Gradle 2.8 delivers performance improvements and a collection of general fixes and enhancements.
Large builds with many source files should see major improvements in incremental build speed. This release also brings faster build script compilation, faster source compilation when using continuous build, and general performance improvements that apply to all builds.
Building upon the recent releases, this release brings more improvements to the Gradle TestKit. It is now easier to inject plugins under test into test builds.
Work continues on the new managed model. This release brings richer modelling capabilities along with interoperability improvements when dynamically depending on rule based tasks.
A Gradle release would not be complete without contributions from the wonderful Gradle community. This release provides support for file name encoding in Zip files, support for more PMD features and other fixes from community pull requests.
Here are the new features introduced in this Gradle release.
One of Gradle's key features is the ability to perform “incremental builds”. This allows Gradle to avoid doing redundant work, by detecting that it can safely reuse files created by a previous build. For example, the “class files” created by the Java compiler can be reused if there were no changes to source code or compiler settings since the previous build. This is a generic capability available to all kinds of work performed by Gradle.
This feature relies on tracking checksums of files in order to detect changes. In this Gradle release, improvements have been made to the management of file checksums, resulting in significantly improved build times when the build is mostly up to date (i.e. many previously created files were able to be reused).
Highly incremental builds of projects with greater than 140,000 files have been measured at 35-50% faster with Gradle 2.8. Very large projects (greater than 400,000 files) are also significantly faster again, if there is ample memory available to the build process (see “The Build Environment” in the Gradle User Guide for how to control memory allocation). Smaller projects also benefit from these changes.
No build script or configuration changes, beyond upgrading to Gradle 2.8, are required to leverage these performance improvements.
Build script compilation times have been reduced by up to 30% in this version of Gradle.
This improvement is noticeable when building a project for the first time with a certain version of Gradle, or after making changes to build scripts. This is due to Gradle caching the compiled form of the build scripts.
The reduction in compilation time per script is dependent on the size and complexity of script. Additionally, the reduction for the entire build is dependent on the number of build scripts that need to be compiled.
In many cases Gradle will spawn a daemon process to host a compiler tool. This is done for performance reasons, as well as to accommodate special heap size settings, classpath configurations, etc. A compiler daemon is started on first use, and stopped at the end of the build.
With Gradle 2.8, compiler daemons are kept running throughout the lifetime of a continuous build session and will only be stopped when the continuous build is cancelled. This improves the performance of continuous builds, since the cost of re-spawning these compilers is avoided for subsequent builds. This change has no impact on non-continuous builds: in this case each compiler daemon will be stopped at the end of the build.
The following compilers are forked and should demonstrate improved performance with continuous builds:
options.fork = true (default is false)scala plugin and scalaCompileOptions.useAnt = false (default is true)options.fork = true (default is true)play pluginThis release also contains various other performance improvements that are generally applicable to most builds. These improvements are due to the use of more efficient data structures and smarter caching. Build time reductions vary depending on the size and nature of the build.
Previous releases of Gradle required the end user to provide classes under test (e.g. plugin and custom task implementations) to the TestKit by assigning them to the buildscript's classpath.
This release makes it more convenient to inject classes under test through the GradleRunner API with the method withPluginClasspath(Iterable<File>). This classpath is then available to use to locate plugins in a test build via the plugins DSL. The following code example demonstrates the use of the new TestKit API in a test class based on the test framework Spock:
class BuildLogicFunctionalTest extends Specification {
@Rule final TemporaryFolder testProjectDir = new TemporaryFolder()
File buildFile
List<File> pluginClasspath
def setup() {
buildFile = testProjectDir.newFile('build.gradle')
pluginClasspath = getClass().classLoader.findResource("plugin-classpath.txt")
.readLines()
.collect { new File(it) }
}
def "execute helloWorld task"() {
given:
buildFile << """
plugins {
id 'com.company.helloworld'
}
"""
when:
def result = GradleRunner.create()
.withProjectDir(testProjectDir.root)
.withArguments('helloWorld')
.withPluginClasspath(pluginClasspath)
.build()
then:
result.standardOutput.contains('Hello world!')
result.taskPaths(SUCCESS) == [':helloWorld']
}
}Future versions of Gradle will aim for automatically injecting the classpath without additional configuration from the end user.
