Building Scala Projects with Mill

Diagram

Mill is a fast multi-language JVM build tool that supports Scala, making your common development workflows 5-10x faster to Maven, or 2-4x faster than Gradle, and easier to use than SBT. Mill aims to make your JVM project’s build process performant, maintainable, and flexible even as it grows from a small project to a large codebase or monorepo with hundreds of modules:

Mill is used to build many mainstream Scala projects, such as the Coursier dependency resolver, Scala-CLI, and the Ammonite REPL

Mill borrows ideas from other tools like Maven, Gradle, Bazel, or SBT. It tries to learn from the strengths of each tool, while improving on their weaknesses.

Compared to SBT:

  • Mill makes customizing the build yourself much easier: most of what build tools do work with files and call subprocesses, and Mill makes doing that yourself easy. This means you can always make your Mill build do exactly what you want, and are not beholden to third-party plugins that may not exist, be well maintained, or interact well with each other.

  • Mill is much more performant: SBT has enough overhead that even a dozen subprojects is enough to slow it down, while Mill can handle hundreds of modules without issue. Custom tasks in SBT re-execute every time, whereas in Mill they are cached automatically. Mill’s watch-for-changes-and-re-run implementation has much lower latency than SBT’s. The list of ways Mill improves upon SBT’s performance is long, and at the command line you can really feel it

  • Mill builds are much easier to understand: Your Mill build is made of bog-standard objects and defs, rather than SBT’s four-dimensional task matrix. Your IDE’s "jump-to-definition" in Mill actually brings you to the implementation of a task, rather than an SBT taskKey declaration. Customizing things is as simple as writing or overriding `def`s. The net effect is that despite both tools' build files being written in Scala, Mill’s build files are much easier to understand and maintain.

For a more detailed dive into the problems with SBT or how Mill improves upon them, check out the following blog posts:

If you’re interested in the fundamental ideas behind Mill, rather than the user-facing benefits discussed above, check out the section on Mill Design Principles:

The rest of this page contains a quick introduction to getting start with using Mill to build a simple Scala program. The other pages of this doc-site go into more depth, with more examples of how to use Mill and more details of how the Mill build tool works.

If you are using Mill, you will find the following book by the Author useful in using Mill and its supporting libraries to the fullest:

Simple Scala Module

build.mill (download, browse)
package build
import mill._, scalalib._

object `package` extends RootModule with ScalaModule {
  def scalaVersion = "2.13.11"
  def ivyDeps = Agg(
    ivy"com.lihaoyi::scalatags:0.12.0",
    ivy"com.lihaoyi::mainargs:0.6.2"
  )

  object test extends ScalaTests {
    def ivyDeps = Agg(ivy"com.lihaoyi::utest:0.8.4")
    def testFramework = "utest.runner.Framework"
  }
}

This is a basic Mill build for a single ScalaModule, with two third-party dependencies and a test suite using the uTest framework. As a single-module project, it extends RootModule to mark object package as the top-level module in the build. This lets us directly perform operations ./mill compile or ./mill run without needing to prefix it as foo.compile or foo.run.

You can download this example project using the download link above if you want to try out the commands below yourself. The only requirement is that you have some version of the JVM installed; the ./mill script takes care of any further dependencies that need to be downloaded.

The source code for this module lives in the src/ folder. Output for this module (compiled files, resolved dependency lists, …​) lives in out/.

build.mill
src/
    Foo.scala
resources/
    ...
test/
    src/
        FooTests.scala
out/
    compile.json
    compile.dest/
    ...
    test/
        compile.json
        compile.dest/
        ...

This example project uses two third-party dependencies - MainArgs for CLI argument parsing, Scalatags for HTML generation - and uses them to wrap a given input string in HTML templates with proper escaping.

Typical usage of a ScalaModule is shown below:

> ./mill resolve _ # List what tasks are available to run
assembly
...
clean
...
compile
...
run
...
show
...
inspect
...

