Using OSGi and having trouble to get all the dependencies as proper OSGi bundles? Even worse, sometimes you need to adapt bundles due to class loading issues or make an annoying dependency optional?
bnd-platform can help you solve that problem - it builds OSGi bundles and even Eclipse Update Sites from existing JARs, for instance retrieved from Maven repositories together with transitive dependencies. If needed you can adapt the creation of bundles via general or individual configuration options. bnd-platform is a Gradle plugin and uses bnd to create bundles and Eclipse for the creation of p2 repositories.
For a quick start, check out the (outdated) sample project on GitHub or use this minimal example as a template for your build.
What bnd-platform can do:
The simplest way to apply the plugin to your Gradle build is using the latest release hosted on Maven Central:
buildscript {
repositories {
mavenCentral()
}
dependencies {
classpath 'org.standardout:bnd-platform:3.0.1'
}
}
apply plugin: 'org.standardout.bnd-platform'
Alternatives are including the repository content in the buildSrc folder as done in the sample project or by installing the plugin to your local Maven repository using gradlew install
and adding it as dependency to your build script via mavenLocal()
repository.
Snapshot versions are available in the related sonatype repository:
buildscript {
repositories {
mavenCentral()
maven {
url 'https://oss.sonatype.org/content/repositories/snapshots/'
}
}
dependencies {
classpath 'org.standardout:bnd-platform:3.1.0-SNAPSHOT'
}
}
apply plugin: 'org.standardout.bnd-platform'
The platform plugin comes with several Gradle tasks - the following are the main tasks and build upon each other:
In addition, the clean task deletes all previously created bundles or update site artifacts. Usually you will want to clean the created bundles when building an update site, e.g. gradle clean updateSite
.
Be aware that for building the p2 repository Eclipse is used. If no path to a local Eclipse installation is configured (see the settings section later on) the plugin will by default download Eclipse Indigo and use it for that purpose.
bnd-platform adds a configuration named bndplatform to a Gradle build. Prior to version 3 the configuration was named platform, but this clashes with a platform method that was introduced in Gradle 5.
You can add dependencies to the bndplatform configuration and configure them like you would with any other Gradle build - for example:
// add Maven Central so the dependency can be resolved
repositories {
mavenCentral()
}
dependencies {
// add pegdown as dependency to the platform configuration
bndplatform 'org.pegdown:pegdown:1.4.2'
}
That's it - if you combine the previous code snippet with this one you have your first bnd-platform build script. A call gradle updateSite
would create a p2 repository containing bundles for the pegdown library and its dependencies.
Please see the Gradle documentation on basic and advanced dependency management for more details on the dependency configration and advanced issues like resolving version conflicts or dealing with transitive dependencies.
As an alternative to the conventional dependency declaration you can use the platform extension to add a dependency using the bundle keyword:
platform {
// add pegdown as dependency to the platform configuration
bundle 'org.pegdown:pegdown:1.4.2'
}
Both notations support adapting the dependency configuration as supported by Gradle, e.g. excluding specific transitive dependencies. However, adapting the OSGi bundle creation is only possible with the second notation, as it supports an additional bnd configuration.
The bnd library is used as a tool to create OSGi bundles from JARs. bnd analyzes the class files in a JAR to determine packages to import and export. bnd also allows to configure its behavior through a set of header instructions that resemble the OSGi manifest headers (see the bnd website for a detailed description).
The bundle configuration can be applied when adding a dependency:
platform {
bundle(group: 'net.sf.ehcache', name: 'ehcache-core', version:'2.6.6') {
bnd {
// make hibernate packages optional
optionalImport 'org.hibernate', 'org.hibernate.*'
}
}
}
Or independently - just registering the configuration without adding a dependency. The configuration is only applied if the dependency is either added directly or as a transitive dependency at another point in the script.
platform {
bnd(group: 'net.sf.ehcache', name: 'ehcache-core') {
// make hibernate packages optional
optionalImport 'org.hibernate', 'org.hibernate.*'
}
}
Note that in the example above the version was omitted - the configuration applies to any dependency matching the group and name, regardless of its version.
