@bazel/concatjs
Concatjs is a JavaScript bundler, in a trivial sense: the UNIX cat
command is a basic implementation:
$ cat one.js two.js > bundle.js
Clearly self-evident is that this bundler is super-fast and simple. A performant implementation adds some in-memory caching, and for developer ergonomics you add a simple IIFE wrapper around each file so that the Chrome DevTools shows the files in the tree as if they had been independently loaded.
However at its core, concatjs requires a big tradeoff of a migration cost to buy-in, to get this incredible performance. The path of the JavaScript files is lost in the bundling process, so they must contain their module ID internally.
Named AMD/UMD modules and goog.module
are the two JS module formats that are compatible with concatjs.
Most packages do not ship with this format, so in order to use concatjs tooling, you have to shim your code and dependencies. See the Compatibility section below.
This is at the core of how Google does JavaScript development.
So Bazel rules that originated in Google's codebase have affordances for concatjs.
For example ts_library
produces named AMD modules in its "devmode" output, and
karma_web_test
expects to bundle inputs using concatjs.
Compatibility
First-party code
First-party code has to be authored as named AMD/UMD modules. This is also historically referred to as "RequireJS" modules since that's the JS loader that is typically used with them.
If you write TypeScript, you can do this following their documentation.
There is an example in this repository: we have an index.ts
file that wants
to be used with require.js require("@bazel/concatjs")
.
So it
declares
that module name
using the TS triple-slash syntax:
///<amd-module name="@bazel/concatjs"/>
it is also compiled with
the "compilerOptions": { "module": "umd" }
TypeScript setting.
Third-party code
To make it easier to produce a UMD version of a third-party npm package, we automatically generate a target that uses Browserify to build one, using the main
entry from the package's package.json
.
In most cases this will make the package loadable under concatjs.
This target has a __umd
suffix. For example, if your library is at @npm//foo
then the UMD target is @npm//foo:foo__umd
.
An example where this fixes a users issue: https://github.com/bazelbuild/rules_nodejs/issues/2317#issuecomment-735921318
In some cases, the generated UMD bundle is not sufficient, and in others it fails to build because it requires some special Browserify configuration.
You can always write your own shim that grabs a symbol from a package you use, and exposes it in an AMD/require.js-compatible way.
For example, even though RxJS ships with a UMD bundle, it contains multiple entry points and uses anonymous modules, not named modules. So our Angular/concatjs example has a rxjs_shims.js
file that exposes some RxJS operators, then at https://github.com/bazelbuild/rules_nodejs/blob/2.3.1/examples/angular/src/BUILD.bazel#L65-L71 this is combined in a filegroup
with the rxjs.umd.js
file. Now we use this filegroup target when depending on RxJS in a concatjs_*
rule.
Ultimately by using concatjs, you're signing up for at least a superficial understanding of these shims and may need to update them when you change your dependencies.
Serving JS in development mode under Bazel
There are two choices for development mode:
- Use the
concatjs_devserver
rule to bring up our simple, fast development server. This is intentionally very simple, to help you get started quickly. However, since there are many development servers available, we do not want to mirror their features in yet another server we maintain. - Teach your real frontend server to serve files from Bazel's output directory.
This is not yet documented. Choose this option if you have an existing server
used in development mode, or if your requirements exceed what the
concatjs_devserver
supports. Be careful that your development round-trip stays fast (should be under two seconds).
To use concatjs_devserver
, you simply load
the rule, and call it with deps
that
point to your ts_library
target(s):
load("//packages/concatjs:index.bzl", "concatjs_devserver", "ts_library")
ts_library(
name = "app",
srcs = ["app.ts"],
)
concatjs_devserver(
name = "devserver",
# We'll collect all the devmode JS sources from these TypeScript libraries
deps = [":app"],
# This is the path we'll request from the browser, see index.html
serving_path = "/bundle.js",
# The devserver can serve our static files too
static_files = ["index.html"],
)
The index.html
should be the same one you use for production, and it should
load the JavaScript bundle from the path indicated in serving_path
.