> ./mill inspect compile # Show documentation and inputs of a task
compile(ScalaModule.scala:...)
    Compiles the current module to generate compiled classfiles/bytecode.
Inputs:
    scalaVersion
    upstreamCompileOutput
    allSourceFiles
    compileClasspath

> ./mill compile # compile sources into classfiles
...
compiling 1 Scala source to...

> ./mill run # run the main method, if any
error: Missing argument: --text <str>

> ./mill run --text hello
<h1>hello</h1>

> ./mill test
...
+ foo.FooTests...simple ...  <h1>hello</h1>
+ foo.FooTests...escaping ...  <h1>&lt;hello&gt;</h1>

> ./mill assembly # bundle classfiles and libraries into a jar for deployment

> ./mill show assembly # show the output of the assembly task
".../out/assembly.dest/out.jar"

> java -jar ./out/assembly.dest/out.jar --text hello
<h1>hello</h1>

> ./out/assembly.dest/out.jar --text hello # mac/linux
<h1>hello</h1>

The output of every Mill task is stored in the out/ folder under a name corresponding to the task that created it. e.g. The assembly task puts its metadata output in out/assembly.json, and its output files in out/assembly.dest. You can also use show to make Mill print out the metadata output for a particular task.

Additional Mill tasks you would likely need include:

$ mill runBackground # run the main method in the background

$ mill clean <task>  # delete the cached output of a task, terminate any runBackground

$ mill launcher      # prepares a foo/launcher.dest/run you can run later

$ mill jar           # bundle the classfiles into a jar suitable for publishing

$ mill -i console    # start a Scala console within your project

$ mill -i repl       # start an Ammonite Scala REPL within your project

You can run mill resolve __ to see a full list of the different tasks that are available, mill resolve _ to see the tasks within foo, mill inspect compile to inspect a task’s doc-comment documentation or what it depends on, or mill show foo.scalaVersion to show the output of any task.

The most common tasks that Mill can run are cached targets, such as compile, and un-cached commands such as foo.run. Targets do not re-evaluate unless one of their inputs changes, whereas commands re-run every time.

Custom Build Logic

Mill makes it very easy to customize your build graph, overriding portions of it with custom logic. In this example, we override the JVM resources of our ScalaModule - normally the resources/ folder - to instead contain a single generated text file containing the line count of all the source files in that module

build.mill (download, browse)
package build
import mill._, scalalib._

object `package` extends RootModule with ScalaModule {
  def scalaVersion = "2.13.11"

  /** Total number of lines in module source files */
  def lineCount = Task {
    allSourceFiles().map(f => os.read.lines(f.path).size).sum
  }

  /** Generate resources using lineCount of sources */
  override def resources = Task {
    os.write(Task.dest / "line-count.txt", "" + lineCount())
    Seq(PathRef(Task.dest))
  }
}

The addition of lineCount and resources overrides the previous resource folder provided by ScalaModule (labelled resource.super below), replacing it with the destination folder of the new resources task, which is wired up to lineCount:

Diagram 
> mill run
...
Line Count: 17

> mill show lineCount
17

> mill inspect lineCount
lineCount(build.mill:...)
    Total number of lines in module source files
Inputs:
    allSourceFiles

Above, def lineCount is a new build task we define, which makes use of allSourceFiles (an existing task) and is in-turn used in our override of resources (also an existing task). os.read.lines and os.write come from the OS-Lib library, which is one of Mill’s [Bundled Libraries]. This generated file can then be loaded and used at runtime, as see in the output of mill run

While this is a toy example, it shows how easy it is to customize your Mill build to include the kinds of custom logic common in the build config of most real-world projects.

This customization is done in a principled fashion familiar to most programmers - object-orienting overrides - rather than ad-hoc monkey-patching or mutation common in other build tools. You never have "spooky action at a distance" affecting your build / graph definition, and your IDE can always help you find the final override of any particular build task as well as where any overriden implementations may be defined.

Unlike normal methods, custom user-defined tasks in Mill benefit from all the same things that built-in tasks do: automatic caching (in the out/ folder), parallelism (with the -j/--jobs flag), inspectability (via show/inspect), and so on. While these things may not matter for such a simple example that runs quickly, they ensure that custom build logic remains performant and maintainable even as the complexity of your project grows.

