config.yml

# Configuration file for https://circleci.com/gh/angular/angular

# Note: YAML anchors allow an object to be re-used, reducing duplication.
# The ampersand declares an alias for an object, then later the `<<: *name`
# syntax dereferences it.
# See http://blog.daemonl.com/2016/02/yaml.html
# To validate changes, use an online parser, eg.
# http://yaml-online-parser.appspot.com/

# CircleCI configuration version
# Version 2.1 allows for extra config reuse features
# https://circleci.com/docs/2.0/reusing-config/#getting-started-with-config-reuse
version: 2.1

# We don't want to include the current branch name in the cache key because that would prevent
# PRs from being able to restore the cache since the branch names are always different for PRs.
# The cache key should only consist of dynamic values that change whenever something in the
# cache changes. For example:
# 1) yarn lock file changes --> cached "node_modules" are different.
# 2) bazel repository definitions change --> cached bazel repositories are different.
# Windows needs its own cache key because binaries in node_modules are different.
# **NOTE 1 **: If you change the cache key prefix, also sync the cache_key_fallback to match.
# **NOTE 2 **: Keep the static part of the cache key as prefix to enable correct fallbacks.
# See https://circleci.com/docs/2.0/caching/#restoring-cache for how prefixes work in CircleCI.
var_3: &cache_key v3-angular-node-10.16-{{ checksum "yarn.lock" }}-{{ checksum "WORKSPACE" }}-{{ checksum "packages/bazel/package.bzl" }}-{{ checksum "aio/yarn.lock" }}
var_4: &cache_key_fallback v3-angular-node-10.16-
var_3_win: &cache_key_win v5-angular-win-node-12.0-{{ checksum "yarn.lock" }}-{{ checksum "WORKSPACE" }}-{{ checksum "packages/bazel/package.bzl" }}-{{ checksum "aio/yarn.lock" }}
var_4_win: &cache_key_win_fallback v5-angular-win-node-12.0-

# Cache key for the Material unit tests job. **Note** when updating the SHA in the cache keys,
# also update the SHA for the "MATERIAL_REPO_COMMIT" environment variable.
var_5: &material_unit_tests_cache_key v4-angular-material-818a8c7d6bb3f25d0637e5d6462cf9a84403b78f
var_6: &material_unit_tests_cache_key_fallback v4-angular-material-

# Workspace initially persisted by the `setup` job, and then enhanced by `build-npm-packages` and
# `build-ivy-npm-packages`.
# https://circleci.com/docs/2.0/workflows/#using-workspaces-to-share-data-among-jobs
# https://circleci.com/blog/deep-diving-into-circleci-workspaces/
var_7: &workspace_location ~/

# Filter to run a job on builds for pull requests only.
var_8: &only_on_pull_requests
  filters:
    branches:
      only:
        - /pull\/\d+/

# Filter to skip a job on builds for pull requests.
var_9: &skip_on_pull_requests
  filters:
    branches:
      ignore:
        - /pull\/\d+/

# Filter to run a job on builds for the master branch only.
var_10: &only_on_master
  filters:
    branches:
      only:
        - master

# Executor Definitions
# https://circleci.com/docs/2.0/reusing-config/#authoring-reusable-executors
# **NOTE 1**: Pin to exact images using an ID (SHA). See https://circleci.com/docs/2.0/circleci-images/#using-a-docker-image-id-to-pin-an-image-to-a-fixed-version.
#             (Using the tag in not necessary when pinning by ID, but include it anyway for documentation purposes.)
# **NOTE 2**: If you change the version of the docker images, also change the `cache_key` suffix.
# **NOTE 3**: If you change the version of the `*-browsers` docker image, make sure the
#             `CI_CHROMEDRIVER_VERSION_ARG` env var (in `.circleci/env.sh`) points to a ChromeDriver
#             version that is compatible with the Chrome version in the image.
executors:
  default-executor:
    parameters:
      resource_class:
        type: string
        default: medium
    docker:
      - image: circleci/node:10.16@sha256:75c05084fff4afa3683a03c5a04a4a3ad95c536ff2439d8fe14e7e1f5c58b09a
    resource_class: << parameters.resource_class >>
    working_directory: ~/ng

  browsers-executor:
    parameters:
      resource_class:
        type: string
        default: medium
    docker:
      # The browser docker image comes with Chrome and Firefox preinstalled. This is just
      # needed for jobs that run tests without Bazel. Bazel runs tests with browsers that will be
      # fetched by the Webtesting rules. Therefore for jobs that run tests with Bazel, we don't need a
      # docker image with browsers pre-installed.
      - image: circleci/node:10.16-browsers@sha256:d2a96fe1cbef51257ee626b5f645e64dade3e886f00ba9cb7e8ea65b4efe8db1
    resource_class: << parameters.resource_class >>
    working_directory: ~/ng

