使用常春藤解决具有相同依赖关系的两个工件

Question

I'm having trouble resolving kafka dependencies using ivy.

I am trying to create a kafka server in a junit test.

TestUtils.createServer(kafkaConfig, Time.SYSTEM);

This requires kafka_2.11-0.10.2.1.jar , kafka-clients-0.10.2.1-test.jar , and kafka_2.11-0.2.1-test.jar to be on the test class path since kafka.utils.TestUtils is in the *-test.jar . The problem is how can I resolve the *-test.jar using ivy and ant? In the ivy.xml I have:

<configurations defaultconfigmapping="default>
    <conf name="compile"/>
    <conf name="test" extends="compile"/>
    <conf name="master"/>
    <conf name="sources"/>
    <conf name="javadoc"/>
    <conf name="runtime" extends="compile"/>
    <conf name="default" extends="master,runtime"/>
</configurations>
...
    <dependency org="org.apache.kafka" name="kafka-clients" rev="0.10.2.1" conf="compile->default"/>
    <dependency org="org.apache.kafka" name="kafka_2.11" rev="0.10.2.1" conf="test->default">
        <artifact name="kafka_2.11" ext="jar"/>
        <artifact name="kafka_2.11" e:classifier="test" ext="jar"/>
    </dependency>
    <dependency org="org.apache.kafka" name="kafka-clients" rev="0.10.2.1" conf="test->default">
        <artifact name="kafka-clients" ext="jar"/>
        <artifact name="kafka-clients" e:classifier="test" ext="jar"/>
    </dependency>

The build has something similar this

<ivy:resolve log="quiet" conf="compile" transitive="false"/>
<ivy:cachepth pathid="test-classpath" conf="test" log="quiet"/>

This results in NoClassDefFoundError org/apache/kafka/test/TestUtils .

I noticed that conf is set to "compile" but when I add test to the list I get other errors for missing classes.

Answer 1

Don't set the transitive dependency resolution to false. You'll lose one of the major benefits of using ivy.

Instead create configuration mappings in each of your dependencies.

Example

The following contrived example saves the dependencies in separate directories to illustrate which jars are on the desired classpath. As you can see the kafka client jars are downloaded along with the jars they depend on (the transitive dependencies).

├── build.xml
├── ivy.xml
└── target
    └── lib
        ├── compile
        │   ├── kafka-clients-0.10.2.1.jar
        │   ├── lz4-1.3.0.jar
        │   ├── slf4j-api-1.7.21.jar
        │   └── snappy-java-1.1.2.6.jar
        └── test
            ├── kafka-clients-0.10.2.1.jar
            ├── kafka-clients-test-0.10.2.1.jar
            ├── lz4-1.3.0.jar
            ├── slf4j-api-1.7.21.jar
            └── snappy-java-1.1.2.6.jar

build.xml

The retrieve task is used to save the files under the "target/lib" directory. The ivycachepath task is used to create an ANT cache with the appropriate files.

<project name="demo" default="build" xmlns:ivy="antlib:org.apache.ivy.ant">

    <target name="build">
        <ivy:retrieve pattern="${build.dir}/lib/[conf]/[artifact](-[classifier])-[revision].[ext]"/>

        <ivy:cachepath pathid="compile-classpath" conf="compile"/>
        <ivy:cachepath pathid="test-classpath" conf="test"/>
    </target>

</project>

ivy.xml

The magic is in the configuration mappings "compile->default" or "test->default".

<ivy-module version="2.0" xmlns:e="http://ant.apache.org/ivy/extra">
    <info organisation="com.myspotontheweb" module="demo"/>

    <configurations>
        <conf name="compile" description="Required to compile application"/>
        <conf name="test" description="Additional test dependencies" extends="compile"/>
    </configurations>

    <dependencies>
        <!-- compile dependencies -->
        <dependency org="org.apache.kafka" name="kafka-clients" rev="0.10.2.1" conf="compile->default"/>

        <!-- test dependencies -->
        <dependency org="org.apache.kafka" name="kafka-clients" rev="0.10.2.1" conf="test->default">
            <artifact name="kafka-clients"/>
            <artifact name="kafka-clients" e:classifier="test"/>
        </dependency>
    </dependencies>

</ivy-module>