We have a Maven 2 project with lots of modules in it. Example:
<modules>
<module>common</module>
<module>foo</module>
<module>data</module>
<module>bar</module>
... more ...
</module>
Let's say the "data" module is time consuming to build and we want to exclude it when the project is build by a CI server. Currently we use two pom.xml files to achieve this. One has all modules in it and the other one has all modules except the ones which can be left out for CI. But that's pretty annoying because sometimes we forget to put a new module into both files.
Is there a solution which doesn't need two separate module lists?
With Maven 3.2.1, you can now use -pl !<module_name>,!<module_name> to exclude certain modules from the reactor build.
See this feature request: https://issues.apache.org/jira/browse/MNG-5230
The easiest might be to use profiles like this:
<project>
...
<modules>
<module>common</module>
<module>foo</module>
<module>bar</module>
<modules>
...
<profiles>
<profile>
<id>expensive-modules-to-build</id>
<modules>
<module>data</module>
</modules>
</profile>
</profiles>
</project>
You should then check out ways you can activate profiles
The projects to build can also be specified on the mvn command line. This would remove the need for a separate pom, but instead you would have to change the CI configuration everytime there is a new module.
-pl,--projects <arg> Comma-delimited list of specified
reactor projects to build instead
of all projects. A project can be
specified by [groupId]:artifactId
or by its relative path.
Maybe a combination of this flag and --also-make-dependents or --also-make would reduce this maintenance burden again.
-am,--also-make If project list is specified, also
build projects required by the
list
-amd,--also-make-dependents If project list is specified, also
build projects that depend on
projects on the list
I assume you want the default build to always build everything, regardless of speed, so that new developers can get started quickly without having to understand lots about the POM. You can use profiles like this:
<modules>
<module>common</module>
<module>foo</module>
<module>bar</module>
</modules>
...
<profiles>
<profile>
<id>expensive-modules-to-build</id>
<activation>
<activeByDefault>true</activeByDefault>
</activation>
<modules>
<module>data</module>
</modules>
</profile>
</profiles>
</project>
The problem with this is that if a developer specifies another profile on the command line, then the expensive-modules-to-build isn't included (unless the developer also specifies it). This makes it complicated to remember which profiles need to be included.
Here is a hacky way around that. Both profiles are always included, because the pom.xml file always exists. So to exclude the expensive modules, you can use -P!full-build on the command line.
<profiles>
<profile>
<id>full-build</id>
<activation>
<file>
<exists>pom.xml</exists>
</file>
</activation>
<modules>
<module>data</module>
</modules>
</profile>
<profile>
<id>short-build</id>
<activation>
<file>
<exists>pom.xml</exists>
</file>
</activation>
<modules>
<module>common</module>
<module>foo</module>
<module>bar</module>
</modules>
</profile>
</profiles>
Another idea: Reactor modules can be nested, so it should be possible to group your fast and slow-building modules into separate poms and then add another aggregator pom containing these two as modules. Your CI Server could then only reference the pom containing the fast building modules.
<artifactId>fast</artifactId>
<modules>
<module>fast-a</module>
<module>fast-b</module>
<module>fast-c</module>
</module>
<artifactId>all</artifactId>
<modules>
<module>fast</module>
<module>slow</module>
</module>
You could be to use maven profiles. In our build environment, we created a profile quick that disables many plugins and test execution.
This is done by
<profile>
<id>quick</id>
<properties>
<skipTests>true</skipTests>
<!-- others... -->
</properties>
<build>
<plugins>
<!-- configuration... -->
</plugins>
</build>
</profile>
And then we invoke maven the following way
mvn groupId:artifactId:goal -P quick
You could maybe disable compilation and other standard plugins in the pom of your module to speed it up.
Not exactly the answer these folks were asking for. My situation was I wanted to deploy only the parent pom. I'm using the spring-boot-thin-layout in a child module. This requires the parent module be deployed into artifactory. I added the following into my project. It enables skipping of install and/or deploy phase.