  windows-executor:
    working_directory: ~/ng
    resource_class: windows.medium
    # CircleCI windows VMs do have the GitBash shell available:
    # https://github.com/CircleCI-Public/windows-preview-docs#shells
    # But in this specific case we really should not use it because Bazel must not be ran from
    # GitBash. These issues discuss why:
    # https://github.com/bazelbuild/bazel/issues/5751
    # https://github.com/bazelbuild/bazel/issues/5724#issuecomment-410194038
    # https://github.com/bazelbuild/bazel/issues/6339#issuecomment-441600879
    shell: powershell.exe -ExecutionPolicy Bypass
    machine:
      # Windows preview image that includes the following:
      # - Visual Studio 2019 build tools
      # - Node 12
      # - yarn 1.17
      # - Python 3 3.7.4
      image: windows-server-2019-vs2019:201908-02

# Command Definitions
# https://circleci.com/docs/2.0/reusing-config/#authoring-reusable-commands
commands:
  custom_attach_workspace:
    description: Attach workspace at a predefined location
    steps:
      - attach_workspace:
          at: *workspace_location

  # Initializes the CI environment by setting up common environment variables.
  init_environment:
    description: Initializing environment (setting up variables, overwriting Yarn)
    steps:
      - run: ./.circleci/env.sh
      - run:
          # Overwrite the yarn installed in the docker container with our own version.
          name: Overwrite yarn with our own version
          command: |
            ourYarn=$(realpath ./third_party/github.com/yarnpkg/yarn/releases/download/v1.17.3/bin/yarn.js)
            sudo chmod a+x $ourYarn
            sudo ln -fs $ourYarn /usr/local/bin/yarn
      - run: echo "Yarn version $(yarn --version)"
      - run:
          # Configure git as the CircleCI `checkout` command does.
          # This is needed because we only checkout on the setup job.
          # Add GitHub to known hosts
          name: Configure git
          command: |
            mkdir -p ~/.ssh
            echo 'github.com ssh-rsa AAAAB3NzaC1yc2EAAAABIwAAAQEAq2A7hRGmdnm9tUDbO9IDSwBK6TbQa+PXYPCPy6rbTrTtw7PHkccKrpp0yVhp5HdEIcKr6pLlVDBfOLX9QUsyCOV0wzfjIJNlGEYsdlLJizHhbn2mUjvSAHQqZETYP81eFzLQNnPHt4EVVUh7VfDESU84KezmD5QlWpXLmvU31/yMf+Se8xhHTvKSCZIFImWwoG6mbUoWf9nzpIoaSjB+weqqUUmpaaasXVal72J+UX2B+2RPW3RcT0eOzQgqlJL3RKrTJvdsjE3JEAvGq3lGHSZXy28G3skua2SmVi/w4yCE6gbODqnTWlg7+wC604ydGXA8VJiS5ap43JXiUFFAaQ==' >> ~/.ssh/known_hosts
            git config --global url."ssh://git@github.com".insteadOf "https://github.com" || true
            git config --global gc.auto 0 || true

  # Normally this would be an individual job instead of a command.
  # But startup and setup time for each invidual windows job are high enough to discourage
  # many small jobs, so instead we use a command for setup unless the gain becomes significant.
  setup_win:
    description: Setup windows node environment
    steps:
      # Use the Linux workspace directly, as it already has checkout, rebased and node modules.
      - custom_attach_workspace
      # Install Bazel pre-requisites that aren't in the preconfigured CircleCI Windows VM.
      - run: ./.circleci/windows-env.ps1
      - run: node --version
      - run: yarn --version
      - restore_cache:
          keys:
            - *cache_key_win
            - *cache_key_win_fallback
      # Reinstall to get windows binaries.
      - run: yarn install --frozen-lockfile --non-interactive
      - setup_circleci_bazel_config_win
      # Install @bazel/bazel globally and use that for the first run.
      # Workaround for https://github.com/bazelbuild/rules_nodejs/issues/894
      - run: yarn global add @bazel/bazel@$env:BAZEL_VERSION
      - run: bazel info

  setup_circleci_bazel_config:
    description: Set up CircleCI bazel configuration on Linux
    steps:
      - run: sudo cp .circleci/bazel.linux.rc $HOME/.bazelrc

  setup_circleci_bazel_config_win:
    description: Set up CircleCI bazel configuration on Windows
    steps:
      - run: copy .circleci\bazel.windows.rc $env:USERPROFILE\.bazelrc

