Create and synchronize a PR in a dependent repository

Description

If repository A depends on repository B, cascadinging-pr can be used by a workflow in repository B to trigger the CI on repository A and verify it passes when it will upgrade with the proposed change from repository B.

When used in a workflow triggered by a PR event in origin-repo, cascading-pr will create, update and close a matching PR in another repository (destination-repo). When the PR is updated, cascading-pr subsequently will update the matching PR. The worfklows in origin-repo will wait for the workflow in destination-repo to complete. If the workflow in destination-repo fails, the workflow in origin-repo will also fail.

As an example, when a PR is created in forgejo/runner, a matching PR is created in actions/setup-forgejo with the proposed change. cascading-pr will wait until the CI in actions/setup-forgejo is successful.

The update script is expected to be found in the origin repository running the PR. It is given four arguments:

• A directory in which the destination repository (or a fork) is checked-out on the base branch
• A file with the JSON describing the pull request in the destination repository
• A directory in which the origin repository is checked-out on the head branch
• A file with the JSON describing the pull request in the origin repository

If changes are found in the destination repository directory after the update script runs, they will be pushed as a new commit in the PR.

origin-token is used when accessing origin-repo and needs the read:user, read:repository and write:issue scopes.

destination-token is used to push the branch that contains an update to destination-repo and to open a pull request. It needs the read:user, write:repository and write:issue scopes.

It is recommended that a dedicated user is used to create destination-token and that destination-fork-repo is always used unless the users who are able to create pull requests are trusted.

When the PR is from a forked repository, the update script is run from the default branch of the base repository instead of the head branch of the fork. The pull request author must not be trusted and it is imperative that the update script never runs anything found in the head branch of the pull request.

If the fork of the destination repository is specified and it does not exist, it is created.

Inputs

parameter	description	required	default
origin-url	URL of the Forgejo instance where the PR that triggers the action is located (e.g. https://code.forgejo.org)	true
origin-repo	the repository in which the PR was created	true
origin-token	a token with write permission on origin-repo	true
origin-pr	number of the PR in {orign-repo}	true
destination-url	URL of the Forgejo instance where the cascading PR is created or updated (e.g. https://code.forgejo.org)	true
destination-repo	the repository in which the cascading PR is created or updated	true
destination-fork-repo	the fork of {destination-repo} in which the {destination-branch} will be created or updated	false
destination-branch	the base branch of the destination repository for the cascading PR	true
destination-token	a token with write permission on destination-repo	true
update	path to the script to update the content of the cascading PR	true
prefix	prefix of the cascading PR created on destination-repo (default to {origin-repo})	false
close-merge	if true the cascading PR will be closed and the branch deleted when the PR is merged	false	false
verbose	if true print verbose information	false	false
debug	if true print debug information	false	false

Forgejo dependencies

The Forgejo repositories that depend on each other are linked with workflows using cascading-pr as follows.

flowchart TD
    lxc-helper(lxc-helper) --> act(act)
    act --> runner(Forgejo runner)
    runner --> setup-forgejo(setup-forgejo)
    setup-forgejo --> e2e(end-to-end)
    forgejo-curl(forgejo-curl.sh) --> setup-forgejo
    forgejo(forgejo) --> e2e

    click lxc-helper "https://code.forgejo.org/forgejo/lxc-helpers/src/branch/main/.forgejo/workflows/cascade-act.yml"
    click act "https://code.forgejo.org/forgejo/act/src/branch/main/.forgejo/workflows/cascade-runner.yml"
    click runner "https://code.forgejo.org/forgejo/runner/src/branch/main/.forgejo/workflows/cascade-setup-forgejo.yml"
    click setup-forgejo "https://code.forgejo.org/actions/setup-forgejo/src/branch/main/.forgejo/workflows/cascade-end-to-end.yml"
    click e2e "https://code.forgejo.org/actions/end-to-end"
    click forgejo-curl "https://code.forgejo.org/forgejo/forgejo-curl/src/branch/main/.forgejo/workflows/cascade-setup-forgejo.yml"
    click forgejo "https://codeberg.org/forgejo/forgejo/src/branch/forgejo/.forgejo/workflows/cascade-setup-end-to-end.yml"