In my parent pom:
<properties>
<disable.install>true</disable.install>
<disable.deploy>true</disable.deploy>
<enable.deployAtEnd>true</enable.deployAtEnd>
</properties>
<profiles>
<profile>
<id>deploy-parent</id>
<activation>
<activeByDefault>true</activeByDefault>
</activation>
<properties>
<disable.install>true</disable.install>
<disable.deploy>true</disable.deploy>
<deployAtEnd>${enable.deployAtEnd}</deployAtEnd>
</properties>
<build>
<finalName>${project.version}</finalName>
</build>
</profile>
</profiles>
And the in my child pom(s) or any module you don't want deployed with parent:
<properties>
<maven.install.skip>${disable.install}</maven.install.skip>
<maven.deploy.skip>${disable.deploy}</maven.deploy.skip>
<deployAtEnd>${enable.deployAtEnd}</deployAtEnd>
</properties>
So effectively when I run mvn deploy on the parent pom, it will compile all the modules, not run install on anything, and then at the end deploy any module not having <maven.deploy.skip>${disable.deploy}</maven.deploy.skip> in it's properties. So in my case only deploying the parent.
Related
I have two profiles in my pom.xml, dev and stage:
<profiles>
<profile>
<id>dev</id>
<properties>
<hostname>vl-wlp1.hk.oracle.com</hostname>
</properties>
<id>stage</id>
<properties>
<hostname>vl-wcfs.hk.oracle.com</hostname>
</properties>
</profile>
</profiles>
And I'd like use these in my site documentation at:
src/site/apt/index.apt
like so:
Dev Site: ${dev.hostname}
Stage Site: ${stage.hostname}
Can I do that or something else that has the same effect?
Not without a huge hack, no.
If you want to read both property values independently, they will have to be two different properties.
How about a pragmatic solution like this:
<properties>
<dev.host>vl-wlp1.hk.oracle.com</dev.host>
<stage.host>vl-wcfs.hk.oracle.com</stage.host>
</properties>
<profiles>
<profile>
<id>dev</id>
<properties>
<hostname>${dev.host}</hostname>
</properties>
<id>stage</id>
<properties>
<hostname>${stage.host}</hostname>
</properties>
</profile>
</profiles>
src/site/apt/index.apt:
Dev Site: ${dev.host}
Stage Site: ${stage.host}
(The huge hack mentioned above would mean programmatically iterating over the current project's profiles and parsing each profile's properties manually, you could do that in a custom maven plugin or in a Groovy Script using GMaven)
If I have Maven builds set up for an app with profiles set up for different environments (say like prod vs. dev, defining different DB settings and stuff like that) the 'install' goal doesn't seem to make sense, as I don't know which environment got installed into my repo - I've just got com.example.myproject:myapp:0.0.1.
Have I misunderstood something, or are profiles supposed to be used with other goals?
Well, you could use the classifier attribute so that each profile creates a jar with the classifier, i.e. a unique jar for each environment. Here is a code snippet to illustrate this. When run with the dev profile (mvn -P dev install), it creates a jar with -dev classifier, like myapp-dev-0.0.1.jar
<project>
...
<properties>
<env></env>
</properties>
...
<build>
<plugins>
...
<plugin>
<groupId>org.apache.maven.plugins</groupId>
<artifactId>maven-jar-plugin</artifactId>
<configuration>
<classifier>${env}</classifier>
</configuration>
</plugin>
</plugins>
</build>
<profiles>
...
<profile>
<id>dev</id>
<properties>
<env>dev</env>
</properties>
...
</profile>
</profiles>
</project>
You run the usual mvn commands and can select the appropriate profile with -P http://maven.apache.org/guides/introduction/introduction-to-profiles.html So it dependends on which profile you chose, what gets installed in the repository.
Suppose I have a module A:jar, whose runtime and compilation set of dependencies depends on the JDK version. In my example, I have a pre-jdk6-profile for JAXB API: prior to JDK 1.6.0 I need to include jaxb-api-nnn.jar as a compile dependency. This profile is placed to A.pom.
I also have module B:war, which depends on A:jar. I want to be able to activate this profile on a build server to build the JDK 1.5.x deliverable. When I execute Maven with a given profile activated, I get the message:
mvn -Ppre-jdk6-profile -o install
[WARNING]
Profile with id: 'pre-jdk6-profile' has not been activated.
and jaxb-api-nnn.jar is missing in resulting B.war. However if I activate this profile when building from the parent pom.xml, everything is OK. That means the profiles are not inherited from dependencies, and the parent multi-module pom.xml was able to build everything correctly because it seems like all profiles are merged in reactor.