  setup_bazel_rbe:
    description: Setup bazel RBE remote execution
    steps:
      # We need ensure that the same default digest is used for encoding and decoding
      # with openssl. Openssl versions might have different default digests which can
      # cause decryption failures based on the openssl version. https://stackoverflow.com/a/39641378/4317734
      - run: openssl aes-256-cbc -d -in .circleci/gcp_token -md md5 -k "$CI_REPO_NAME" -out /home/circleci/.gcp_credentials
      - run: echo "export GOOGLE_APPLICATION_CREDENTIALS=/home/circleci/.gcp_credentials" >> $BASH_ENV
      - run: ./.circleci/setup-rbe.sh .bazelrc.user

  notify_webhook_on_fail:
    description: Notify a webhook about failure
    parameters:
      # `webhook_url_env_var` are secret env vars defined in CircleCI project settings.
      # The URLs come from https://angular-team.slack.com/apps/A0F7VRE7N-circleci.
      webhook_url_env_var:
        type: env_var_name
    steps:
      - run:
          when: on_fail
          command: |
            notificationJson="{\"text\":\":x: \`$CIRCLE_JOB\` job for $CIRCLE_BRANCH branch failed on build $CIRCLE_BUILD_NUM: $CIRCLE_BUILD_URL :scream:\"}"
            curl --request POST --header "Content-Type: application/json" --data "$notificationJson" ${<< parameters.webhook_url_env_var >>}

# Job definitions
# Jobs can include parameters that are passed in the workflow job invocation.
# https://circleci.com/docs/2.0/reusing-config/#authoring-parameterized-jobs
jobs:
  setup:
    executor: default-executor
    steps:
      - checkout
      - run:
          name: Rebase PR on target branch
          # After checkout, rebase on top of target branch.
          command: >
            if [[ -n "${CIRCLE_PR_NUMBER}" ]]; then
              # User is required for rebase.
              git config user.name "angular-ci"
              git config user.email "angular-ci"
              # Rebase PR on top of target branch.
              node tools/rebase-pr.js angular/angular ${CIRCLE_PR_NUMBER}
            else
              echo "This build is not over a PR, nothing to do."
            fi
      # This cache is saved in the build-npm-packages so that Bazel cache is also included.
      - restore_cache:
          keys:
            - *cache_key
            - *cache_key_fallback
      - init_environment
      - run:
          name: Running Yarn install
          command: yarn install --frozen-lockfile --non-interactive
          # Yarn's requests sometimes take more than 10mins to complete.
          no_output_timeout: 45m
      - run: yarn --cwd aio install --frozen-lockfile --non-interactive
      # Make the bazel directories and add a file to them if they don't exist already so that
      # persist_to_workspace does not fail.
      - run: |
          if [ ! -d ~/bazel_repository_cache ]; then
            mkdir ~/bazel_repository_cache
            touch ~/bazel_repository_cache/MARKER
          fi
      # Persist any changes at this point to be reused by further jobs.
      # **NOTE**: To add new content to the workspace, always persist on the same root.
      - persist_to_workspace:
          root: *workspace_location
          paths:
            - ./ng
            - ./bazel_repository_cache

  lint:
    executor: default-executor
    steps:
      - custom_attach_workspace
      - init_environment

      - run: 'yarn bazel:format -mode=check ||
              (echo "BUILD files not formatted. Please run ''yarn bazel:format''" ; exit 1)'
      # Run the skylark linter to check our Bazel rules
      - run: 'yarn bazel:lint ||
              (echo -e "\n.bzl files have lint errors. Please run ''yarn bazel:lint-fix''"; exit 1)'

      - run: yarn gulp lint
      - run: node tools/verify-codeownership

  test:
    executor:
      name: default-executor
      resource_class: xlarge
    steps:
      - custom_attach_workspace
      - init_environment
      - setup_circleci_bazel_config
      # Setup remote execution and run RBE-compatible tests.
      - setup_bazel_rbe
      - run:
          command: yarn bazel test //... --build_tag_filters=-ivy-only --test_tag_filters=-ivy-only
          no_output_timeout: 20m

  # Temporary job to test what will happen when we flip the Ivy flag to true
  test_ivy_aot:
    executor:
      name: default-executor
      resource_class: xlarge
    steps:
      - custom_attach_workspace
      - init_environment
      - setup_circleci_bazel_config
      - setup_bazel_rbe

        # We need to explicitly specify the --symlink_prefix option because otherwise we would
        # not be able to easily find the output bin directory when uploading artifacts for size
        # measurements.
      - run:
          command: yarn test-ivy-aot //... --symlink_prefix=dist/
          no_output_timeout: 20m