Example workflow

on:
  pull_request:
    types:
      - opened
      - synchronize
      - closed

jobs:
  test:
    runs-on: docker
    steps:
      - uses: actions/checkout@v4
      - uses: actions/cascading-pr@v1
        with:
          origin-url: https://code.forgejo.org
          origin-repo: forgejo/lxc-helpers
          origin-token: ${{ secrets.ORIGIN_TOKEN }}
          origin-pr: ${{ github.event.pull_request.number }}
          destination-url: https://code.forgejo.org
          destination-repo: forgejo/act
          destination-branch: main
          destination-token: ${{ secrets.DESTINATION_TOKEN }}
          update: ./upgrade-lxc-helpers

Pull requests from forked repositories

When cascading-pr runs as a consequence of pull request from a forked repository, the workflow must be triggered by a pull_request_target event otherwise it will not have access to secrets.

Prevent privilege escalation

When cascading-pr runs as a consequence of a pull request from a repository forked from orgin-repo, it should create a pull request from a forked repository of destination-repo by specifying the destination-fork-repo.

If the destination-fork-repo repository does not exist, it will be created as a fork of the destination-repo repository, using destination-token.

Hacking

The test environment consists of the following (all users password is admin1234)

• A forgejo instance with a runner
• An unprivileged user user1
• The repository user1/originrepo
  • contains a pull_request workflow using cascading-pr that targets user2/destinationrepo
  • contains a script that will modify user2/destinationrepo
  • a branch1 at the same commit as main
• The repository user1/cascading-pr with the action under test
• An unprivileged user user2
• The repository user2/destinationrepo

git clone https://code.forgejo.org/actions/setup-forgejo
export PATH=$(pwd)/setup-forgejo:$PATH
git clone https://code.forgejo.org/actions/cascading-pr
cd cascading-pr
forgejo-curl.sh logout
forgejo-runner.sh teardown
forgejo.sh teardown
forgejo.sh setup root admin1234 codeberg.org/forgejo/forgejo 1.21
FORGEJO_RUNNER_CONFIG=$(pwd)/tests/runner-config.yaml forgejo-runner.sh setup
url=http://$(cat forgejo-ip):3000
firefox $url

The test for a successful run of the cascading-pr action consists of:

• creating a PR from branch1 to main
• wait for the commit status until it is successful

testing an update on the action

• run tests/run.sh --debug once so all is in place
• commit changes to the files that are in the cascading-pr action (action.yml, cascading-pr.sh etc.)
• push the modified action to user1/cascading-pr
• visit $url/user1/originrepo/actions/runs/1 and click re-run

interactive debugging

Following the steps below recreate the same environment as the integration workflow locally. It is helpful for forensic analysis when something does not run as expected and the error displayed are unclear.

To help with the development loop all steps are idempotent and running tests/run.sh --debug multiple times must succeed.

Individual steps can be run independendely by using the name of the function. For instance:

• tests/run.sh --debug create_pull_request will only call the create_pull_request function found in tests/run.sh to (re)create the pull request in user1/originrepo.
• ./cascading-pr.sh --debug --origin-url ... upsert_branch will only call the upsert_branch function found in cascading-pr.sh.

directories

The tests/run.sh script stores all its files in /tmp/cascading-pr-test. The temporary directories created by cascading-pr.sh are disposed of when the script ends.

logging

If --debug is used a full debug log is displayed, very complete and very verbose. Otherwise it is stashed in a temporary file and only displayed if an error happens.

snippets for copy/pasting

tests/run.sh --debug
tests/run.sh --debug no_change_no_cascade_pr
./cascading-pr.sh --debug --origin-url "$url" --origin-repo "user1/originrepo" --origin-token "$(cat /tmp/cascading-pr-test/user1/repo-token)" --origin-pr 1 --destination-url "$url" --destination-repo "user2/destinationrepo" --destination-token "$(cat /tmp/cascading-pr-test/user2/repo-token)" --destination-branch "main" --update "upgraded" run

Update the README

With https://github.com/npalm/action-docs action-docs --update-readme