If you don't have an index.html file, a simple one will be generated by the
concatjs_devserver
.
See examples/app
in this repository for a working example. To run the
devserver, we recommend you use ibazel:
$ ibazel run examples/app:devserver
ibazel
will keep the devserver program running, and provides a LiveReload
server so the browser refreshes the application automatically when each build
finishes.
Testing with Karma
The karma_web_test
rule runs karma tests with Bazel.
It depends on rules_webtesting, so you need to add this to your WORKSPACE
if you use the web testing rules in @bazel/concatjs
:
# Fetch transitive Bazel dependencies of karma_web_test
http_archive(
name = "io_bazel_rules_webtesting",
sha256 = "e9abb7658b6a129740c0b3ef6f5a2370864e102a5ba5ffca2cea565829ed825a",
urls = ["https://github.com/bazelbuild/rules_webtesting/releases/download/0.3.5/rules_webtesting.tar.gz"],
)
# Set up web testing, choose browsers we can test on
load("@io_bazel_rules_webtesting//web:repositories.bzl", "web_test_repositories")
web_test_repositories()
load("@io_bazel_rules_webtesting//web/versioned:browsers-0.3.3.bzl", "browser_repositories")
browser_repositories(
chromium = True,
firefox = True,
)
Known issues with running Chromium for macOS/Windows in Bazel
For macOS and Windows, Chromium comes with files that contain spaces in their file names. This breaks runfile tree creation within Bazel due to a bug. There are various workarounds that allow for Chromium on these platforms:
- Instruct Bazel to automatically disable runfile tree creation if not needed. More details here
- Instruct Bazel to use an alternative experimental approach for creating runfile trees. More details here
Installing with user-managed dependencies
If you didn't use the yarn_install
or npm_install
rule to create an npm
workspace, you'll have to declare a rule in your root BUILD.bazel
file to execute karma:
# Create a karma rule to use in karma_web_test_suite karma
# attribute when using user-managed dependencies
nodejs_binary(
name = "karma/karma",
entry_point = "//:node_modules/karma/bin/karma",
# Point bazel to your node_modules to find the entry point
data = ["//:node_modules"],
)
Rules
concatjs_devserver
concatjs_devserver is a simple development server intended for a quick "getting started" experience.
Additional documentation here
Example usage (generated)
load("@build_bazel_rules_nodejs//packages/concatjs:index.bzl", "concatjs_devserver")
concatjs_devserver(
# A unique name for this target.
name = "",
)
name
A unique name for this target.
additional_root_paths
Additional root paths to serve static_files
from.
Paths should include the workspace name such as ["__main__/resources"]
bootstrap
Scripts to include in the JS bundle before the module loader (require.js)
deps
Targets that produce JavaScript, such as ts_library
devserver
Go based devserver executable.
With cross-platform RBE for OSX & Windows ctx.executable.devserver will be linux as --cpu and --host_cpu must be overridden to k8. However, we still want to be able to run the devserver on the host machine so we need to include the host devserver binary, which is ctx.executable.devserver_host, in the runfiles. For non-RBE and for RBE with a linux host, ctx.executable.devserver & ctx.executable.devserver_host will be the same binary.
Defaults to precompiled go binary setup by @bazel/typescript npm package
devserver_host
Go based devserver executable for the host platform. Defaults to precompiled go binary setup by @bazel/typescript npm package
entry_module
The entry_module
should be the AMD module name of the entry module such as "__main__/src/index".
concatjs_devserver
concats the following snippet after the bundle to load the application:
require(["entry_module"]);
port
The port that the devserver will listen on.
scripts
User scripts to include in the JS bundle before the application sources
serving_path
The path you can request from the client HTML which serves the JavaScript bundle. If you don't specify one, the JavaScript can be loaded at /_/ts_scripts.js
static_files
Arbitrary files which to be served, such as index.html. They are served relative to the package where this rule is declared.
ts_library
type-check and compile a set of TypeScript sources to JavaScript.
It produces declarations files (.d.ts
) which are used for compiling downstream
TypeScript targets and JavaScript for the browser and Closure compiler.
