13 KiB
Building Angular with Bazel
Note: this doc is for developing Angular, it is not public documentation for building an Angular application with Bazel.
The Bazel build tool (http://bazel.build) provides fast, reliable incremental builds. We plan to migrate Angular's build scripts to Bazel.
Installation
In order to ensure that everyone builds Angular in a consistent way, Bazel will be installed through NPM and therefore it's not necessary to install Bazel manually.
The binaries for Bazel will be provided by the @bazel/bazel
NPM package and its platform-specific dependencies.
You can access Bazel with the yarn bazel
command
Configuration
The WORKSPACE
file indicates that our root directory is a
Bazel project. It contains the version of the Bazel rules we
use to execute build steps, from build_bazel_rules_typescript
.
The sources on GitHub are published from Google's internal
repository (google3).
Bazel accepts a lot of options. We check in some options in the
.bazelrc
file. See the bazelrc doc. For example, if you don't
want Bazel to create several symlinks in your project directory
(bazel-*
) you can add the line build --symlink_prefix=/
to your
.bazelrc
file.
Building Angular
- Build a package:
yarn bazel build packages/core
- Build all packages:
yarn bazel build packages/...
You can use ibazel to get a "watch mode" that continuously keeps the outputs up-to-date as you save sources. Note this is new as of May 2017 and not very stable yet.
Testing Angular
- Test package in node:
yarn bazel test packages/core/test:test
- Test package in karma:
yarn bazel test packages/core/test:test_web
- Test all packages:
yarn bazel test packages/...
You can use ibazel to get a "watch mode" that continuously keeps the outputs up-to-date as you save sources.
Various Flags Used For Tests
If you're experiencing problems with seemingly unrelated tests failing, it may be because you're not using the proper flags with your Bazel test runs in Angular.
See also: //.bazelrc
where --define=compile=legacy
is defined as default.
--config=debug
: build and launch in debug mode (see debugging instructions below)--test_arg=--node_options=--inspect=9228
: change the inspector port.--define=compile=<option>
Controls if ivy or legacy mode is enabled. This switches which compiler is used (ngc, ngtsc, or a tsc pass-through mode).legacy
: (default behavior) compile against View Engine, e.g.--define=compile=legacy
aot
: Compile in ivy AOT mode, e.g.--define=compile=aot
--test_tag_filters=<tag>
: filter tests down to tags defined in thetag
config of your rules in any givenBUILD.bazel
.no-ivy-aot
: Useful for excluding build and test targets that are not meant to be executed in Ivy AOT mode (--define=compile=aot
).ivy-only
: Useful for excluding all Ivy build and tests targets with--define=compile=legacy
.fixme-ivy-aot
: Useful for including/excluding build and test targets that are currently broken in Ivy AOT mode (--define=compile=aot
).
Debugging a Node Test
- Open chrome at: chrome://inspect
- Click on
Open dedicated DevTools for Node
to launch a debugger. - Run test:
yarn bazel test packages/core/test:test --config=debug
The process should automatically connect to the debugger. For additional info and testing options, see the nodejs_test documentation.
Debugging a Node Test in VSCode
First time setup:
- Go to Debug > Add configuration (in the menu bar) to open
launch.json
- Add the following to the
configurations
array:
{
"name": "Attach (inspect)",
"type": "node",
"request": "attach",
"port": 9229,
"address": "localhost",
"restart": false,
"sourceMaps": true,
"localRoot": "${workspaceRoot}",
"remoteRoot": null
},
{
"name": "Attach (no-sm,inspect)",
"type": "node",
"request": "attach",
"port": 9229,
"address": "localhost",
"restart": false,
"sourceMaps": false,
"localRoot": "${workspaceRoot}",
"remoteRoot": null
},
Setting breakpoints directly in your code files may not work in VSCode. This is because the files you're actually debugging are built files that exist in a ./private/...
folder.
The easiest way to debug a test for now is to add a debugger
statement in the code
and launch the bazel corresponding test (yarn bazel test <target> --config=debug
).
Bazel will wait on a connection. Go to the debug view (by clicking on the sidebar or
Apple+Shift+D on Mac) and click on the green play icon next to the configuration name
(ie Attach (inspect)
).
Debugging a Karma Test
- Run test:
yarn bazel run packages/core/test:test_web
- Open chrome at: http://localhost:9876/debug.html
- Open chrome inspector
Debugging Bazel rules
Open external
directory which contains everything that bazel downloaded while executing the workspace file:
open $(bazel info output_base)/external
See subcommands that bazel executes (helpful for debugging):
yarn bazel build //packages/core:package -s
To debug nodejs_binary executable paths uncomment find . -name rollup 1>&2
(~ line 96) in
open $(bazel info output_base)/external/build_bazel_rules_nodejs/internal/node_launcher.sh
Stamping
Bazel supports the ability to include non-hermetic information from the version control system in built artifacts. This is called stamping. You can see an overview at https://www.kchodorow.com/blog/2017/03/27/stamping-your-builds/ In our repo, here is how it's configured:
- In
tools/bazel_stamp_vars.sh
we run thegit
commands to generate our versioning info. - In
.bazelrc
we register this script as the value for theworkspace_status_command
flag. Bazel will run the script when it needs to stamp a binary.