Gradle will use the default character encoding for file names when creating Zip archives. Depending on where the archive will be extracted, this may not be the best possible encoding to use due to the way various operating systems and archive tools interpret file names in the archive. Some tools assume the extracting platform character encoding is the same encoding used to create the archive. A mismatch between encodings will manifest itself as "corrupted" or "mangled" file names.
Zip tasks can now be configured with an explicit character encoding to use for file names and comment fields in the archive. (This option does not affect the content of the files in the archive.) The default behavior has not been changed, so no changes should be necessary for existing builds.
By default, the PMD plugin will report all rule violations and fail the build if any violations are found. This means the only way to ignore low priority violations was to create a custom ruleset.
Gradle now supports configuring a "rule priority" threshold. The PMD report will contain only violations higher than or equal to the priority configured, and the build will only fail if one of these "priority" violations is discovered.
You configure the rule priority threshold via the PmdExtension. You can also configure the property on a per-task level through the Pmd task.
pmd {
rulePriority = 3
}Some PMD rules require access to the dependencies of your project to perform type resolution. If the dependencies are available on PMD's auxclasspath, additional problems can be detected.
Gradle now automatically adds the compile dependencies of each analyzed source set to PMD's auxclasspath. No additional configuration should be necessary to enable this in existing builds.
A number of improvements have been made to the experimental managed model feature. This feature is currently used by the native and Play plugins and will evolve into a general purpose feature for describing Gradle builds.
Set and List of scalar typesThe managed model now supports collections of scalar types. This means that it is possible to use a Set or a List with element type:
Number type (Integer, Double, ...)BooleanStringFileA collection of a scalar type can be attached to any @Managed type as a property, or used for the elements of a ModelSet or ModelMap, or as a top level element. For example:
@Managed
interface User {
Set<String> getGroups();
}Collections of scalar types are available as read-only properties, in which case they default to an empty collection, or as read-write properties, in which case they default to null. Read-only properties are similar to final values: they can be mutated as long as they are the subject of a rule. Read-write properties can also be mutated, but they are not final: the collection can be overwritten by a configuration rule.
A read-only (non nullable) property is created by defining only a setter, while a read-write property is created by defining both a setter and a getter:
@Managed
interface User {
Set<String> getGroups();
void setGroups(Set<String> groups);
}FunctionalSourceSetThis release facilitates adding source sets (LanguageSourceSet) to arbitrary locations in the model space through the use of the language-base plugin and FunctionalSourceSet. A FunctionalSourceSet can be attached to any @Managed type as a property, or used for the elements of a ModelSet or ModelMap, or as a top level element.
Using FunctionalSourceSet allows build and plugin authors to strongly model things that use collections of sources as LanguageSourceSets. This kind of strongly typed modelling also allows build and plugin authors to access LanguageSourceSets in a controlled and consistent way using rules.
Adding a top level FunctionalSourceSet is as simple as:
model {
sources(FunctionalSourceSet)
}or from a RuleSource
class Rules extends RuleSource {
@Model
void functionalSources(FunctionalSourceSet sources) {
}
}
apply plugin: RulesHere's an example of creating a managed type with FunctionalSourceSet properties.