By default, ts_library
uses the tsconfig.json
file in the workspace root
directory. See the notes about the tsconfig
attribute below.
Serving TypeScript for development
ts_library
is typically served by the concatjs_devserver rule, documented in the @bazel/concatjs
package.
Accessing JavaScript outputs
The default output of the ts_library
rule is the .d.ts
files.
This is for a couple reasons:
- help ensure that downstream rules which access default outputs will not require a cascading re-build when only the implementation changes but not the types
- make you think about whether you want the
devmode
(namedUMD
) orprodmode
outputs
You can access the JS output by adding a filegroup
rule after the ts_library
,
for example
ts_library(
name = "compile",
srcs = ["thing.ts"],
)
filegroup(
name = "thing.js",
srcs = ["compile"],
# Change to es6_sources to get the 'prodmode' JS
output_group = "es5_sources",
)
my_rule(
name = "uses_js",
deps = ["thing.js"],
)
Example usage (generated)
load("@build_bazel_rules_nodejs//packages/concatjs:index.bzl", "ts_library")
ts_library(
# A unique name for this target.
name = "",
# The TypeScript source files to compile.
srcs = [],
)
name
A unique name for this target.
angular_assets
Additional files the Angular compiler will need to read as inputs. Includes .css and .html files
compiler
Sets a different TypeScript compiler binary to use for this library.
For example, we use the vanilla TypeScript tsc.js for bootstrapping,
and Angular compilations can replace this with ngc
.
The default ts_library compiler depends on the //@bazel/typescript
target which is setup for projects that use bazel managed npm deps and
install the @bazel/typescript npm package.
You can also use a custom compiler to increase the NodeJS heap size used for compilations.
To do this, declare your own binary for running tsc_wrapped
, e.g.:
nodejs_binary(
name = "tsc_wrapped_bin",
entry_point = "@npm//:node_modules/@bazel/typescript/internal/tsc_wrapped/tsc_wrapped.js",
templated_args = [
"--node_options=--max-old-space-size=2048",
],
data = [
"@npm//protobufjs",
"@npm//source-map-support",
"@npm//tsutils",
"@npm//typescript",
"@npm//@bazel/typescript",
],
)
then refer to that target in the compiler
attribute.
Note that nodejs_binary
targets generated by npm_install
/yarn_install
can include data dependencies
on packages which aren't declared as dependencies.
For example, if you use tsickle to generate Closure Compiler-compatible JS,
then it needs to be a data dependency of tsc_wrapped
so that it can be loaded at runtime.
data
deps
Compile-time dependencies, typically other ts_library targets
devmode_module
Set the typescript module
compiler option for devmode output.
This value will override the module
option in the user supplied tsconfig.
devmode_target
Set the typescript target
compiler option for devmode output.
This value will override the target
option in the user supplied tsconfig.
expected_diagnostics
generate_externs
internal_testing_type_check_dependencies
Testing only, whether to type check inputs that aren't srcs.
link_workspace_root
Link the workspace root to the bin_dir to support absolute requires like 'my_wksp/path/to/file'.
If source files need to be required then they can be copied to the bin_dir with copy_to_bin.
module_name
module_root
package_name
The package name that the linker will link this ts_library output as.
If package_path is set, the linker will link this package under
package_path
The package path in the workspace that the linker will link this ts_library output to.
If package_path is set, the linker will link this package under
prodmode_module
Set the typescript module
compiler option for prodmode output.
This value will override the module
option in the user supplied tsconfig.
prodmode_target
Set the typescript target
compiler option for prodmode output.
This value will override the target
option in the user supplied tsconfig.
runtime
runtime_deps
srcs
The TypeScript source files to compile.
supports_workers
Intended for internal use only.
Allows you to disable the Bazel Worker strategy for this library. Typically used together with the "compiler" setting when using a non-worker aware compiler binary.
tsconfig
A tsconfig.json file containing settings for TypeScript compilation.
Note that some properties in the tsconfig are governed by Bazel and will be
overridden, such as target
and module
.
