Baseline Java code quality plugins

Baseline is a family of Gradle plugins for configuring Java projects with sensible defaults for code-style, static analysis, dependency versioning, CircleCI and IntelliJ IDEA/Eclipse integration.

Plugin	Description
com.palantir.baseline-idea	Configures Intellij IDEA with code style and copyright headers
com.palantir.baseline-eclipse	Configures Eclipse with code style and copyright headers
com.palantir.baseline-error-prone	Static analysis for your Java code using Google's error-prone.
com.palantir.baseline-checkstyle	Enforces consistent Java formatting using checkstyle
com.palantir.baseline-format	Formats your java files to comply with checkstyle
com.palantir.baseline-scalastyle	Enforces formatting using scalastyle
com.palantir.baseline-class-uniqueness	Analyses your classpath to ensure no fully-qualified class is defined more than once.
com.palantir.baseline-circleci	CircleCI integration using $CIRCLE_ARTIFACTS and $CIRCLE_TEST_REPORTS dirs
com.palantir.baseline-config	Config files for the above plugins
com.palantir.baseline-reproducibility	Sensible defaults to ensure Jar, Tar and Zip tasks can be reproduced
com.palantir.baseline-exact-dependencies	Ensures projects explicitly declare all the dependencies they rely on, no more and no less
com.palantir.baseline-release-compatibility	Ensures projects targetting older JREs only compile against classes and methods available in those JREs.
com.palantir.baseline-testing	Configures test tasks to dump heap dumps (hprof files) for convenient debugging

See also the Baseline Java Style Guide and Best Practices.

Usage

The baseline set of plugins requires at least Gradle 5.0.

It is recommended to add apply plugin: 'com.palantir.baseline' to your root project's build.gradle. Individual plugins will be automatically applied to appropriate subprojects.

buildscript {
    repositories {
        gradlePluginPortal()
        maven { url  "http://palantir.bintray.com/releases" }
    }

    dependencies {
        classpath 'com.palantir.baseline:gradle-baseline-java:<version>'
        classpath 'gradle.plugin.org.inferred:gradle-processors:2.1.0'
    }
}

repositories {
    maven { url  "http://palantir.bintray.com/releases" }
}

apply plugin: 'java'
apply plugin: 'org.inferred.processors'  // installs the "processor" configuration needed for baseline-error-prone
apply plugin: 'com.palantir.baseline'

Run ./gradlew baselineUpdateConfig to download config files and extract them to the .baseline/ directory. These files should be committed to your repository to ensure reproducible builds.

Tip: Install the CheckStyle-IDEA plugin to run checkstyle from within IntelliJ.

Selective usage

Alternatively, you can apply plugins selectively, e.g.:

apply plugin: 'com.palantir.baseline-config'

allprojects {
    apply plugin: 'com.palantir.baseline-idea'
}

subprojects {
    apply plugin: 'java'
    apply plugin: 'com.palantir.baseline-checkstyle'
}

com.palantir.baseline-idea

Run ./gradlew idea to (re-) generate IntelliJ project and module files from the templates in .baseline. The generated project is pre-configured with Baseline code style settings and support for the CheckStyle-IDEA plugin.

The com.palantir.baseline-idea plugin automatically applies the idea plugin.

Generated IntelliJ projects have default per-project code formatting rules as well as Checkstyle configuration. The JDK and Java language level settings are picked up from the Gradle sourceCompatibility property on a per-module basis.

com.palantir.baseline-eclipse

Run ./gradlew eclipse to repopulate projects from the templates in .baseline.

The com.palantir.baseline-eclipse plugin automatically applies the eclipse plugin, but not the java plugin. The com.palantir.baseline-eclipse plugin has no effects if the java plugin is not applied.

If set, sourceCompatibility is used to configure the Eclipse project settings and the Eclipse JDK version. Note that targetCompatibility is also honored and defaults to sourceCompatibility.

Generated Eclipse projects have default per-project code formatting rules as well as Checkstyle configuration.