@Managed
interface BuildType {
FunctionalSourceSet getSources()
FunctionalSourceSet getInputs()
ModelMap<FunctionalSourceSet> getComponentSources()
}It is now possible to attach internal information to an unmanaged ComponentSpec. This way a plugin can make some data about its components visible to build logic via a public component type, while hiding the rest of the data behind the internal view type. The default implementation of the component must implement all internal views declared for the component.
interface SampleLibrarySpec extends ComponentSpec {
String getPublicData()
void setPublicData(String publicData)
}
interface SampleLibrarySpecInternal extends ComponentSpec {
String getInternalData()
void setInternalData(String internalData)
}
class DefaultSampleLibrarySpec extends BaseComponentSpec implements SampleLibrarySpec, SampleLibrarySpecInternal {
String internalData
String publicData
}Components can be targeted by rules via their internal types when those internal types extend ComponentSpec:
class Rules extends RuleSource {
@Mutate
void mutateInternal(ModelMap<SampleLibrarySpecInternal> sampleLibs) {
sampleLibs.each { sampleLib ->
sampleLib.internalData = "internal"
}
}
}For internal view types that do not not extend ComponentSpec, targeting components can be achieved via ComponentSpecContainer.withType():
@Defaults
void finalize(ModelMap<SampleLibrarySpec> sampleLibs) {
sampleLibs.withType(SomeInternalView).all { sampleLib ->
// ...
}
}Interoperability between legacy configuration space and new rule based model configuration space has been improved. More specifically, the tasks.withType(..) construct allows legacy configuration tasks to depend on tasks created via the new rule based approach. See this issue for details.
Currently, the managed model works well for defining a tree of objects. This release improves support for a graph of objects, with better support for references between different model elements. In particular, the inputs for a rule can traverse a "reference" property, which is simply a read-write property on a @Managed type that refers to some other model element.
Features that have become superseded or irrelevant due to the natural evolution of Gradle become deprecated, and scheduled to be removed in the next major Gradle version (Gradle 3.0). See the User guide section on the “Feature Lifecycle” for more information.
The following are the newly deprecated items in this Gradle release. If you have concerns about a deprecation, please raise it via the Gradle Forums.
sonar and sonar-runner plugins have been superseded by an external plugin from SonarQubeSonarQube is an open platform to manage code quality, and in previous versions of Gradle the sonar and sonar-runner plugin has provided integration with this analysis tool. An improved plugin is now available directly from the developers of SonarQube. This new plugin supersedes the sonar and sonar-runner plugins that are part of the Gradle distribution; the sonar and sonar-runner plugin are now deprecated and will be removed in Gradle 3.0.
See the official documentation from SonarQube for more details.
The following classes have been deprecated and will be removed in Gradle 3.0:
org.gradle.api.artifacts.ConflictResolution.org.gradle.api.artifacts.Module.org.gradle.api.tasks.javadoc.AntGroovydoc.org.gradle.api.tasks.scala.AntScalaDoc.org.gradle.BuildExceptionReporter.org.gradle.BuildLogger.org.gradle.BuildResultLogger.org.gradle.TaskExecutionLogger.org.gradle.plugins.ide.eclipse.model.EclipseDomainModel.org.gradle.api.logging.Logging.ANT_IVY_2_SLF4J_LEVEL_MAPPER. This constant is intended to be used only by the Gradle internals, and will be removed from the Gradle API.org.gradle.util.AvailablePortFinder. Although this class is an internal class and not a part of the public API, some users may be utilizing it and should plan to implement an alternative.Setting the Eclipse project name in eclipse.project.file.beforeMerged or eclipse.project.file.whenMerged hook provided by the Eclipse plugin has been deprecated. Support for this will be removed in Gradle 3.0
In previous versions of Gradle, adding a one-item list of FileCollection objects to a FileCollection would succeed.
For example:
FileCollection foo = files()
foo += [files()]However, this behavior was never explicitly supported and only worked incidentally because Groovy selected the appropriate method on FileCollection to add a single FileCollection object. In Groovy 2.4, this behavior has changed and no longer selects the correct method (see "Using the + operator with Iterable objects" under Upgraded to Groovy 2.4.4). As this construct may be in use, we have added a workaround to maintain backwards compatibility, but this unsupported functionality is considered deprecated.
Instead, adding the FileCollection without a list should be used:
FileCollection foo = files()
foo += files()The Gradle API now uses Groovy 2.4.4. Previously, it was using Groovy 2.3.10. This change should be transparent to the majority of users; however, it can cause minor problems with existing build scripts and plugins.