Multi-Module Project

build.mill (download, browse)
package build
import mill._, scalalib._

trait MyModule extends ScalaModule {
  def scalaVersion = "2.13.11"
  object test extends ScalaTests {
    def ivyDeps = Agg(ivy"com.lihaoyi::utest:0.8.4")
    def testFramework = "utest.runner.Framework"
  }
}

object foo extends MyModule {
  def moduleDeps = Seq(bar)
  def ivyDeps = Agg(ivy"com.lihaoyi::mainargs:0.4.0")
}

object bar extends MyModule {
  def ivyDeps = Agg(ivy"com.lihaoyi::scalatags:0.8.2")
}

This example contains a simple Mill build with two modules, foo and bar. We don’t mark either module as top-level using extends RootModule, so running tasks needs to use the module name as the prefix e.g. foo.run or bar.run. You can define multiple modules the same way you define a single module, using def moduleDeps to define the relationship between them.

Note that we split out the test submodule configuration common to both modules into a separate trait MyModule. This lets us avoid the need to copy-paste common settings, while still letting us define any per-module configuration such as ivyDeps specific to a particular module.

The above builds expect the following project layout:

build.mill
foo/
    src/
        Foo.scala
bar/
    src/
        Bar.scala
    test/
        src/
            BarTests.scala
out/
    foo/
        compile.json
        compile.dest/
        ...
    bar/
        compile.json
        compile.dest/
        ...
        test/
            compile.json
            compile.dest/
            ...

Typically, both source code and output files in Mill follow the module hierarchy, so e.g. input to the foo module lives in foo/src/ and compiled output files live in out/foo/compile.dest. You can use mill resolve to list out what tasks you can run, e.g. mill resolve __.run below which lists out all the run tasks:

> mill resolve __.run
foo.run
bar.run

> mill foo.run --foo-text hello --bar-text world
Foo.value: hello
Bar.value: <h1>world</h1>

> mill bar.run world
Bar.value: <h1>world</h1>

> mill bar.test
...
...bar.BarTests...simple...
...bar.BarTests...escaping...

Mill’s evaluator will ensure that the modules are compiled in the right order, and recompiled as necessary when source code in each module changes.

You can use wildcards and brace-expansion to select multiple tasks at once or to shorten the path to deeply nested tasks. If you provide optional task arguments and your wildcard or brace-expansion is resolved to multiple tasks, the arguments will be applied to each of the tasks.

Table 1. Wildcards and brace-expansion

Wildcard

Function

_

matches a single segment of the task path

__

matches arbitrary segments of the task path

{a,b}

is equal to specifying two tasks a and b

You can use the ` symbol to add another task with optional arguments. If you need to feed a ` as argument to your task, you can mask it by preceding it with a backslash (\).

> mill foo._.compile # Runs `compile` for all direct sub-modules of `foo`

> mill foo.__.test # Runs `test` for all transitive sub-modules of `foo`

> mill {foo,bar}.__.testCached # Runs `testCached` for all sub-modules of `foo` and `bar`

> mill __.compile + foo.__.test # Runs all `compile` tasks and all tests under `foo`.

For more details on the query syntax, check out the documentation for [Task Query Syntax]

Watch and Re-evaluate

You can use the --watch flag to make Mill watch a task’s inputs, re-evaluating the task as necessary when the inputs change:

$ mill --watch foo.compile
$ mill --watch foo.run
$ mill -w foo.compile
$ mill -w foo.run

Mill’s --watch flag watches both the files you are building using Mill, as well as Mill’s own build.mill file and anything it imports, so any changes to your build.mill will automatically get picked up.

For long-running processes like web servers, you can use runBackground to make sure they recompile and restart when code changes, forcefully terminating the previous process even though it may be still alive:

$ mill -w foo.compile
$ mill -w foo.runBackground

Parallel Task Execution

By default, mill will evaluate all tasks in parallel, with the number of concurrent tasks equal to the number of cores on your machine.