Note that Bazel has a --stamp
argument to yarn bazel build
, but this has no effect since our stamping takes place in Skylark rules. See https://github.com/bazelbuild/bazel/issues/1054
Remote cache
Bazel supports fetching action results from a cache, allowing a clean build to pick up artifacts from prior builds. This makes builds incremental, even on CI. It works because Bazel assigns a content-based hash to all action inputs, which is used as the cache key for the action outputs. Thanks to the hermeticity property, we can skip executing an action if the inputs hash is already present in the cache.
Of course, non-hermeticity in an action can cause problems. At worst, you can fetch a broken artifact from the cache, making your build non-reproducible. For this reason, we are careful to implement our Bazel rules to depend only on their inputs.
Currently we only use remote caching on CircleCI. We could enable it for developer builds as well, which would make initial builds much faster for developers by fetching already-built artifacts from the cache.
This feature is experimental, and developed by the CircleCI team with guidance from Angular. Contact Alex Eagle with questions.
How it's configured:
- In
.circleci/config.yml
, each CircleCI job downloads a proxy binary, which is built from https://github.com/notnoopci/bazel-remote-proxy. The download is done by running.circleci/setup_cache.sh
. When the feature graduates from experimental, this proxy will be installed by default on every CircleCI worker, and this step will not be needed. - Next, each job runs the
setup-bazel-remote-cache
anchor. This starts up the proxy running in the background. In the CircleCI UI, you'll see this step continues running while later steps run, and you can see logging from the proxy process. - Bazel must be configured to connect to the proxy on a local port. This configuration lives in
.circleci/bazel.rc
and is enabled because we overwrite the system Bazel settings in /etc/bazel.bazelrc with this file. - Each
bazel
command in.circleci/config.yml
picks up and uses the caching flags.
Diagnosing slow builds
If a build seems slow you can use Bazel to diagnose where time is spent.
The first step is to generate a profile of the build using the --profile filename_name.profile
flag.
yarn bazel build //packages/compiler --profile filename_name.profile
This will generate a filename_name.profile
that you can then analyse using analyze-profile command.
Using the console profile report
You can obtain a simple report directly in the console by running:
yarn bazel analyze-profile filename_name.profile
This will show the phase summary, individual phase information and critical path.
You can also list all individual tasks and the time they took using --task_tree
.
yarn bazel analyze-profile filename_name.profile --task_tree ".*"
To show all tasks that take longer than a certain threshold, use the --task_tree_threshold
flag.
The default behaviour is to use a 50ms threshold.
yarn bazel analyze-profile filename_name.profile --task_tree ".*" --task_tree_threshold 5000
--task_tree
takes a regexp as argument that filters by the text shown after the time taken.
Compiling TypeScript shows as:
70569 ACTION_EXECUTE (10974.826 ms) Compiling TypeScript (devmode) //packages/compiler:compiler []
To filter all tasks by TypeScript compilations that took more than 5 seconds, use:
yarn bazel analyze-profile filename_name.profile --task_tree "Compiling TypeScript" --task_tree_threshold 5000
Using the HTML profile report
A more comprehensive way to visualize the profile information is through the HTML report:
yarn bazel analyze-profile filename_name.profile --html --html_details --html_histograms
This will generate a filename_name.profile.html
file that you can open in your browser.
On the upper right corner that is a small table of contents with links to three areas: Tasks, Legend and Statistics.
In the Tasks section you will find a graph of where time is spent. Legend shows what the colors in the Tasks graph mean. Hovering over the background will show what phase that is, while hovering over bars will show more details about that specific action.
The Statistics section shows how long each phase took and how time was spent in that phase. Usually the longest one is the execution phase, which also includes critical path information.
Also in the Statistics section are the Skylark statistic, split in User-Defined and Builtin function execution time. You can click the "self" header twice to order the table by functions where the most time (in ms) is spent.
When diagnosing slow builds you should focus on the top time spenders across all phases and functions. Usually there is a single item (or multiple items of the same kind) where the overwhelming majority of time is spent.
Known issues
Webstorm
The autocompletion in WebStorm can be added via a Bazel plugin intended for IntelliJ IDEA, but the plugin needs to be installed in a special way. See bazelbuild/intellij#246 for more info.
Xcode
If you see the following error:
$ yarn bazel build packages/...
ERROR: /private/var/tmp/[...]/external/local_config_cc/BUILD:50:5: in apple_cc_toolchain rule @local_config_cc//:cc-compiler-darwin_x86_64: Xcode version must be specified to use an Apple CROSSTOOL
ERROR: Analysis of target '//packages/core/test/render3:render3' failed; build aborted: Analysis of target '@local_config_cc//:cc-compiler-darwin_x86_64' failed; build aborted
It might be linked to an interaction with VSCode. If closing VSCode fixes the issue, you can add the following line to your VSCode configuration:
"files.exclude": {"bazel-*": true}
source: https://github.com/bazelbuild/bazel/issues/4603
If VSCode is not the root cause, you might try:
- Quit VSCode (make sure no VSCode is running).
bazel clean --expunge
sudo xcode-select -s /Applications/Xcode.app/Contents/Developer
sudo xcodebuild -license
yarn bazel build //packages/core # Run a build outside VSCode to pre-build the xcode; then safe to run VSCode