        # Publish bundle artifacts which will be used to calculate the size change. **Note**: Make
        # sure that the size plugin from the Angular robot fetches the artifacts from this CircleCI
        # job (see .github/angular-robot.yml). Additionally any artifacts need to be stored with the
        # following path format: "{projectName}/{context}/{fileName}". This format is necessary
        # because otherwise the bot is not able to pick up the artifacts from CircleCI. See:
        # https://github.com/angular/github-robot/blob/master/functions/src/plugins/size.ts#L392-L394
      - store_artifacts:
          path: dist/bin/packages/core/test/bundling/hello_world/bundle.min.js
          destination: core/hello_world/bundle
      - store_artifacts:
          path: dist/bin/packages/core/test/bundling/todo/bundle.min.js
          destination: core/todo/bundle
      - store_artifacts:
          path: dist/bin/packages/core/test/bundling/hello_world/bundle.min.js.br
          destination: core/hello_world/bundle.br
      - store_artifacts:
          path: dist/bin/packages/core/test/bundling/todo/bundle.min.js.br
          destination: core/todo/bundle.br

  # This job is currently a PoC for running tests on SauceLabs via bazel. It runs a subset of the
  # tests in `legacy-unit-tests-saucelabs` (see
  # [BUILD.bazel](https://github.com/angular/angular/blob/ef44f51d5/BUILD.bazel#L66-L92)).
  #
  # NOTE: This is currently limited to master builds only. See the `default_workflow` configuration.
  test_saucelabs_bazel:
    executor:
      name: default-executor
      # In order to avoid the bottleneck of having a slow host machine, we acquire a better
      # container for this job. This is necessary because we launch a lot of browsers concurrently
      # and therefore the tunnel and Karma need to process a lot of file requests and tests.
      resource_class: xlarge
    steps:
      - custom_attach_workspace
      - init_environment
      - setup_circleci_bazel_config
      - setup_bazel_rbe
      - run:
          name: Run Bazel tests in saucelabs
          # All web tests are contained within a single //:test_web_all target for Saucelabs
          # as running each set of tests as a separate target will attempt to acquire too
          # many browsers on Saucelabs (7 per target currently) and some tests will always
          # fail to acquire browsers. For example:
          # 14 02 2019 19:52:33.170:WARN [launcher]: chrome beta on SauceLabs have not captured in 180000 ms, killing.
          # //packages/forms/test:web_test_sauce TIMEOUT in 315.0s
          command:  |
            ./scripts/saucelabs/run-bazel-via-tunnel.sh \
              --tunnel-id angular-${CIRCLE_BUILD_NUM}-${CIRCLE_NODE_INDEX} \
              --username $SAUCE_USERNAME \
              --key $(echo $SAUCE_ACCESS_KEY | rev) \
              yarn bazel test //:test_web_all
          no_output_timeout: 20m
      - notify_webhook_on_fail:
          webhook_url_env_var: SLACK_DEV_INFRA_CI_FAILURES_WEBHOOK_URL

  test_aio:
    # Needed because the AIO tests and the PWA score test depend on Chrome being available.
    executor: browsers-executor
    steps:
      - custom_attach_workspace
      - init_environment
        # Compile dependencies to ivy
        # Running `ngcc` here (instead of implicitly via `ng build`) allows us to take advantage of
        # the parallel, async mode speed-up (~20-25s on CI).
      - run: yarn --cwd aio ivy-ngcc --properties es2015
        # Build aio
      - run: yarn --cwd aio build --progress=false
        # Lint the code
      - run: yarn --cwd aio lint
        # Run unit tests
      - run: yarn --cwd aio test --progress=false --watch=false
        # Run e2e tests
      - run: yarn --cwd aio e2e --configuration=ci
        # Run PWA-score tests
      - run: yarn --cwd aio test-pwa-score-localhost $CI_AIO_MIN_PWA_SCORE
        # Run accessibility tests
      - run: yarn --cwd aio test-a11y-score-localhost
        # Check the bundle sizes.
      - run: yarn --cwd aio payload-size
        # Run unit tests for Firebase redirects
      - run: yarn --cwd aio redirects-test

  deploy_aio:
    # Needed because before deploying the deploy-production script runs the PWA score tests.
    executor: browsers-executor
    steps:
      - custom_attach_workspace
      - init_environment
        # Deploy angular.io to production (if necessary)
      - run: setPublicVar_CI_STABLE_BRANCH
      - run: yarn --cwd aio deploy-production

  test_aio_local:
    parameters:
      viewengine:
        type: boolean
        default: false
    # Needed because the AIO tests and the PWA score test depend on Chrome being available.
    executor: browsers-executor
    steps:
      - custom_attach_workspace
      - init_environment
        # Build aio (with local Angular packages)
      - run: yarn --cwd aio build-local<<# parameters.viewengine >>-with-viewengine<</ parameters.viewengine >>-ci
        # Run unit tests
      - run: yarn --cwd aio test --progress=false --watch=false
        # Run e2e tests
      - run: yarn --cwd aio e2e --configuration=ci
        # Run PWA-score tests
      - run: yarn --cwd aio test-pwa-score-localhost $CI_AIO_MIN_PWA_SCORE
        # Check the bundle sizes.
      - run: yarn --cwd aio payload-size aio-local<<# parameters.viewengine >>-viewengine<</ parameters.viewengine >>