The configuration options inside the bnd call are sketched below:
platform {
bnd(<DependencyNotation>) {
// override/set the symbolic name
symbolicName = <SymbolicNameString>
// override/set the bundle name
bundleName = <BundleNameString>
// override/set bundle version
version = <VersionString>
// generic bnd header instruction
instruction <Header>, <Instruction>
// adapt the Import-Package instruction to import the given packages optionally
optionalImport <Package1>, <Package2>, ...
// adapt the Import-Package instruction to add the given package instruction
prependImport <PackageInstruction1>, <PackageInstruction2>, ...
// override the default behavior, if a (generated) qualifier should be added to
// the version of wrapped bundles (see plugin settings)
addQualifier = true | false
}
}
You can enable auto-determining versions for package imports by enabling the determineImportVersions
plugin setting. For each bundle to be created from a JAR retrieved via Maven/Ivy, their direct dependencies are analysed in turn and package imports are determined by the packages present there and the version of the dependency modules. This works for most cases, but is not as good as if the information would be determined based on the packages exported by the dependencies. What comes as a bonus is that for packages that are not found in the direct dependencies the imports are made optional automatically.
A default strategy defines how the versions are represented for the Import-Package instructions, i.e. what lower and upper bounds are allowed for an imported package. Pre-defined strategies that can be used are:
Set the version strategy for the whole platform like this:
platform {
determineImportVersions = true
importVersionStrategy = MINIMUM
}
You can also adapt the configuration for specific dependencies, or define a custom version strategy using a Closure. See the definitions of the pre-defined strategies in the sources for PlatformPluginExtension
for more information on how this can be done. The following example demonstrates both:
platform {
imports(group: 'com.google.inject', name: 'guice') {
versionStrategy = {
// guice uses a strange versioning scheme for its package exports
// e.g. version 2.0 of exports packages with version 1.2, version 3.0 with 1.3 etc.
"[1.${it.major},1.${it.major + 1})"
}
}
}
The above example influences the package imports for the packages provided by guice for all bundles that have guice as their dependency. This is needed in this case as guice already is provided as OSGi bundle with the exported package versions differing from the module version - and because bnd-platform currently only uses the information of the module version and applies it to all imports instead of using available package export information. This might be improved in the future if there is need.
bnd-platform will by default leave JARs that are recognized as bundles (meaning they have a Bundle-SymbolicName header already defined) as they are, except to eventually added Bundle-License and Bundle-Vendor headers if not yet present. If an existing bundle is wrapped because a bundle configuration applies to it, the configration from the bundle manifest applies as default configuration.
To other JARs the global default configuration applies, which exports packages with the dependency's version number and imports any identified packages as mandatory import. You can override or extend the global default configuration by adding a bnd configuration without giving a dependency notation, for example:
platform {
bnd {
// make the package junit.framework an optional import
// for all JARs that were not bundles already
optionalImport 'junit.framework'
}
}
But be careful what you put into the default configuration - setting the symbolicName or version here will not be seen as error, but does not make any sense and may lead to unpredicatable behavior (as there can't be two bundles with the same symbolic name and version).
A bundle configuration that is more concrete will always override/extend a more general configuration. A configuration applied to a dependency group takes precedence over the default configuration, while a configuration specified for a combination of group and name in turn takes precedence over a group configuration. If configurations are defined on the same level, the configuration that is defined later in the script will override/extend a previous one.
Please note that in addition, in the combined configurations for a bundle, all assignments (e.g. version = '1.0.0'
) take precedence over the method calls like instruction
and optionalImport
. This allows for instance to both override the version of a bundle in a concrete configuration and making use of it in a more general configuration:
platform {
bnd(group: 'org.standardout') {
// packages with version number (uses the version provided further below)
instruction 'Export-Package', "org.*;version=$version"
}
bnd(group: 'org.standardout', name: 'bnd-platform', version: '0.1') {
// override the version
version = '0.1.0.RELEASE'
}
}
Starting with version 0.3 it is possible to override all bundle configurations at once. This also applies to dependencies that already are an OSGi bundle. You can use any instructions you would use inside bnd to apply them to all dependencies in a call to override:
platform {
override {
// JUnit optional everywhere - so we can exclude it from products
optionalImport 'junit.framework.*', 'org.junit.*'
}
}
Many third party libraries have optional dependencies being specified by using <optional>true</optional>
in a pom.xml file.