You can use the --jobs (-j) to configure explicitly how many concurrent tasks you wish to run

Example: Use up to 4 parallel threads to compile all modules:

mill -j4 __.compile

To disable parallel execution use -j1.

mill generates an output file in out/mill-chrome-profile.json that can be loaded into the Chrome browser’s chrome://tracing page for visualization. This can make it much easier to analyze your parallel runs to find out what’s taking the most time:

ChromeTracing.png

Note that the maximal possible parallelism depends both on the number of cores available as well as the task and module structure of your project, as tasks that depend on one another other cannot be processed in parallel

Mill Cheat Sheet

Mill is a command-line tool and supports various options. Run mill --help for a complete list of options and a cheat-sheet of how to work with tasks:

Output of mill --help
Mill Build Tool, version 0.12.0-RC3-6-98487d
Usage: mill [options] task [task-options] [+ task ...]

task cheat sheet:
  mill resolve _                 # see all top-level tasks and modules
  mill resolve __.compile        # see all `compile` tasks in any module (recursively)

  mill foo.bar.compile           # compile the module `foo.bar`

  mill foo.run --arg 1           # run the main method of the module `foo` and pass in `--arg 1`
  mill -i foo.console            # run the Scala console for the module `foo` (if it is a ScalaModule)

  mill foo.__.test               # run tests in modules nested within `foo` (recursively)
  mill foo.test arg1 arg2        # run tests in the `foo` module passing in test arguments `arg1 arg2`
  mill foo.test + bar.test       # run tests in the `foo` module and `bar` module
  mill '{foo,bar,qux}.test'      # run tests in the `foo` module, `bar` module, and `qux` module

  mill foo.assembly              # generate an executable assembly of the module `foo`
  mill show foo.assembly         # print the output path of the assembly of module `foo`
  mill inspect foo.assembly      # show docs and metadata for the `assembly` task on module `foo`

  mill clean foo.assembly        # delete the output of `foo.assembly` to force re-evaluation
  mill clean                     # delete the output of the entire build to force re-evaluation

  mill path foo.run foo.sources  # print the task chain showing how `foo.run` depends on `foo.sources`
  mill visualize __.compile      # show how the `compile` tasks in each module depend on one another

options:
  -D --define <k=v>    Define (or overwrite) a system property.
  --allow-positional   Allows command args to be passed positionally without `--arg` by default
  -b --bell            Ring the bell once if the run completes successfully, twice if it fails.
  --bsp                Enable BSP server mode.
  --color <bool>       Toggle colored output; by default enabled only if the console is interactive
  -d --debug           Show debug output on STDOUT
  --disable-callgraph  Disables fine-grained invalidation of tasks based on analyzing code changes.
                       If passed, you need to manually run `clean` yourself after build changes.
  --help               Print this help message and exit.
  -i --interactive     Run Mill in interactive mode, suitable for opening REPLs and taking user
                       input. This implies --no-server. Must be the first argument.
  --import <str>       Additional ivy dependencies to load into mill, e.g. plugins.
  -j --jobs <str>      The number of parallel threads. It can be an integer e.g. `5` meaning 5
                       threads, an expression e.g. `0.5C` meaning half as many threads as available
                       cores, or `C-2` meaning 2 threads less than the number of cores. `1` disables
                       parallelism and `0` (the default) uses 1 thread per core.
  -k --keep-going      Continue build, even after build failures.
  --meta-level <int>   Select a meta-level to run the given tasks. Level 0 is the main project in
                       `build.mill`, level 1 the first meta-build in `mill-build/build.mill`, etc.
  --no-server          Run without a background server. Must be the first argument.
  -s --silent          Make ivy logs during script import resolution go silent instead of printing
  --ticker <bool>      Enable ticker log (e.g. short-lived prints of stages and progress bars).
  -v --version         Show mill version information and exit.
  -w --watch           Watch and re-run the given tasks when when their inputs change.
  task <str>...        The name or a pattern of the tasks(s) you want to build.

Please see the documentation at https://mill-build.org for more details

Note that options and flags given before the task name are passed to Mill, while options and flags after the task name are passed to the task. This mirrors how most command line script or task runners work (e.g. python <python-args> script.py <script-args>)