Shifting the profile to parent pom makes things worse, as the dependencies are applied to all other projects (e.g. to C:ear). Are there nice solutions for this task, namely, if any module A depends on module B, then all compile and runtime dependencies which are activated by a profile, are correctly handled?
The profile in project A:jar follows:
<project ...>
<artifactId>A</artifactId>
<packaging>jar</packaging>
...
<parent>
<artifactId>P</artifactId>
...
</parent>
<profiles>
<profile>
<id>pre-jdk6-profile</id>
<activation>
<jdk>(,1.6.0)</jdk>
</activation>
<dependencies>
<dependency>
<groupId>javax.xml.ws</groupId>
<artifactId>jaxws-api</artifactId>
</dependency>
</dependencies>
</profile>
</profiles>
...
</project>
a) In a multi-module build, you should always build from the top pom, never from an individual module. If you want to build only one module, use advanced reactor options (see mvn --help) like this:
mvn -pl mymodule
b) Define and activate the profile in the parent pom, but add the configuration in the child pom.
parent pom.xml
<profiles>
<profile>
<id>pre-jdk-6</id>
<activation>
<jdk>(,1.6.0)</jdk>
</activation>
</profile>
</profiles>
child pom.xml
<profiles>
<profile>
<id>pre-jdk-6</id>
<dependencies>
<dependency>
<groupId>javax.xml.ws</groupId>
<artifactId>jaxws-api</artifactId>
</dependency>
</dependencies>
</profile>
</profiles>
Several notes way after the fact:
When you use -P profName, it activates a profile named 'profName'
After that, it disables all profiles that have an <activation> tag for them. It doesn't matter whether they are activated by the java version, as in the example, or by default or env value or anything.
That means the -P causes any otherwise activated profile to become deactivated.
Solution: Either use <activation><jdk>...</jdk></activation> or use -P but do not use both.
I have a POM which declares web application stuff that is common to my projects. I use this as the parent for all web applications.
Is it possible to activate a profile only when the packaging is war? I have tried the property approach, but that doesn't work (as it isn't a system/environment property).
Since this fails the build, I can simply disable that profile when installing the POM, but I'd like it to be more intelligent on its own.
Walter
You can simply check the existence of src/main/webapp. Each web application that uses the Maven standard directory layout should contain this folder. So you avoid unnecessary dummy files.
<profile>
<id>custom-profile-eclipse-project-generation-webapp</id>
<activation>
<file>
<exists>${basedir}/src/main/webapp</exists>
</file>
</activation>
<build>
</build>
</profile>
More precise you can also check for the the existence of ${basedir}/src/main/webapp/WEB-INF/web.xml. That should definitively identify a war-project.
For myself I use this configuration in my common super-pom to configure the maven-eclipse-plugin for different project types. Thats very handy to get homogenous eclipse-configurations over the same project type in our organization, especially when developers straightforwardly run eclipse:eclipse on multi-module-projects.
I know this isn't answering your question directly, but the usual workaround for problems like this is to just use specialization (as with classes).
So you have your MasterPom with all common behavior.
MasterWarPom that extends MasterPom (is it's parent), and put any 'packing is war' specializations in here.
Likewise you could have MasterJarPom, etc ...
That way the differences are split out nicely.
There's no clean way to do that, the parent module has no way of knowing the child's packaging. (Non-clean solutions would involve creating a plugin that parses the child module's pom etc.)
The best I've been able to come up with for these sorts scenarios has been to use a file-based activation trigger.
eg my parent pom has
<profile>
<id>maven-war-project</id>
<activation>
<file><!-- add a file named .maven-war-project-marker to webapp projects to activate this profile -->
<exists>${basedir}/.maven-war-project-marker</exists>
</file>
</activation>
<build>
<plugins>
<!-- configuration for webapp plugins here -->
</plugins>
</build>
and webapp projects that inherit from this parent contain a file named
'.maven-war-project-marker'
that activates the profile
This looks pretty obtuse but works fairly reliably whereas
- using property-activation is unreliable if a different person or system does the build,
- inheriting from type-specific parents became a bit cumbersome for me as the grandparent-pom changes version relatively frequently as it is used to define 'standard' or preferred versions of common dependencies which in turn required corresponding releases of all of the type-specific parents with no change other than the grandparent version
Try in this way ?