  test_aio_tools:
    executor: default-executor
    steps:
      - custom_attach_workspace
      - init_environment
        # Install
      - run: yarn --cwd aio install --frozen-lockfile --non-interactive
      - run: yarn --cwd aio extract-cli-command-docs
        # Run tools tests
      - run: yarn --cwd aio tools-test
      - run: ./aio/aio-builds-setup/scripts/test.sh

  test_docs_examples:
    parameters:
      ivy:
        type: boolean
        default: false
    executor:
      # Needed because the example e2e tests depend on Chrome.
      name: browsers-executor
      resource_class: xlarge
    parallelism: 5
    steps:
      - custom_attach_workspace
      - init_environment
        # Install aio
      - run: yarn --cwd aio install --frozen-lockfile --non-interactive
      - when:
          condition: << parameters.ivy >>
          steps:
              # Rename the Ivy packages dist folder to "dist/packages-dist" as the AIO
              # package installer picks up the locally built packages from that location.
              # *Note*: We could also adjust the packages installer, but given we won't have
              # two different folders of Angular distributions in the future, we should keep
              # the packages installer unchanged.
            - run: mv dist/packages-dist-ivy-aot dist/packages-dist
        # Run examples tests. The "CIRCLE_NODE_INDEX" will be set if "parallelism" is enabled.
        # Since the parallelism is set to "5", there will be five parallel CircleCI containers.
        # with either "0", "1", etc as node index. This can be passed to the "--shard" argument.
      - run: yarn --cwd aio example-e2e --setup --local <<# parameters.ivy >>--ivy<</ parameters.ivy >> --cliSpecsConcurrency=5 --shard=${CIRCLE_NODE_INDEX}/${CIRCLE_NODE_TOTAL} --retry 2

  # This job should only be run on PR builds, where `CI_PULL_REQUEST` is not `false`.
  aio_preview:
    executor: default-executor
    environment:
       AIO_SNAPSHOT_ARTIFACT_PATH: &aio_preview_artifact_path 'aio/tmp/snapshot.tgz'
    steps:
      - custom_attach_workspace
      - init_environment
      - run: ./aio/scripts/build-artifacts.sh $AIO_SNAPSHOT_ARTIFACT_PATH $CI_PULL_REQUEST $CI_COMMIT
      - store_artifacts:
          path: *aio_preview_artifact_path
          # The `destination` needs to be kept in synch with the value of
          # `AIO_ARTIFACT_PATH` in `aio/aio-builds-setup/Dockerfile`
          destination: aio/dist/aio-snapshot.tgz
      - run: node ./aio/scripts/create-preview $CIRCLE_BUILD_NUM

  # This job should only be run on PR builds, where `CI_PULL_REQUEST` is not `false`.
  test_aio_preview:
    # Needed because the test-preview script runs e2e tests and the PWA score test with Chrome.
    executor: browsers-executor
    steps:
      - custom_attach_workspace
      - init_environment
      - run: yarn --cwd aio install --frozen-lockfile --non-interactive
      - run:
          name: Wait for preview and run tests
          command: node aio/scripts/test-preview.js $CI_PULL_REQUEST $CI_COMMIT $CI_AIO_MIN_PWA_SCORE


  # The `build-npm-packages` tasks exist for backwards-compatibility with old scripts and
  # tests that rely on the pre-Bazel `dist/packages-dist` output structure (build.sh).
  # Having multiple jobs that independently build in this manner duplicates some work; we build
  # the bazel packages more than once. Even though we have a remote cache, these jobs will
  # typically run in parallel so up-to-date outputs will not be available at the time the build
  # starts.

  # Build the view engine npm packages. No new jobs should depend on this.
  build-npm-packages:
    executor:
      name: default-executor
      resource_class: xlarge
    steps:
      - custom_attach_workspace
      - init_environment
      - setup_circleci_bazel_config
      - setup_bazel_rbe

      - run: scripts/build-packages-dist.sh

      # Save the npm packages from //packages/... for other workflow jobs to read
      - persist_to_workspace:
          root: *workspace_location
          paths:
            - ng/dist/packages-dist

      # Save dependencies and bazel repository cache to use on subsequent runs.
      - save_cache:
          key: *cache_key
          paths:
            - "node_modules"
            - "aio/node_modules"
            - "~/bazel_repository_cache"

  # Build the ivy npm packages.
  build-ivy-npm-packages:
    executor:
      name: default-executor
      resource_class: xlarge
    steps:
      - custom_attach_workspace
      - init_environment
      - setup_circleci_bazel_config
      - setup_bazel_rbe

