Devcontainers plugin: Plugin is not able to process GitLab URLs correctly #117

Closed Parkreiner closed 4 months ago

Issue comes from a user at a prospective company.

Problem

When using the devcontainers plugin, the plugin is sometimes unable to parse GitLab repo URLs, and detect whether the repo has a devcontainers plugin

Likely issue

The issue seems to be that the target field in the LocationSpec value includes additional paths at the end that make the URL not valid. The URL becomes valid when those additional pieces are removed

Other notes

Brandon provided a sample catalog-info file that isn't working (though he obviously can't share the full repo):

apiVersion: backstage.io/v1alpha1
kind: Component
metadata:
  namespace: default
  annotations:
    backstage.io/managed-by-location: url:https://my.gitlab.com/devexportal/sandbox/myreactapp/-/blob/main/catalog-info.yaml
    backstage.io/managed-by-origin-location: url:https://my.gitlab.com/devexportal/sandbox/myreactapp/-/blob/main/catalog-info.yaml
    backstage.io/view-url: https://my.gitlab.com/devexportal/sandbox/myreactapp/-/blob/main/catalog-info.yaml
    backstage.io/edit-url: https://my.gitlab.com/devexportal/sandbox/myreactapp/-/edit/main/catalog-info.yaml
    backstage.io/source-location: url:https://my.gitlab.com/devexportal/sandbox/myreactapp/-/blob/main/
    swf-backstage-dev.com/template-source-url: https://my.gitlab.com/devexportal/templates/simple-react-app/-/blob/0.5.0/
    swf-backstage-dev.com/template-source-name: simple-react-app
    gitlab.com/project-slug: devexportal/sandbox/myreactapp
    backstage.io/techdocs-ref: dir:.
    argocd/app-name: myreactapp-main
    argocd/proxy-url: /argocd/api
    backstage.io/kubernetes-namespace: myreactapp
    backstage.io/kubernetes-label-selector: app.kubernetes.io/component=myreactapp
    gitlab.com/instance: my.gitlab.com
  title: myReactApp
  name: myreactapp
  description: Creating React App from Template
  tags:
    - helm
    - kubernetes
    - react
    - typescript
    - vite
    - devcontainers
  links:
    - url: https://my.gitlab.com/devexportal/templates/simple-react-app/-/blob/0.5.0/
      title: Template Source Url
      type: templates
  uid: e375af26-1e87-44aa-b02f-78498e879628
  etag: d26e609289dbb3d09c7550f5c57d17c3f45a9c3e
relations:
  - type: ownedBy
    targetRef: group:default/gitlab-users
    target:
      kind: group
      namespace: default
      name: gitlab-users
  - type: scaffoldedFrom
    targetRef: template:default/simple-react-app
    target:
      kind: template
      namespace: default
      name: simple-react-app
spec:
  type: website
  lifecycle: experimental
  owner: group:default/my-group
  dependsOn: []
  consumesApis: []
  coder:
    templateName: docker
    mode: auto
    params:
      repo: custom
      region: us-east-1
  createdDate: 2024-04-17T17:08:58.204Z

The culprit might be the fact that the backstage.io/source-location property has a value of url:https://my.gitlab.com/devexportal/sandbox/myreactapp/-/blob/main/, and that the /~/blob/main/ at the end shouldn't be there?

We aren't working with this property directly, but it might be shared with the APIs we are using

This reminds me of the work we are doing on the Coder git-clone module, we have to parse git URLs there as well.

My idea was to have a set of matchers we can use for figuring out the base repository URL and branch, so it could look something like this:

github = [ "{owner}/{repo}/tree/{branch}" "{owner}/{repo}/blob/{branch}" ]
gitlab = [ "{owner}/{repo}/-/tree/{branch}" "{owner}/{repo}/-/blob/{branch}"  ]
bitbucket = [
  "{owner}/{repo}/src/{ignore}?at={branch}"
  "{owner}/{repo}/src/{branch}"
]

Each matcher is tried in order, this syntax is inspired by go-chi but we could do :var if we wanted it Express-like. Or we could just directly write them as regular expressions.

Or we use something like https://github.com/IonicaBizau/git-url-parse

coder / backstage-plugins