For example the Retrofit library has android as optional dependency:
<dependency>
<groupId>com.google.android</groupId>
<artifactId>android</artifactId>
<optional>true</optional>
</dependency>
Since it is unlikely to happen that you want an android dependency in your OSGi application it can be marked as optional:
plugin('com.squareup.retrofit2:retrofit:2.4.0'){
bnd {
optionalImport 'android.os'
optionalImport 'android.net'
}
}
Unfortunately the information about the <optional>true</optional>
instruction is lost during a Gradle build because Gradle's dependency management does not support it yet. (FYI https://docs.gradle.org/4.6/release-notes.html#support-for-optional-dependencies-in-pom-consumption)
So the bnd-platform plug-in cannot add the optionalImport instructions automatically, yet.
For the Retrofit library it is fairly easy to write the optionalImport instructions like above, but there are other libraries, which have plenty of these <optional>true</optional>
instruction.
In order to generate the list of potential optional imports a potentialOptionalImports task has been created. This task creates a potentialOptionalImports.txt file, which lists potential optional imports of each and every bundle, which is created during a build.
Even though setting up an extensive platform of OSGi bundles can be done quite fast using bnd-platform, in many cases additional configuration is necessary. It comes naturally that it should be possible to reuse and share those configurations.
With Gradle you can use apply from: 'someScript.gradle'
to include other build scripts. In those you can define dependencies, bnd configuration or remote repositories like you would do in the main build script.
An alternative is using the gradle-include-plugin which allows you to include specific methods from an external script and thus provide parameters to the include. In context of bnd-platform it often makes sense to provide a version number as parameter. See the sample project for some nice examples, e.g. the logging or geotools platform modules defined there. Using the include plugin these modules are applied to the sample build like this:
include {
from('modules/logging.groovy') {
slf4jAndLogback '1.7.2', '1.0.10' // slf4j and logback with given versions
}
from('modules/geotools.groovy') {
geotools() // include geotools with default modules and version
}
}
We have created a repository on GitHub to collect the platform modules and configurations we use for our projects and you are welcome to fork and contribute: shared-platform. The repository is designed to be used with the gradle-include-plugin and to enable the sharing of configurations without imposing them on you - just include the configuration that makes sense for you and augment it with your own.
You can combine dependencies to an Eclipse feature. The feature will contain the dependencies, as well as their transitive dependencies. You can use the features for more fine-grained control in you target platform or in a feature based Eclipse product. The platform feature contains all features you define. If fetching sources is enabled, a matching source feature will be created for each feature.
A feature includes all plugins (bundles) defined in its context, here an example:
platform {
// define a feature
feature(id: 'platform.restclient', name: 'REST client dependencies', version: '1.0.0') {
// define what's in the feature
plugin 'org.codehaus.groovy.modules.http-builder:http-builder:0.6', {
// exclude this transitive dependency
exclude group: 'net.sourceforge.nekohtml', module: 'nekohtml'
}
plugin 'commons-io:commons-io:2.4'
// include a feature that is defined elsewhere given its ID
includes << 'platform.geotools'
}
}
Providing a feature ID is mandatory, but name and version may be omitted (the version defaults to the platform feature version). plugin
and bundle
can be used synonymously inside the feature
block, but when using bundle
you may experience problems when used in inner closures.
You can easily add local JARs to the platform. If the JAR is not an OSGi bundle yet, you have add it on its own and at least provide symbolicName and version:
platform {
bundle file('someLibrary.jar'), {
bnd {
version = '1.0.0' // mandatory
symbolicName = 'com.example.library.some' // mandatory
bundleName = 'Some Library'
instruction 'Export-Package', "com.example.library.some.*;version=$version"
}
}
// depends on groovy
bundle 'org.codehaus.groovy:groovy:1.8.5'
}
As in the example above, you should make sure to add additional dependencies that might be needed by the JAR. Please note that for a JAR filename.jar
sources provided in a filename-sources.jar
will be wrapped automatically in a corresponding source bundle.
