Overriding transitive dependency versions

Gradle resolves any dependency version conflicts by selecting the latest version found in the dependency graph. Some projects might need to divert from the default behavior and enforce an earlier version of a dependency e.g. if the source code of the project depends on an older API of a dependency than some of the external libraries.

Forcing a version of a dependency requires a conscious decision. Changing the version of a transitive dependency might lead to runtime errors if external libraries do not properly function without them. Consider upgrading your source code to use a newer version of the library as an alternative approach.

In general, forcing dependencies is done to downgrade a dependency. There might be different use cases for downgrading:

  • a bug was discovered in the latest release

  • your code depends on a lower version which is not binary compatible

  • your code doesn’t depend on the code paths which need a higher version of a dependency

In all situations, this is best expressed saying that your code strictly depends on a version of a transitive. Using strict versions, you will effectively depend on the version you declare, even if a transitive dependency says otherwise.

Strict dependencies are to some extent similar to Maven’s nearest first strategy, but there are subtle differences:

Let’s say a project uses the HttpClient library for performing HTTP calls. HttpClient pulls in Commons Codec as transitive dependency with version 1.10. However, the production source code of the project requires an API from Commons Codec 1.9 which is not available in 1.10 anymore. A dependency version can be enforced by declaring it as strict it in the build script:

build.gradle.kts
dependencies {
    implementation("org.apache.httpcomponents:httpclient:4.5.4")
    implementation("commons-codec:commons-codec") {
        version {
            strictly("1.9")
        }
    }
}
build.gradle
dependencies {
    implementation 'org.apache.httpcomponents:httpclient:4.5.4'
    implementation('commons-codec:commons-codec') {
        version {
            strictly '1.9'
        }
    }
}

Consequences of using strict versions

Using a strict version must be carefully considered, in particular by library authors. As the producer, a strict version will effectively behave like a force: the version declaration takes precedence over whatever is found in the transitive dependency graph. In particular, a strict version will override any other strict version on the same module found transitively.

However, for consumers, strict versions are still considered globally during graph resolution and may trigger an error if the consumer disagrees.

For example, imagine that your project B strictly depends on C:1.0. Now, a consumer, A, depends on both B and C:1.1.

Then this would trigger a resolution error because A says it needs C:1.1 but B, within its subgraph, strictly needs 1.0. This means that if you choose a single version in a strict constraint, then the version can no longer be upgraded, unless the consumer also sets a strict version constraint on the same module.

In the example above, A would have to say it strictly depends on 1.1.

For this reason, a good practice is that if you use strict versions, you should express them in terms of ranges and a preferred version within this range. For example, B might say, instead of strictly 1.0, that it strictly depends on the [1.0, 2.0[ range, but prefers 1.0. Then if a consumer chooses 1.1 (or any other version in the range), the build will no longer fail (constraints are resolved).

Forced dependencies vs strict dependencies

If the project requires a specific version of a dependency at the configuration-level this can be achieved by calling the method ResolutionStrategy.force(java.lang.Object[]).

build.gradle.kts
configurations {
    "compileClasspath" {
        resolutionStrategy.force("commons-codec:commons-codec:1.9")
    }
}

dependencies {
    implementation("org.apache.httpcomponents:httpclient:4.5.4")
}
build.gradle
configurations {
    compileClasspath {
        resolutionStrategy.force 'commons-codec:commons-codec:1.9'
    }
}

dependencies {
    implementation 'org.apache.httpcomponents:httpclient:4.5.4'
}

Excluding transitive dependencies

While the previous section showed how you can enforce a certain version of a transitive dependency, this section covers excludes as a way to remove a transitive dependency completely.

Similar to forcing a version of a dependency, excluding a dependency completely requires a conscious decision. Excluding a transitive dependency might lead to runtime errors if external libraries do not properly function without them. If you use excludes, make sure that you do not utilise any code path requiring the excluded dependency by sufficient test coverage.

