ee/user/usage_quotas.html#manage-you...

[1] https://docs.gitlab.com/ee/user/project/repository/reducing_...

[2] https://docs.gitlab.com/ee/topics/git/lfs/

[3] https://docs.gitlab.com/ee/user/packages/generic_packages/

It's not just Git. Gitlab hosts a container registry for every repo, so if you build and store a new container image regularly then you could easily top 5GB.

Larger heavily contributed to repositories can be quite large, even if they are just plain code. Sure, not quite 5 GB large, but still.

Linux:

  $ git clone https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
  $ du -h linux/.git
  ...
  2.8G    linux/.git

Git:

  $ git clone git://git.kernel.org/pub/scm/git/git.git
  $ du -h git/.git
  ...
  120M    git/.git

DefinitelyTyped:

  $ git clone https://github.com/DefinitelyTyped/DefinitelyTyped
  $ du -h DefinitelyTyped/.git
  850M    DefinitelyTyped/.git

home-assistant/core:

  $ git clone https://github.com/home-assistant/core
  $ du -h DefinitelyTyped/.git
  380M    core/.git

The above is also only taking into account repo size, while the GitLab limit applies to everything including release artifacts, CI build artifacts, hosted containers, etc. Many of which GitLab currently provides poor or sometimes even zero support for implementing purging.

I don’t think using the Linux kernel, which is the OG repo that git was created for, is a fair comparison. Though, I guess since it’s barely 3GB, it’s proof that you almost never need >5GB.

Build react hello world and you’ve got gigabyte docker.

I have recently migrated my projects to fossil (https://fossil-scm.org), hosting them myself on a cheap vps.

I also really got hooked on their code/project management philosophy, rethinking a lot on how I want to run collaborative projects in general

Personally I use codeberg[1] which doesn't have all the fancy bells and whistles but for personal projects it runs like butter.

[1] https://codeberg.org/

Git works perfectly fine without a web interface like that. It's not popular nowadays, but you can just keep the primary repository on your own server and merge patches by email.

>So, we either go to Github, where our licenses are abused for their shitty M

As long as it's on the internet MS can scrape the data so not being on Github is only a temporary defense.

But GitHub's TOS can only protect them from GitHub repos.

If it has impact on CoPilot or not, is not clear.

Except not all contributors to projects signed the TOS since you can upload an existing project with existing non-Github contributors. One person uploading a project does not override or change the license of that project. So MS is already not relying on the TOS for legal protection which means they feel legally they can train on any OSS projects.

Well, at least in the US training an AI will probably fall under Fair Use. In the EU there is an explicit copyright exception for data mining. So I don't think there's a legal obligation for Microsoft to only train within the bounds of public GitHub repos.

AI fair use is not really clear anywhere.

I think SFconservancy's articles about CoPilot are very helpful.

> Well, at least in the US training an AI will probably fall under Fair Use.

Provoking question: so, will it be Ok to train an AI on leaked Microsoft code the publish the models (not the code)?

Of course, MSFT won't accept this, but w know that big corps are hypocrite by default.

Hi there, in another reply I mentioned GitLab for Open Source provides Ultimate features and higher limits to Open Source projects for free: https://news.ycombinator.com/item?id=32387621

Yes, and even the Ultimate plan limits will be significantly impacted. If they are not already self hosting now is a good time to look into it or explore other lightweight options like Gitea.

Reply on news.ycombinator.com