      - run: scripts/build-ivy-npm-packages.sh

      # Save the npm packages from //packages/... for other workflow jobs to read
      - persist_to_workspace:
          root: *workspace_location
          paths:
            - ng/dist/packages-dist-ivy-aot

  # We run the integration tests outside of Bazel for now.
  # They are a separate workflow job so that they can be easily re-run.
  # When the tests are ported to bazel test targets, they should move to the "test"
  # job above, as part of the bazel test command. That has flaky_test_attempts so the
  # need to re-run manually should be alleviated.
  # See comments inside the integration/run_tests.sh script.
  integration_test:
    executor:
      # Needed because the integration tests expect Chrome to be installed (e.g cli-hello-world)
      name: browsers-executor
      # Note: we run Bazel in one of the integration tests, and it can consume >2G
      # of memory. Together with the system under test, this can exhaust the RAM
      # on a 4G worker so we use a larger machine here too.
      resource_class: xlarge
    parallelism: 4
    steps:
      - custom_attach_workspace
      - init_environment
      # Runs the integration tests in parallel across multiple CircleCI container instances. The
      # amount of container nodes for this job is controlled by the "parallelism" option.
      - run: ./integration/run_tests.sh ${CIRCLE_NODE_INDEX} ${CIRCLE_NODE_TOTAL}

  # This job creates compressed tarballs (`.tgz` files) for all Angular packages and stores them as
  # build artifacts. This makes it easy to try out changes from a PR build for testing purposes.
  # More info CircleCI build artifacts: https://circleci.com/docs/2.0/artifacts
  #
  # NOTE: Currently, this job only runs for PR builds. See `publish_snapshot` for non-PR builds.
  publish_packages_as_artifacts:
    executor: default-executor
    environment:
      NG_PACKAGES_DIR: &ng_packages_dir 'dist/packages-dist'
      NG_PACKAGES_ARCHIVES_DIR: &ng_packages_archives_dir 'dist/packages-dist-archives'
    steps:
      - custom_attach_workspace
      - init_environment
      - run:
          name: Create artifacts
          command: ./scripts/ci/create-package-archives.sh $CI_PULL_REQUEST $CI_COMMIT $NG_PACKAGES_DIR $NG_PACKAGES_ARCHIVES_DIR
      - store_artifacts:
          path: *ng_packages_archives_dir
          destination: angular

  # This job updates the content of repos like github.com/angular/core-builds
  # for every green build on angular/angular.
  publish_snapshot:
    executor: default-executor
    steps:
      # See below - ideally this job should not trigger for non-upstream builds.
      # But since it does, we have to check this condition.
      - run:
          name: Skip this job for Pull Requests and Fork builds
          # Note: Using `CIRCLE_*` env variables (instead of those defined in `env.sh` so that this
          #       step can be run before `init_environment`.
          command: >
            if [[ -n "${CIRCLE_PR_NUMBER}" ]] ||
                [[ "$CIRCLE_PROJECT_USERNAME" != "angular" ]] ||
                [[ "$CIRCLE_PROJECT_REPONAME" != "angular" ]]; then
              circleci step halt
            fi
      - custom_attach_workspace
      - init_environment
      # CircleCI has a config setting to force SSH for all github connections
      # This is not compatible with our mechanism of using a Personal Access Token
      # Clear the global setting
      - run: git config --global --unset "url.ssh://git@github.com.insteadof"
      - run:
          name: Decrypt github credentials
          # We need ensure that the same default digest is used for encoding and decoding with
          # openssl. Openssl versions might have different default digests which can cause
          # decryption failures based on the installed openssl version. https://stackoverflow.com/a/39641378/4317734
          command: 'openssl aes-256-cbc -d -in .circleci/github_token -md md5 -k "${KEY}" -out ~/.git_credentials'
      - run: ./scripts/ci/publish-build-artifacts.sh

  aio_monitoring_stable:
    # This job needs Chrome to be globally installed because the tests run with Protractor
    # which does not load the browser through the Bazel webtesting rules.
    executor: browsers-executor
    steps:
      - custom_attach_workspace
      - init_environment
      - run: setPublicVar_CI_STABLE_BRANCH
      - run:
          name: Check out `aio/` from the stable branch
          command: |
            git fetch origin $CI_STABLE_BRANCH
            git checkout --force origin/$CI_STABLE_BRANCH -- aio/
      - run:
          name: Run tests against https://angular.io/
          command: ./aio/scripts/test-production.sh https://angular.io/ $CI_AIO_MIN_PWA_SCORE
      - notify_webhook_on_fail:
          webhook_url_env_var: SLACK_CARETAKER_WEBHOOK_URL
      - notify_webhook_on_fail:
          webhook_url_env_var: SLACK_DEV_INFRA_CI_FAILURES_WEBHOOK_URL