Transitive dependencies can be excluded on the level of a declared dependency. Exclusions are spelled out as a key/value pair via the attributes group and/or module as shown in the example below. For more information, refer to ModuleDependency.exclude(java.util.Map).

build.gradle.kts
dependencies {
    implementation("commons-beanutils:commons-beanutils:1.9.4") {
        exclude(group = "commons-collections", module = "commons-collections")
    }
}
build.gradle
dependencies {
    implementation('commons-beanutils:commons-beanutils:1.9.4') {
        exclude group: 'commons-collections', module: 'commons-collections'
    }
}

In this example, we add a dependency to commons-beanutils but exclude the transitive dependency commons-collections. In our code, shown below, we only use one method from the beanutils library, PropertyUtils.setSimpleProperty(). Using this method for existing setters does not require any functionality from commons-collections as we verified through test coverage.

src/main/java/Main.java
import org.apache.commons.beanutils.PropertyUtils;

public class Main {
    public static void main(String[] args) throws Exception {
        Object person = new Person();
        PropertyUtils.setSimpleProperty(person, "name", "Bart Simpson");
        PropertyUtils.setSimpleProperty(person, "age", 38);
    }
}

Effectively, we are expressing that we only use a subset of the library, which does not require the commons-collection library. This can be seen as implicitly defining a feature variant that has not been explicitly declared by commons-beanutils itself. However, the risk of breaking an untested code path increased by doing this.

For example, here we use the setSimpleProperty() method to modify properties defined by setters in the Person class, which works fine. If we would attempt to set a property not existing on the class, we should get an error like Unknown property on class Person. However, because the error handling path uses a class from commons-collections, the error we now get is NoClassDefFoundError: org/apache/commons/collections/FastHashMap. So if our code would be more dynamic, and we would forget to cover the error case sufficiently, consumers of our library might be confronted with unexpected errors.

This is only an example to illustrate potential pitfalls. In practice, larger libraries or frameworks can bring in a huge set of dependencies. If those libraries fail to declare features separately and can only be consumed in a "all or nothing" fashion, excludes can be a valid method to reduce the library to the feature set actually required.

On the upside, Gradle’s exclude handling is, in contrast to Maven, taking the whole dependency graph into account. So if there are multiple dependencies on a library, excludes are only exercised if all dependencies agree on them. For example, if we add opencsv as another dependency to our project above, which also depends on commons-beanutils, commons-collection is no longer excluded as opencsv itself does not exclude it.

build.gradle.kts
dependencies {
    implementation("commons-beanutils:commons-beanutils:1.9.4") {
        exclude(group = "commons-collections", module = "commons-collections")
    }
    implementation("com.opencsv:opencsv:4.6") // depends on 'commons-beanutils' without exclude and brings back 'commons-collections'
}
build.gradle
dependencies {
    implementation('commons-beanutils:commons-beanutils:1.9.4') {
        exclude group: 'commons-collections', module: 'commons-collections'
    }
    implementation 'com.opencsv:opencsv:4.6' // depends on 'commons-beanutils' without exclude and brings back 'commons-collections'
}

If we still want to have commons-collections excluded, because our combined usage of commons-beanutils and opencsv does not need it, we need to exclude it from the transitive dependencies of opencsv as well.

build.gradle.kts
dependencies {
    implementation("commons-beanutils:commons-beanutils:1.9.4") {
        exclude(group = "commons-collections", module = "commons-collections")
    }
    implementation("com.opencsv:opencsv:4.6") {
        exclude(group = "commons-collections", module = "commons-collections")
    }
}
build.gradle
dependencies {
    implementation('commons-beanutils:commons-beanutils:1.9.4') {
        exclude group: 'commons-collections', module: 'commons-collections'
    }
    implementation('com.opencsv:opencsv:4.6') {
        exclude group: 'commons-collections', module: 'commons-collections'
    }
}

Historically, excludes were also used as a band aid to fix other issues not supported by some dependency management systems. Gradle however, offers a variety of features that might be better suited to solve a certain use case. You may consider to look into the following features:

  • Update or downgrade dependency versions: If versions of dependencies clash, it is usually better to adjust the version through a dependency constraint, instead of attempting to exclude the dependency with the undesired version.

  • Component Metadata Rules: If a library’s metadata is clearly wrong, for example if it includes a compile time dependency which is never needed at compile time, a possible solution is to remove the dependency in a component metadata rule. By this, you tell Gradle that a dependency between two modules is never needed — i.e. the metadata was wrong — and therefore should never be considered. If you are developing a library, you have to be aware that this information is not published, and so sometimes an exclude can be the better alternative.

  • Resolving mutually exclusive dependency conflicts: Another situation that you often see solved by excludes is that two dependencies cannot be used together because they represent two implementations of the same thing (the same capability). Some popular examples are clashing logging API implementations (like log4j and log4j-over-slf4j) or modules that have different coordinates in different versions (like com.google.collections and guava). In these cases, if this information is not known to Gradle, it is recommended to add the missing capability information via component metadata rules as described in the declaring component capabilities section. Even if you are developing a library, and your consumers will have to deal with resolving the conflict again, it is often the right solution to leave the decision to the final consumers of libraries. I.e. you as a library author should not have to decide which logging implementation your consumers use in the end.