The Eclipse plugin is compatible with the following versions: Checkstyle 7.5+, JDK 1.7, 1.8

com.palantir.baseline-error-prone

The com.palantir.baseline-error-prone plugin brings in the net.ltgt.errorprone-javacplugin plugin. We recommend applying the org.inferred.processors plugin 1.3.0+ in order to avoid error: plug-in not found: ErrorProne. The minimal setup is as follows:

buildscript {
    dependencies {
        classpath 'gradle.plugin.org.inferred:gradle-processors:1.2.18'
    }
}

apply plugin: 'org.inferred.processors'
apply plugin: 'com.palantir.baseline-error-prone'

Error-prone rules can be suppressed on a per-line or per-block basis just like Checkstyle rules:

@SuppressWarnings("Slf4jConstantLogMessage")

Rules can be suppressed at the project level, or have their severity modified, by adding the following to the project's build.gradle:

tasks.withType(JavaCompile).configureEach {
    options.errorprone.errorproneArgs += ['-Xep:Slf4jLogsafeArgs:OFF']
    // alternatively, using the DSL:
    options.errorprone {
        check('Slf4jLogsafeArgs', CheckSeverity.OFF)
    }
}

More information on error-prone severity handling can be found at errorprone.info/docs/flags.

Baseline error-prone checks

Baseline configures the following checks in addition to the error-prone's out-of-the-box checks:

• DangerousParallelStreamUsage: Discourage the use of Java parallel streams.
• Slf4jConstantLogMessage: Allow only compile-time constant slf4j log message strings.
• Slf4jLevelCheck: Slf4j level checks (if (log.isInfoEnabled()) {) must match the most severe level in the containing block.
• Slf4jLogsafeArgs: Allow only com.palantir.logsafe.Arg types as parameter inputs to slf4j log messages. More information on Safe Logging can be found at github.com/palantir/safe-logging.
• PreferCollectionTransform: Prefer Guava's Lists.transform or Collections2.transform instead of Iterables.transform when first argument's declared type is a List or Collection type for performance reasons.
• PreferListsPartition: Prefer Guava's Lists.partition(List, int) instead of Iterables.partition(Iterable, int) when first argument's declared type is a list for performance reasons.
• PreferSafeLoggableExceptions: Users should throw SafeRuntimeException instead of RuntimeException so that messages will not be needlessly redacted when logs are collected:

  -throw new RuntimeException("explanation", e); // this message will be redacted when logs are collected
  +throw new SafeRuntimeException("explanation", e); // this message will be preserved (allowing easier debugging)

• PreferSafeLoggingPreconditions: Users should use the safe-logging versions of Precondition checks for standardization when there is equivalent functionality

  -com.google.common.base.Preconditions.checkNotNull(variable, "message");
  +com.palantir.logsafe.Preconditions.checkNotNull(variable, "message"); // equivalent functionality is available in the safe-logging variant