The default value is set to //:tsconfig.json
by a macro. This means you must
either:
- Have your
tsconfig.json
file in the workspace root directory - Use an alias in the root BUILD.bazel file to point to the location of tsconfig:
alias(name="tsconfig.json", actual="//path/to:tsconfig-something.json")
and also make the tsconfig.json file visible to other Bazel packages:exports_files(["tsconfig.json"], visibility = ["//visibility:public"])
- Give an explicit
tsconfig
attribute to allts_library
targets
tsickle_typed
If using tsickle, instruct it to translate types to ClosureJS format
use_angular_plugin
Run the Angular ngtsc compiler under ts_library
Macros and Functions
karma_web_test
Runs unit tests in a browser with Karma.
When executed under bazel test
, this uses a headless browser for speed.
This is also because bazel test
allows multiple targets to be tested together,
and we don't want to open a Chrome window on your machine for each one. Also,
under bazel test
the test will execute and immediately terminate.
Running under ibazel test
gives you a "watch mode" for your tests. The rule is
optimized for this case - the test runner server will stay running and just
re-serve the up-to-date JavaScript source bundle.
To debug a single test target, run it with bazel run
instead. This will open a
browser window on your computer. Also you can use any other browser by opening
the URL printed when the test starts up. The test will remain running until you
cancel the bazel run
command.
This rule will use your system Chrome by default. In the default case, your
environment must specify CHROME_BIN so that the rule will know which Chrome binary to run.
Other browsers
and customLaunchers
may be set using the a base Karma configuration
specified in the config_file
attribute.
By default we open a headless Chrome. To use a real Chrome browser window, you can pass
--define DISPLAY=true
to Bazel, along with configuration_env_vars = ["DISPLAY"]
on
karma_web_test
.
Example usage (generated)
load("@build_bazel_rules_nodejs//packages/concatjs:index.bzl", "karma_web_test")
karma_web_test(
)
srcs
A list of JavaScript test files
deps
Other targets which produce JavaScript such as ts_library
data
Runtime dependencies
configuration_env_vars
Pass these configuration environment variables to the resulting binary. Chooses a subset of the configuration environment variables (taken from ctx.var), which also includes anything specified via the --define flag. Note, this can lead to different outputs produced by this rule.
bootstrap
JavaScript files to include before the module loader (require.js). For example, you can include Reflect,js for TypeScript decorator metadata reflection, or UMD bundles for third-party libraries.
runtime_deps
Dependencies which should be loaded after the module loader but before the srcs and deps.
These should be a list of targets which produce JavaScript such as ts_library
.
The files will be loaded in the same order they are declared by that rule.
static_files
Arbitrary files which are available to be served on request.
Files are served at:
/base/<WORKSPACE_NAME>/<path-to-file>
, e.g.
/base/npm_bazel_typescript/examples/testing/static_script.js
config_file
User supplied Karma configuration file. Bazel will override certain attributes of this configuration file. Attributes that are overridden will be outputted to the test log.
tags
Standard Bazel tags, this macro adds tags for ibazel support
peer_deps
list of peer npm deps required by karma_web_test
kwargs
Passed through to karma_web_test
karma_web_test_suite
Defines a test_suite of web_test targets that wrap a karma_web_test target.
This macro accepts all parameters in karma_web_test and adds additional parameters for the suite. See karma_web_test docs for all karma_web_test.
The wrapping macro is web_test_suite
which comes from rules_websting:
https://github.com/bazelbuild/rules_webtesting/blob/master/web/web.bzl.
Example usage (generated)
load("@build_bazel_rules_nodejs//packages/concatjs:index.bzl", "karma_web_test_suite")
karma_web_test_suite(
# The base name of the test
name = "",
)
name
The base name of the test
browsers
A sequence of labels specifying the browsers to use.
web_test_data
Data dependencies for the wrapper web_test targets.
wrapped_test_tags
A list of test tag strings to use for the wrapped karma_web_test target.
kwargs
Arguments for the wrapped karma_web_test target.