Support Forgejo Actions if Forgejo is hosted on a subpath #443

JulianWgs commented 2026年02月06日 16:49:45 +01:00

Hi,

I talked with someone at FOSDEM about this, but I don't remember their name. They told be to open an issue here.

There is a long standing issue with the checkout action so that it doesn't work when the Git Forge (Github or Forgejo) is hosted under a supath. Github does not want to support this as it is only applicable to Forgejo instances. As the the repo for checkout action for Forgejo is a read-only mirror of the Github checkout action there is no way to contribute a fix that.

The fix (including test and doc changes) includes less than a 100 changed lines. I think it would be worthwhile considering to maintaining this out of tree patchset for Forgejo users. I am unfortunately not a JavaScipt / TypeScript programmer so I can't really take care of the code, but I would volunteer to test it.

There is also a rebased version from 2024.

Thank you & best regards
Julian

PS: When using Tailscale (as I am) to my knowledge it is not trivially possible to host Forgejo under a subdomain, so subpaths are kind of necessary in that case.

Hi, I talked with someone at FOSDEM about this, but I don't remember their name. They told be to open an issue here. There is a long standing [issue](https://github.com/actions/checkout/issues/1242) with the checkout action so that it doesn't work when the Git Forge (Github or Forgejo) is hosted under a supath. Github does not want to support this as it is only applicable to Forgejo instances. As the the [repo for checkout action](https://code.forgejo.org/actions/checkout) for Forgejo is a read-only mirror of the Github checkout action there is no way to contribute a fix that. The [fix](https://github.com/actions/checkout/pull/1305) (including test and doc changes) includes less than a 100 changed lines. I think it would be worthwhile considering to maintaining this out of tree patchset for Forgejo users. I am unfortunately not a JavaScipt / TypeScript programmer so I can't really take care of the code, but I would volunteer to test it. There is also a [rebased version](https://github.com/Arniiiii/checkout_path_fix) from 2024. Thank you & best regards Julian PS: When using Tailscale (as I am) to my knowledge it is not trivially possible to host Forgejo under a subdomain, so subpaths are kind of necessary in that case.

aahlenst commented 2026年02月07日 14:50:48 +01:00

GitHub's actions/checkout has multiple problems. This is one of them. Another one is lack of support for Git repositories that use SHA-256. It's also not great that Forgejo Actions is reliant on GitHub for such an essential functionality.

I don't think it makes sense to soft-fork GitHub actions/checkout. Soft-forking does not seem to work well in case of actions/upload-artifact and actions/download-artifact.

It would be great if Forgejo Runner could perform essential operations like checkout and upload/download artifact without an action. An interesting option to explore would be whether Forgejo Runner could inject pre-built binaries into a job, similarly to what Docker/Podman does with --init. That could also solve some problems mentioned in #422 (somebody posted their alternative to actions/checkout in there). I'd also welcome if a group of people decided to build and maintain a completely new action that replaces GitHub's actions/checkout.

GitHub's actions/checkout has multiple problems. This is one of them. Another one is lack of support for Git repositories that use SHA-256. It's also not great that Forgejo Actions is reliant on GitHub for such an essential functionality. I don't think it makes sense to soft-fork GitHub actions/checkout. Soft-forking does not seem to work well [in case of actions/upload-artifact and actions/download-artifact](https://code.forgejo.org/forgejo/forgejo-actions-feature-requests/issues/85). It would be great if Forgejo Runner could perform essential operations like checkout and upload/download artifact without an action. An interesting option to explore would be whether Forgejo Runner could inject pre-built binaries into a job, similarly to what Docker/Podman does with `--init`. That could also solve some problems mentioned in https://codeberg.org/forgejo/discussions/issues/422 (somebody posted their alternative to actions/checkout in there). I'd also welcome if a group of people decided to build and maintain a completely new action that replaces GitHub's actions/checkout.