Please refer to the Groovy language changelogs for further details.
In earlier versions of Groovy, an assignment to a property that has multiple setter methods that each take a different parameter type would select the setter method in a JVM-implementation dependent way. This is documented in GROOVY-6084.
For example:
class Foo {
void setProperties(Properties p) { }
void setProperties(File f) { }
}
def foo = new Foo()
foo.properties = [:]In Groovy 2.3 and earlier, this would sometimes fail depending on which setter method was selected.
In Groovy 2.4, this always fails with an exception:
No signature of method: Foo.setProperties() is applicable for argument types: (java.util.LinkedHashMap) values: [[:]]
Possible solutions: setProperties(java.io.File), setProperties(java.util.Properties), getProperties(), getProperty(java.lang.String), setProperty(java.lang.String, java.lang.Object)To fix this problem, select the correct method by supplying the appropriate type (in this case, Properties).
In earlier versions of Groovy, when the '+' operator was used with objects that implement java.lang.Iterable and there was not a direct match on the signature of the plus() method, the behavior was not defined and could result in success.
For example:
class Foo implements Iterable<String> {
public Foo plus(Foo bar) { }
}
Foo foo = new Foo()
foo += [ new Foo() ]In Groovy 2.3 and earlier, this may work, invoking the plus(Foo) method.
In Groovy 2.4, this always fails with an exception:
Cannot cast object '[Foo@563a3ada]' with class 'java.util.ArrayList' to class 'Foo' due to: groovy.lang.GroovyRuntimeException: Could not find matching constructor for: Foo(Foo)To fix this problem, add a plus() method that accepts an appropriate java.lang.Iterable object so that there is a direct match on the method signature.
PMD can perform additional analysis for some rules (see above), therefore new violations may be found in existing projects. Previously, these rules were unable to detect problems because classes outside of your project were not available during analysis.
The default version of CodeNarc has been updated from 0.23 to 0.24.1. Should you want to stay on older version, it is possible to downgrade it using the codenarc configuration:
dependencies {
codenarc 'org.codenarc:CodeNarc:0.17'
}The built-in OSGi plugin uses the bndlib library internally. The library has been updated from 2.1.0 to 2.4.0.
To ensure unique project names in the IDE, Gradle applies de-duplication logic when generating IDE metadata for Eclipse and Idea projects. This de-duplication logic has been improved such that all projects with non-unique names are now de-duplicated by adding a prefix based on the parent project(s). For example:
Given a Gradle multiproject build with the following project structure:
root
|-foo
| \- app
|
\-bar
\- appThe following IDE project name mapping will result:
root
|-foo
| \- foo-app
|
\-bar
\- bar-appDuplicate words in a row within the de-duplication prefix are removed from the generated ide project name.
Given a project with the following structure:
myapp
|-myapp-foo
| \- app
|
\-myapp-bar
\- appThe following IDE project name mapping will result:
myapp
|-myapp-foo
| \- myapp-foo-myapp-app
|
\-myapp-bar
\- myapp-bar-myapp-appThe Java plugins make some details about the project source sets visible in the managed model, to allow integration between rules based plugins and the stable Java plugins. This integration has changed in this Gradle release, to move more of the integration into the managed model:
sources container is no longer added as a project extension. It is now visible only to rules, as part of the managed model.ClassDirectoryBinarySpec instances can no longer be added to the binaries container. Instances are still added to this container by the Java plugins for each source set, however, additional instances cannot be added. This capability will be added again in a later release, to allow rules based plugins to define arbitrary class directory binaries.binaries container until they are required by rules. Previously, the plugins would add these instances eagerly when the source set was defined.We would like to thank the following community members for making contributions to this release of Gradle.
We love getting contributions from the Gradle community. For information on contributing, please see gradle.org/contribute.
Known issues are problems that were discovered post release that are directly related to changes made in this release.