• ShutdownHook: Applications should not use Runtime#addShutdownHook.
• SwitchStatementDefaultCase: Switch statements should avoid using default cases. Default cases prevent the MissingCasesInEnumSwitch check from detecting when an enum value is not explicitly handled. This check is important to help avoid incorrect behavior when new enum values are introduced.
• GradleCacheableTaskAction: Gradle plugins should not call Task.doFirst or Task.doLast with a lambda, as that is not cacheable. See gradle/gradle#5510 for more details.
• PreferBuiltInConcurrentKeySet: Discourage relying on Guava's com.google.common.collect.Sets.newConcurrentHashSet(), when Java's java.util.concurrent.ConcurrentHashMap.newKeySet() serves the same purpose.
• JUnit5RuleUsage: Prevent accidental usage of org.junit.Rule/org.junit.ClassRule within Junit5 tests
• DangerousCompletableFutureUsage: Disallow CompletableFuture asynchronous operations without an Executor.
• NonComparableStreamSort: Stream.sorted() should only be called on streams of Comparable types.
• DangerousStringInternUsage: Disallow String.intern() invocations in favor of more predictable, scalable alternatives.
• OptionalOrElseThrowThrows: Optional.orElseThrow argument must return an exception, not throw one.
• LambdaMethodReference: Lambda should use a method reference.
• SafeLoggingExceptionMessageFormat: SafeLoggable exceptions do not interpolate parameters.
• StrictUnusedVariable: Functions shouldn't have unused parameters.
• StringBuilderConstantParameters: StringBuilder with a constant number of parameters should be replaced by simple concatenation.
• JUnit5SuiteMisuse: When migrating from JUnit4 -> JUnit5, classes annotated with @RunWith(Suite.class) are dangerous because if they reference any JUnit5 test classes, these tests will silently not run!
• ThrowError: Prefer throwing a RuntimeException rather than Error.
• ReverseDnsLookup: Calling address.getHostName may result in an unexpected DNS lookup.
• ReadReturnValueIgnored: The result of a read call must be checked to know if EOF has been reached or the expected number of bytes have been consumed.
• FinalClass: A class should be declared final if all of its constructors are private.
• RedundantModifier: Avoid using redundant modifiers.
• StrictCollectionIncompatibleType: Likely programming error due to using the wrong type in a method that accepts Object.
• InvocationHandlerDelegation: InvocationHandlers which delegate to another object must catch and unwrap InvocationTargetException.
• Slf4jThrowable: Slf4j loggers require throwables to be the last parameter otherwise a stack trace is not produced.
• JooqResultStreamLeak: Autocloseable streams and cursors from jOOQ results should be obtained in a try-with-resources statement.
• StreamOfEmpty: Stream.of() should be replaced with Stream.empty() to avoid unnecessary varargs allocation.
• RedundantMethodReference: Redundant method reference to the same type.
• ExceptionSpecificity: Prefer more specific catch types than Exception and Throwable.
• ThrowSpecificity: Prefer to declare more specific throws types than Exception and Throwable.
• UnsafeGaugeRegistration: Use TaggedMetricRegistry.registerWithReplacement over TaggedMetricRegistry.gauge.
• BracesRequired: Require braces for loops and if expressions.
• CollectionStreamForEach: Collection.forEach is more efficient than Collection.stream().forEach.
• LoggerEnclosingClass: Loggers created using getLogger(Class<?>) must reference their enclosing class.
• UnnecessaryLambdaArgumentParentheses: Lambdas with a single parameter do not require argument parentheses.
• RawTypes: Avoid raw types; add appropriate type parameters if possible.
• VisibleForTestingPackagePrivate: @VisibleForTesting members should be package-private.

Programmatic Application

There exist a number of programmatic code modifications available via refaster. You can run these on your code to apply some refactorings automatically:

./gradlew compileJava compileTestJava -PrefasterApply -PerrorProneApply

You may apply specific error-prone refactors including those which are not enabled by default by providing a comma delimited list of check names to the -PerrorProneApply option.

./gradlew compileJava compileTestJava -PerrorProneApply=ThrowSpecificity

com.palantir.baseline-checkstyle

Checkstyle rules can be suppressed on a per-line or per-block basis. (It is good practice to first consider formatting the code block in question according to the project's style guidelines before adding suppression statements.) To suppress a particular check, say MagicNumberCheck, from an entire class or method, annotate the class or method with the lowercase check name without the "Check" suffix:

@SuppressWarnings("checkstyle:magicnumber")

Checkstyle rules can also be suppressed using comments, which is useful for checks such as IllegalImport where annotations cannot be used to suppress the violation. To suppress checks for particular lines, add the comment // CHECKSTYLE:OFF before the first line to suppress and add the comment // CHECKSTYLE:ON after the last line.

To disable certain checks for an entire file, apply custom suppressions in .baseline/checkstyle/custom-suppressions.xml. Avoid adding suppressions to the autogenerated .baseline/checkstyle/checkstyle-suppressions.xml, as that file will be overridden on updates.

Copyright Checks

Baseline enforces Palantir copyright at the beginning of files when applying com.palantir.baseline-format. To change this, edit the template copyrights in .baseline/copyright/*.txt. The largest file (sorted lexicographically) will be used to generate a new copyright if one is missing, or none of the existing templates match.�

To automatically update all files with mismatching/missing copyrights, run ./gradlew format.

com.palantir.baseline-class-uniqueness

When applied to a java project, this inspects all the jars in your runtimeClasspath configuration and records any conflicts to a baseline-class-uniqueness.lock file. For example:

# Danger! Multiple jars contain identically named classes. This may cause different behaviour depending on classpath ordering.
# Run ./gradlew checkClassUniqueness --write-locks to update this file