  aio_monitoring_next:
    # This job needs Chrome to be globally installed because the tests run with Protractor
    # which does not load the browser through the Bazel webtesting rules.
    executor: browsers-executor
    steps:
      - custom_attach_workspace
      - init_environment
      - run:
          name: Run tests against https://next.angular.io/
          command: ./aio/scripts/test-production.sh https://next.angular.io/ $CI_AIO_MIN_PWA_SCORE
      - notify_webhook_on_fail:
          webhook_url_env_var: SLACK_CARETAKER_WEBHOOK_URL
      - notify_webhook_on_fail:
          webhook_url_env_var: SLACK_DEV_INFRA_CI_FAILURES_WEBHOOK_URL

  legacy-unit-tests-saucelabs:
    executor:
      name: default-executor
      # In order to avoid the bottleneck of having a slow host machine, we acquire a better
      # container for this job. This is necessary because we launch a lot of browsers concurrently
      # and therefore the tunnel and Karma need to process a lot of file requests and tests.
      resource_class: xlarge
    steps:
      - custom_attach_workspace
      - init_environment
      - run:
          name: Preparing environment for running tests on Saucelabs.
          command: |
            setPublicVar KARMA_JS_BROWSERS $(node -e 'console.log(require("./browser-providers.conf").sauceAliases.CI_REQUIRED.join(","))')
            setSecretVar SAUCE_ACCESS_KEY $(echo $SAUCE_ACCESS_KEY | rev)
      - run:
          name: Starting Saucelabs tunnel
          command: ./scripts/saucelabs/start-tunnel.sh
          background: true
      - run: yarn tsc -p packages
      - run: yarn tsc -p modules
        # Waits for the Saucelabs tunnel to be ready. This ensures that we don't run tests
        # too early without Saucelabs not being ready.
      - run: ./scripts/saucelabs/wait-for-tunnel.sh
      - run: yarn karma start ./karma-js.conf.js --single-run --browsers=${KARMA_JS_BROWSERS}
      - run: ./scripts/saucelabs/stop-tunnel.sh

  legacy-misc-tests:
    executor: default-executor
    steps:
      - custom_attach_workspace
      - init_environment
      - run: yarn gulp check-cycle
      # TODO: disabled because the Bazel packages-dist does not seem to have map files for
      # the ESM5/ES2015 output. See: https://github.com/angular/angular/issues/27966
      # - run: yarn gulp source-map-test

  # Job to run unit tests from angular/components. Needs a browser since all
  # component unit tests assume they're running in the browser environment.
  material-unit-tests:
    executor:
      name: browsers-executor
      resource_class: xlarge
    steps:
      - custom_attach_workspace
      - init_environment
      # Although RBE is configured below for the Material repo, also setup RBE in the Angular repo
      # to provision Angular's GCP token into the environment variables.
      - setup_bazel_rbe
      # Restore the cache before cloning the repository because the clone script re-uses
      # the restored repository if present. This reduces the amount of times the components
      # repository needs to be cloned (this is slow and increases based on commits in the repo).
      - restore_cache:
          keys:
            - *material_unit_tests_cache_key
            - *material_unit_tests_cache_key_fallback
      - run:
          name: "Fetching Material repository"
          command: ./scripts/ci/clone_angular_material_repo.sh
      - run:
          # Run yarn install to fetch the Bazel binaries as used in the Material repo.
          name: Installing Material dependencies.
          command: yarn --cwd ${MATERIAL_REPO_TMP_DIR} install --frozen-lockfile --non-interactive
      - save_cache:
          key: *material_unit_tests_cache_key
          paths:
            # Material directory must be kept in sync with the `$MATERIAL_REPO_TMP_DIR` env variable.
            # It needs to be hardcoded here, because env variables interpolation is not supported.
            - "/tmp/material2"
      - run:
          name: "Setup Bazel RBE remote execution in Material repo"
          command: |
            ./.circleci/setup-rbe.sh "${MATERIAL_REPO_TMP_DIR}/.bazelrc.user"
      - run:
          name: "Running Material unit tests"
          command: ./scripts/ci/run_angular_material_unit_tests.sh

  test_zonejs:
    executor:
      name: default-executor
      resource_class: xlarge
    steps:
      - custom_attach_workspace
      - init_environment
      - setup_circleci_bazel_config
      - setup_bazel_rbe
        # Install
      - run: yarn --cwd packages/zone.js install --frozen-lockfile --non-interactive
        # Run zone.js tools tests
      - run: yarn --cwd packages/zone.js promisetest
      - run: yarn --cwd packages/zone.js promisefinallytest
      - run: yarn bazel build //packages/zone.js:npm_package &&
             cp dist/bin/packages/zone.js/npm_package/dist/zone-mix.js ./packages/zone.js/test/extra/ &&
             cp dist/bin/packages/zone.js/npm_package/dist/zone-patch-electron.js ./packages/zone.js/test/extra/ &&
             yarn --cwd packages/zone.js electrontest

