This directory contains end-to-end tests for Angular. Each directory is a self-contained application that exactly mimics how a user might expect Angular to work, so they allow high-fidelity reproductions of real-world issues.
For this to work, we first build the Angular distribution via yarn build
, then
install the distribution into each app.
To test Angular CLI applications, we use the cli-hello-world-*
integration tests.
When a significant change is released in the CLI, the applications should be updated with
ng update
:
$ cd integration/cli-hello-world[-*]
$ yarn install
$ yarn ng update @angular/cli @angular-devkit/build-angular
$ yarn build
$ yarn test
Afterwards the @angular/cli
and @angular-devkit/build-angular
should be reverted to the file:../
urls
and the main package.json
should be updated with the new versions.
The directory cli-hello-world-ivy-compat
contains a test for render3 used with the angular cli.
The cli-hello-world-ivy-minimal
contains a minimal ivy app that is meant to mimic the bazel
equivalent in packages/core/test/bundling/hello_world
, and should be kept similar.
The API for each test is:
- Each sub-directory here is an integration test
- Each test should have a
package.json
file - The test runner will run
yarn
andyarn test
on the package
This means that the test should be started by test script, like
"scripts": {"test": "runProgramA && assertResultIsGood"}
Note that the package.json
file uses a special file:../../dist
scheme to reference the Angular
packages, so that the locally-built Angular is installed into the test app.
Also, beware of floating (non-locked) dependencies. If in doubt, you can install the package
directly from file:../../node_modules
.
WARNING
Always ensure that
yarn.lock
files are up-to-date with the correspondingpackage.json
files (wrt the non-local dependencies - i.e. dependencies whose versions do not start withfile:
).You can update a
yarn.lock
file by runningyarn install
in the project subdirectory.
$ ./integration/run_tests.sh
The test runner will first re-build any stale npm packages, then cd
into each subdirectory to
execute the test.
The PR angular#33927 added the ability to run integration tests with Bazel. These tests can be resource intensive so it is recommended to limit the number of concurrent test jobs with the --local_test_jobs
bazel flag.
Locally, if Bazel uses all of your cores to run the maximum number of integration tests in parallel then this can lead to test timeouts and flakes and freeze up your machine while these tests are running. You can limit the number of concurrent local integration tests that run with:
yarn bazel test --local_test_jobs=<N> //integration/...
Set a reasonable local_test_jobs
limit for your local machine to prevent full cpu utilization during local development test runs.
To avoid having to specify this command line flag, you may want to include it in your .bazelrc.user
file:
test --local_test_jobs=<N>
The downside of this is that this will apply to all tests and not just the resource intensive integration tests.
Two of the integration tests that run Bazel-in-Bazel are particularly resource intensive and are tagged "manual" and "exclusive". To run these tests use,
yarn bazel test //integration/bazel:test
yarn bazel test //integration/cli-hello-world-ivy-minimal:test
When adding a new integration test, follow the steps below to add a bazel test target for the new test.
- Add a build file using the
ng_integration_test
rule from//integration:index.bzl
. - If test requires ports and does not support ethereal ports then make sure the port is unique and add it to the "manually configured ports" section to document which port it is using
- Add at least the following two entries
.bazelignore
(as they may contain BUILD files)integration/new_test/node_modules
integration/new_test/.yarn_local_cache
- Add any other untracked folders to
.bazelignore
that may containBUILD
files - If there are BUILD files in the integration test folder (except for the top-level one defining the test), add those folders to the
--deleted_packages
in the.bazelrc
. An example is thebazel_ngtsc_plugin
test within//integration/bazel_workspace_tests
.
Some integration ports must be managed manually to be unique and in other cases the tests are able to select a random free port.
Where ng e2e
is used we pass ng e2e --port 0
which prompts the cli
to select a random free port for the e2e test. The protractor.conf is
automatically updated to use this port.
Karma automatically finds a free port so no effort is needed there.
The manually configured ports are as follows:
TEST | PORT | CONFIGURATION |
---|---|---|
dynamic-compiler | 4201 | /e2e/browser.config.json: "port": 4201 |
ng_elements | 4205 | /e2e/browser.config.json: "port": 4205 |
platform-server | 4206 | /src/server.ts: app.listen(4206,... |
Note: This will become obsolete soon once we start running integration tests with RBE and within a sandbox environment.
For integration tests we use the Bazel-managed versions of chromium
. For both Karma and Protractor tests we set a number of browser testing flags. To avoid duplication, they will be listed and explained here and the code will reference this file for more information.
The sandbox needs to be disabled with the --no-sandbox
flag for both Karma and Protractor tests, because it causes Chrome to crash on some environments.
See: https://chromedriver.chromium.org/help/chrome-doesn-t-start See: https://github.com/puppeteer/puppeteer/blob/v1.0.0/docs/troubleshooting.md#chrome-headless-fails-due-to-sandbox-issues
So that browsers are not popping up and tearing down when running these tests we run Chrome in headless mode. The --headless
flag puts Chrome in headless mode and a number of other flags are recommended in this mode as well:
--headless
--disable-gpu
--disable-dev-shm-usage
--hide-scrollbars
--mute-audio
These come from the flags that puppeteer passes to chrome when it launches it in headless mode: https://github.com/puppeteer/puppeteer/blob/18f2ecdffdfc70e891750b570bfe8bea5b5ca8c2/lib/Launcher.js#L91
And from the flags that the Karma ChromeHeadless
browser passes to Chrome: https://github.com/karma-runner/karma-chrome-launcher/blob/5f70a76de87ecbb57f3f3cb556aa6a2a1a4f643f/index.js#L171
The --disable-dev-shm-usage
flag disables the usage of /dev/shm
because it causes Chrome to crash on some environments.
On CircleCI, the puppeteer provisioned Chrome crashes with CI we get Root cause: org.openqa.selenium.WebDriverException: unknown error: DevToolsActivePort file doesn't exist which resolves
without this flag.
See: https://github.com/puppeteer/puppeteer/blob/v1.0.0/docs/troubleshooting.md#tips See: https://stackoverflow.com/questions/50642308/webdriverexception-unknown-error-devtoolsactiveport-file-doesnt-exist-while-t
If size regression occurs, one way to debug is to get a build which shows the code before and after. Here are the steps to do that.
- Check out both the
main
branch as well as your change (let's refer to it aschange
branch) into two different working locations. (A suggested way to do this is usinggit worktree
.) - In both
main
andchange
locations update the failing testspackage.json
withNG_BUILD_DEBUG_OPTIMIZE=minify
environment variable so that the resulting build would contain a human readable but optimized output. As an example:- Open
integration/cli-hello-world/package.json
and prefixNG_BUILD_DEBUG_OPTIMIZE=minify
into thebuild
rule. Resulting in something like:"build": "NG_BUILD_DEBUG_OPTIMIZE=minify ng build --prod",
- Run
bazel test //integration/cli-hello-world:test --test_output=streamed --cache_test_results=no
to run the test. - Open the test temporary directory as printed out by Bazel.
- Diff the
main
vschange
to see the differences.myDiffTool change/integration/cli-hello-world/dist/main-es2015.*.js main/integration/cli-hello-world/dist/main-es2015.*.js
- The above should give you a better understanding as to what has changed and what is causing the regression.
- Open