JARs that are already OSGi bundles you can include en masse, and without the need for additional configuration, for example:
platform {
// all bundles in a directory
bundle fileTree(dir: 'lib') {
include '*.jar'
}
// specific bundles
bundle files('someBundle.jar', 'someOtherBundle.jar')
}
When resolving the platform configuration with bnd-platform, Gradle will only include one version for each dependency. This is intentional as even though OSGi is designed to deal with issues such as multiple versions of a bundle, often it will lead to problems - namely package uses conflicts (please note that you can get package uses conflicts even with only one version of each bundle, as some may conflict with packages provided by the system bundle).
However, if there is the need to have multiple versions of a bundle, these are your options:
Following is an example using the platformaux configuration:
dependencies {
// need old version (pre 4) of asm for some bundles
platformaux 'asm:asm:3.3.1'
}
Sometimes it is necessary to create a bundle out of multiple JARs, most often due to class loading issues. The Geotools library is a famous example of that. It uses Java SPI as extension mechanism, which does only recognize extensions from the same class loader. One way suggested to cope with this is creating a monolithic bundle that includes all the needed geotools modules (which I prefer because with Geotools otherwise you have different bundles partly exporting the same packages). Doing this with bnd-platform is quite easy:
platform {
def geotoolsVersion = '10.4'
// define the merged bundle
merge {
// the match closure is applied to all dependencies/artifacts encountered
// if true, an artifact is included in the bundle
match {
// merge all artifacts in org.geotools group, but not gt-opengis
it.group == 'org.geotools' && it.name != 'gt-opengis'
}
bnd {
symbolicName = 'org.geotools'
bundleName = 'Geotools'
version = geotoolsVersion
instruction 'Export-Package', "org.geotools.*;version=$version"
instruction 'Private-Package', '*'
}
}
// add geotools modules as dependencies
bundle "org.geotools:gt-shapefile:$geotoolsVersion"
// etc.
}
Providing the symbolicName and version as part of the bnd configuration is mandatory for merged bundles, as the information from the original JARs' manifests is discarded. Please note that the example above misses the Maven repositories needed to actually retrieve those artifacts, see the sample project for a more complete example.
If you use match { ... }
to merge bundles, it is called for each artifact. The artifact can be accessed via the variable it. An artifact has the following properties that can be useful to check against in a match:
As alternative to match or in combination with it you can add bundles to merge via bundle or include. The syntax is the same as when adding dependencies. However, using include you just specify an artifact to be included if it is a dependency defined somewhere else, it does not add it as dependency.
platform {
merge {
bundle 'someGroup:someArtifact:1.0.0' // also added as dependency
include group: 'someGroup', name: 'someOtherArtifact' // not added as dependency
bnd {
...
}
}
}
You can supply parameters to merge, currently those are:
META-INF/services
defining extensions via SPI (default: true)You can specify them as named parameters, e.g.:
platform {
merge(failOnDuplicate: false, collectServices: true) {
...
}
}
Via the platform extension there are several settings you can provide:
new File(buildDir, 'updatesite')
)new File(buildDir, 'updatesite.zip')
)false
)<feature-id>_versions.json
, that includes information on the versions of the feature available in the p2 repository (default: false
)null
)new File(buildDir, 'eclipse-downloads')
)false
)MAJOR
)true
)true
)ADLER32
)'i'
)YEAR
, MONTH
, DAY
, MINUTE
, SECOND
, MILLISECOND
(default: MONTH
)false
)true
)false
)true
)For example:
platform {
fetchSources = false
featureVersion = '3.1.0'
eclipseHome = new File('/opt/eclipse')
eclipseMirror = 'http://myeclipsedownload.com/eclipse.tar.gz'
}
Releases are currently not done by the CI but manually.
Steps are:
./gradlew publishToSonatype closeAndReleaseSonatypeStagingRepository
)This software is licensed under the Apache License, Version 2.0.
Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.