mvn package -Dmaven.test.skip=true -Dwar
<project ×××××>
<modelVersion>4.0.0</modelVersion>
<parent>
<groupId>××××</groupId>
<artifactId>×××××</artifactId>
<version>×××××</version>
<relativePath>../../</relativePath>
</parent>
<artifactId>×××××</artifactId>
<name>${project.artifactId}-${project.version}</name>
<description>${project.artifactId}-${project.version}</description>
<properties>
<packaging.type>jar</packaging.type>
</properties>
<profiles>
<profile>
<activation>
<property>
<name>war</name>
</property>
</activation>
<properties>
<packaging.type>war</packaging.type>
</properties>
<build>
<finalName>ROOT</finalName>
</build>
</profile>
</profiles>
<packaging>${packaging.type}</packaging>
<dependencies>
<dependency>
... ...
</dependency>
... ...
</dependencies>
I am using two profiles: development and production.
Development should be active on default; production should be used when I am releasing.
In my pom.xml I have:
[...]
<plugin>
<artifactId>maven-release-plugin</artifactId>
<version>2.0-beta-9</version>
<configuration>
<useReleaseProfile>false</useReleaseProfile>
<goals>deploy</goals>
<arguments>-Pproduction</arguments>
</configuration>
</plugin>
[...]
<profiles>
<profile>
<id>production</id>
<properties>
<profile.name>production</profile.name>
</properties>
[...]
</profile>
<profile>
<id>development</id>
<activation>
<activeByDefault>true</activeByDefault>
</activation>
<properties>
<profile.name>development</profile.name>
</properties>
[...]
</profile>
[...]
It just does not work.
useReleaseProfiles doesn't work either:
http://jira.codehaus.org/browse/MRELEASE-459
The development profile should be always active but not when running mvn release:perform.
How do you achieve this?
[UPDATE]:
I have seen with the debug flag that my production profile is used, but development profile is used too, because it is activeByDefault. This cant be overridden by the releaseProfile argument. It would be nice to force the release plugin to use only the "production" profile.
The maven-release-plugin documentation encourages using the releaseProfiles configuration parameter to automatically invoke profiles during the release process.
This is a better approach than manually invoking release profiles from the command-line. One reason, is because the profiles used in the release will be documented in the pom.xml and stored with the tagged code. This makes the build process easier to understand and easier to repeat later, exactly the same way the project was originally released.
If using maven-release-plugin older than 2.4 see this bug preventing use of the above mentioned parameter.
Be aware that in case of a multi-module project you'll have to put the "releaseProfiles" configuration in the root pom! See also this issue for more information about that.
I think you should simply activate your profiles through a property.
<profiles>
<profile>
<id>production</id>
<activation>
<property>
<name>build</name>
<value>release</value>
</property>
</activation>
[...]
</profile>
<profile>
<id>development</id>
<activation>
<property>
<name>build</name>
<value>develop</value>
</property>
</activation>
[...]
</profile>
<profiles>
Do your builds by executing something like this
mvn -Dbuild=develop package
mvn -Dbuild=develop test
mvn -Dbuild=release release:prepare
mvn -Dbuild=release release:perform
If you check "Introduction to Build Profiles", "Deactivating a profile":
mvn groupId:artifactId:goal -P !profile-1,!profile-2
I guess you could use this to deactivate your default profile?
This is a very old post but I came across this issue quite recently. The releaseProfile only worked for me when I set the releaseProfiles to profile called release. Any other profile gives same error.
Sample code:
<build>
<plugins>
<plugin>
<groupId>org.apache.maven.plugins</groupId>
<artifactId>maven-release-plugin</artifactId>
<version>2.5.3</version>
<configuration>
<tagNameFormat>#{project.artifactId}-#{project.version}</tagNameFormat>
<autoVersionSubmodules>true</autoVersionSubmodules>
<releaseProfiles>release</releaseProfiles>
<allowTimestampedSnapshots>true</allowTimestampedSnapshots>
</configuration>
</plugin>
</plugins>
</build>
<profiles>
<profile>
<id>release</id>
<properties>
<connectionUrl>${scm-base}/tags/${project.artifactId}-${project.version}</connectionUrl>
</properties>
</profile>
</profiles>