Skip to main content
Version: 2.1.26

Low-level API

This page is a direct import of sections of the former README of coursier, and is possibly not up-to-date. The use of mdoc ensures the code snippets below compile fine with the latest version though.

Add to your build.sbt

//
libraryDependencies ++= Seq(
"io.get-coursier" %% "coursier-core" % "2.1.26",
"io.get-coursier" %% "coursier-cache" % "2.1.26"
)

The first module, "io.get-coursier" %% "coursier-core" % "2.1.26", contains among others, definitions, mainly in Definitions.scala, Resolution, representing a particular state of the resolution, and ResolutionProcess, that expects to be given metadata, wrapped in any Monad, then feeds these to Resolution, and at the end gives you the final Resolution, wrapped in the same Monad it was given input. This final Resolution has all the dependencies, including the transitive ones.

The second module, "io.get-coursier" %% "coursier-cache" % "2.1.26", is precisely in charge of fetching these input metadata. It uses its own coursier.util.Task as a Monad to wrap them. It also fetches artifacts (JARs, etc.). It caches all of these (metadata and artifacts) on disk, and validates checksums too.

In the code below, we'll assume some imports are around,

import coursier._
import coursier.version.{Version, VersionConstraint}

Resolving dependencies involves create an initial resolution state, with all the initial dependencies in it, like

val start = Resolution(
Seq(
Dependency(
Module(org"org.typelevel", name"cats-core_2.11"), VersionConstraint("0.6.0")
),
Dependency(
Module(org"org.scalaz", name"scalaz-core_2.11"), VersionConstraint("7.2.3")
)
)
)

It goes without saying that a Resolution is immutable, as are all the classes defined in the core module. The resolution process will go on by giving successive Resolutions, until the final one.

start above is only the initial state - it is far from over, as the isDone method on it tells,

start.isDone
// res0: Boolean = false

In order for the resolution to go on, we'll need things from a few repositories,

val repositories = Seq(
coursier.LocalRepositories.ivy2Local,
MavenRepository("https://repo1.maven.org/maven2")
)

The first one, Cache.ivy2Local, is defined in coursier.Cache, itself from the coursier-cache module that we added above. It is an IvyRepository, picking things under ~/.ivy2/local. An IvyRepository is related to the Ivy build tool. This kind of repository involves a so-called pattern, with various properties. These are not of very common use in Scala, although sbt uses them a bit.

The second repository is a MavenRepository. These are simpler than the Ivy repositories. They're the ones we're the most used to in Scala. Common ones like Central like here, or the repositories from Sonatype, are Maven repositories. These originate from the Maven build tool. Unlike the Ivy repositories which involve customisable patterns to point to the underlying metadata and artifacts, the paths of these for Maven repositories all look alike, like for any particular version of the standard library, under paths like this one.

Both IvyRepository and MavenRepository are case classes, so that it's straightforward to specify one's own repositories.

To set credentials for a MavenRepository or IvyRepository, set their authentication field, like

import coursier.core.Authentication

MavenRepository(
"https://nexus.corp.com/content/repositories/releases",
authentication = Some(Authentication("user", "pass"))
)

Now that we have repositories, we're going to mix these with things from the coursier-cache module, for resolution to happen via the cache. We'll create a function of type Seq[(Module, String)] => F[Seq[((Module, String), Either[Seq[String], (Artifact.Source, Project)])]]. Given a sequence of dependencies, designated by their Module (organisation and name in most cases) and version (just a String), it gives either errors (Seq[String]) or metadata ((Artifact.Source, Project)), wrapping the whole in a monad F.

import coursier.cache.Cache

val fetch = ResolutionProcess.fetch0(repositories, Cache.default.fetch)

The monad used by Fetch.from is coursier.util.Task, but the resolution process is not tied to a particular monad - any stack-safe monad would do.

With this fetch method, we can now go on with the resolution. Calling process on start above gives a ResolutionProcess, that drives the resolution. It is loosely inspired by the Process of fs2 (formerly scalaz-stream). It is an immutable structure, that represents the various states the resolution process can be in.

Its method current gives the current Resolution. Calling isDone on the latter says whether the resolution is done or not.

The next method, that expects a fetch method like the one above, gives the "next" state of the resolution process, wrapped in the monad of the fetch method. It allows to do one resolution step.

Lastly, the run method runs the whole resolution until its end. It expects a fetch method too, and will make at most maxIterations steps (50 by default), and return the "final" resolution state, wrapped in the monad of fetch. One should check that the Resolution it returns is done (isDone) - the contrary means that maxIterations were reached, likely signaling an issue, unless the underlying resolution is particularly complex, in which case maxIterations could be increased.

Let's run the whole resolution,

import scala.concurrent.ExecutionContext.Implicits.global

val resolution = start.process.run0(fetch).unsafeRun()

To get additional feedback during the resolution, we can give the Cache.fetch method above a Cache.Logger.

By default, downloads happen in a global fixed thread pool (with 6 threads, allowing for 6 parallel downloads), but you can supply your own thread pool to Cache.fetch.

Now that the resolution is done, we can check for errors in

val errors: Seq[((Module, VersionConstraint), Seq[String])] = resolution.errors0

These would mean that the resolution wasn't able to get metadata about some dependencies.

We can also check for version conflicts, in

val conflicts: Set[Dependency] = resolution.conflicts

which are dependencies whose versions could not be unified.

Then, if all went well, we can fetch and get local copies of the artifacts themselves (the JARs) with

import java.io.File
import coursier.cache.{ArtifactError, Cache}
import coursier.util.{Gather, Task}

val localArtifacts: Seq[Either[ArtifactError, File]] =
Gather[Task].gather(
resolution.artifacts().map(Cache.default.file(_).run)
).unsafeRun()

We're using the Cache.file method, that can also be given a Logger (for more feedback) and a custom thread pool.