git fetch --all fails on forked repos with existing local state in v0.37.1

[therealdwright]

Community Note

• Please vote on this issue by adding a 👍 reaction to the original issue to help the community and maintainers prioritize this request. Searching for pre-existing feature requests helps us consolidate datapoints for identical requirements into a single place, thank you!
• Please do not leave "+1" or other comments that do not add relevant new information or questions, they generate extra noise for issue followers and do not help prioritize the request.
• If you are interested in working on this issue or have submitted a pull request, please leave a comment.

---

Overview of the Issue

After upgrading to Atlantis v0.37.1, git fetch --all fails when processing subsequent pushes to forked repositories that already have local state on disk. The error indicates the head repository (fork) cannot be found, even though it exists and was successfully cloned on the initial plan.

This appears to be a regression introduced in v0.37.1, as the same workflow functioned correctly in previous versions.

Reproduction Steps

1. Create a PR from a forked repository to the base repository
2. Run atlantis plan - this succeeds and clones the repository
3. Force-push an update to the same PR branch
4. Atlantis attempts to run git fetch --all on existing local state
5. Fetch fails with "repository not found" error for the fork remote

Logs

Logs

``` running git fetch --all: Fetching origin Fetching head remote: Repository not found. fatal: repository 'https://github.com/BitGo/infra.git/' not found error: could not fetch head : exit status 1 ```

Environment details

• Atlantis version: v0.37.1
• Deployment method: GitHub App
• If not running the latest Atlantis version have you tried to reproduce this issue on the latest version:
  Atlantis server-side config file:

Additional Context

• Issue only occurs on subsequent pushes to the same PR, not on initial plan
• Issue only affects forked repositories
• The fork repository exists and is accessible (verified via browser and git CLI)
• Likely related to how v0.37.1 handles remote configuration for forks with existing local state

[dosubot]

This looks like a regression in how Atlantis v0.37.1 manages git remotes for forked PRs with existing local state, especially when using the GitHub App integration. Atlantis adds the fork as a remote named source and updates both origin and source before running git fetch --all. If the source remote is missing or its URL is incorrect (for example, due to token or credential handling), fetch will fail with "repository not found" even if the repo exists and was cloned initially. The code tries to reset remote URLs before fetching to handle credential changes, but the workaround for GitHub App tokens is described as brittle and could be the root cause here. You might be able to work around this by deleting the workspace directory for the PR so Atlantis reclones everything, or by verifying that the remote URLs are correct before the fetch step. This issue appears to be specific to v0.37.1 and isn't documented in existing issues or recent pull requests. For more technical details, see the relevant code handling remotes and fetches here and the GitHub App token workaround here.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}