  # Windows jobs
  # Docs: https://circleci.com/docs/2.0/hello-world-windows/
  test_win:
    executor: windows-executor
    steps:
      - setup_win
      - run:
          # Ran into a command parsing problem where `-browser:chromium-local` was converted to
          # `-browser: chromium-local` (a space was added) in https://circleci.com/gh/angular/angular/357511.
          # Probably a powershell command parsing thing. There's no problem using a yarn script though.
          command: yarn circleci-win-ve
          no_output_timeout: 45m
      # Save bazel repository cache to use on subsequent runs.
      # We don't save node_modules because it's faster to use the linux workspace and reinstall.
      - save_cache:
          key: *cache_key_win
          paths:
            - "C:/Users/circleci/bazel_repository_cache"

  test_ivy_aot_win:
    executor: windows-executor
    steps:
      - setup_win
      - run:
          command: yarn circleci-win-ivy
          no_output_timeout: 45m


workflows:
  version: 2
  default_workflow:
    jobs:
      - setup:
          filters:
            branches:
              ignore: g3
      - lint:
          requires:
            - setup
      - test:
          requires:
            - setup
      - test_ivy_aot:
          requires:
            - setup
      - build-npm-packages:
          requires:
            - setup
      - build-ivy-npm-packages:
          requires:
            - setup
      - legacy-misc-tests:
          requires:
            - build-npm-packages
      - legacy-unit-tests-saucelabs:
          requires:
            - setup
      - test_saucelabs_bazel:
          # This job is currently a PoC and a subset of `legacy-unit-tests-saucelabs`. Running on
          # master only to avoid wasting resources.
          #
          # TODO: Run this job on all branches (including PRs) as soon as it is not a PoC.
          <<: *only_on_master
          requires:
            - setup
      - test_aio:
          requires:
            - setup
      - deploy_aio:
          requires:
            - test_aio
      - test_aio_local:
          requires:
            - build-npm-packages
      - test_aio_local:
          name: test_aio_local_viewengine
          viewengine: true
          requires:
            - build-npm-packages
      - test_aio_tools:
          requires:
            - build-npm-packages
      - test_docs_examples:
          requires:
            - build-npm-packages
      - test_docs_examples:
          name: test_docs_examples_ivy
          ivy: true
          requires:
            - build-ivy-npm-packages
      - aio_preview:
          # Only run on PR builds. (There can be no previews for non-PR builds.)
          <<: *only_on_pull_requests
          requires:
            - setup
      - test_aio_preview:
          requires:
            - aio_preview
      - integration_test:
          requires:
            - build-npm-packages
      - publish_packages_as_artifacts:
          # Only run on PR builds. (For non-PR builds, the `publish_snapshot` job.)
          <<: *only_on_pull_requests
          requires:
            - build-npm-packages
      - publish_snapshot:
          # Note: no filters on this job because we want it to run for all upstream branches
          # We'd really like to filter out pull requests here, but not yet available:
          # https://discuss.circleci.com/t/workflows-pull-request-filter/14396/4
          # Instead, the job just exits immediately at the first step.
          requires:
            # Only publish if tests and integration tests pass
            - test
            - test_ivy_aot
            - integration_test
            # Only publish if `aio`/`docs` tests using the locally built Angular packages pass
            - test_aio_local
            - test_aio_local_viewengine
            - test_docs_examples
            - test_docs_examples_ivy
            # Get the artifacts to publish from the build-packages-dist job
            # since the publishing script expects the legacy outputs layout.
            - build-npm-packages
            - build-ivy-npm-packages
            - legacy-unit-tests-saucelabs
            - legacy-misc-tests
      # FIXME: temporarily disabled until the components repo updates to typescript 3.6
      #- material-unit-tests:
      #    requires:
      #      - build-ivy-npm-packages
      - test_zonejs:
          requires:
            - setup
      # Windows Jobs
      # These are very slow so we run them on non-PRs only for now.
      # TODO: remove the filter when CircleCI makes Windows FS faster.
      # The Windows jobs are only run after their non-windows counterparts finish successfully.
      # This isn't strictly necessary as there is no artifact dependency, but helps economize
      # CI resources by not attempting to build when we know should fail.
      - test_win:
          <<: *skip_on_pull_requests
          requires:
            - test
      - test_ivy_aot_win:
          <<: *skip_on_pull_requests
          requires:
            - test_ivy_aot

  aio_monitoring:
    jobs:
      - setup
      - aio_monitoring_stable:
          requires:
            - setup
      - aio_monitoring_next:
          requires:
            - setup
    triggers:
      - schedule:
          <<: *only_on_master
          # Runs AIO monitoring jobs at 10:00AM every day.
          cron: "0 10 * * *"