## runtimeClasspath
[jakarta.annotation:jakarta.annotation-api, javax.annotation:javax.annotation-api]
  - javax.annotation.Resource$AuthenticationType
[jakarta.ws.rs:jakarta.ws.rs-api, javax.ws.rs:javax.ws.rs-api]
  - javax.ws.rs.BadRequestException
  - javax.ws.rs.ClientErrorException
  - javax.ws.rs.ForbiddenException
  - javax.ws.rs.InternalServerErrorException
  - javax.ws.rs.NotAcceptableException
  - javax.ws.rs.NotAllowedException
  - javax.ws.rs.NotAuthorizedException
  - javax.ws.rs.NotFoundException
  - javax.ws.rs.NotSupportedException
  - javax.ws.rs.Priorities

This task can also be used to analyze other configurations in addition to runtimeClasspath, e.g.:

checkClassUniqueness {
  configurations.add project.configurations.myConf
}

If you discover multiple jars on your classpath contain clashing classes, you should ideally try to fix them upstream and then depend on the fixed version. If this is not feasible, you may be able to tell Gradle to use a substituted dependency instead:

configurations.all {
    resolutionStrategy.eachDependency { DependencyResolveDetails details ->
        if (details.requested.name == 'log4j') {
            details.useTarget group: 'org.slf4j', name: 'log4j-over-slf4j', version: '1.7.10'
            details.because "prefer 'log4j-over-slf4j' over any version of 'log4j'"
        }
    }
}

com.palantir.baseline-circleci

The plugin surfaces failures using JUnit XML which is rendered nicely by CircleCI, by

1. Storing JUnit test reports in $CIRCLE_TEST_REPORTS/junit
2. Converting java compilation errors and checkstyle errors into test failures stored under $CIRCLE_TEST_REPORTS/javac and $CIRCLE_TEST_REPORTS/checkstyle respectively
3. Storeing the HTML output of tests in $CIRCLE_ARTIFACTS/junit

com.palantir.baseline-format

Adds a ./gradlew format task which autoformats all Java files using Spotless. Roughly equivalent to:

buildscript {
    dependencies {
        classpath 'com.diffplug.spotless:spotless-plugin-gradle:3.14.0'
    }
}

apply plugin: 'com.diffplug.gradle.spotless'

spotless {
    java {
        target 'src/main/java/**/*.java', 'src/main/test/**/*.java'
        removeUnusedImports
        importOrder ''
        trimTrailingWhitespace
        indentWithSpaces 4
    }
}

Add com.palantir.baseline-format.eclipse=true to your gradle.properties to format entire files with the Eclipse formatter. The Eclipse formatter can be run from IntelliJ using the Eclipse Code Formatter plugin.

To iterate on the eclipse.xml formatter config, you can import it into an instance of Eclipse, edit it through the preferences UI and then export it, or you can manually tune individual values by referring to the master list of DefaultCodeFormatterConstants and DefaultCodeFormatterOptions. Running ./gradlew :gradle-baseline-java:test -Drecreate=true should update all the checked-in snapshot test cases.

Add com.palantir.baseline-format.palantir-java-format=true to your gradle.properties to run our experimental fork of google-java-format. The Palantir Java Formatter can be run from IntelliJ using the palantir-java-format plugin.

com.palantir.baseline-reproducibility

This plugin is a shorthand for the following snippet, which opts-in to reproducible behaviour for all Gradle's Jar, Tar and Zip tasks. (Surprisingly, these tasks are not reproducible by default).

tasks.withType(AbstractArchiveTask) {
    preserveFileTimestamps = false
    reproducibleFileOrder = true
}

It also warns if it detects usage of the nebula.info plugin which is known to violate the reproducibility of Jars by adding a 'Build-Date' entry to the MANIFEST.MF, which will be different on every run of ./gradlew jar.

Complete byte-for-byte reproducibility is desirable because it enables the Gradle build cache to be much more effective.

com.palantir.baseline-exact-dependencies

This plugin adds two tasks to help users ensure they explicitly declare exactly the dependencies they need - nothing more and nothing less:

• checkUnusedDependencies - fails if a project pulls in a jar but never compiles against classes from it. This is undesirable because it inflates published jars and distributions.
• checkImplicitDependencies - fails if source code relies on classes that only appear on the classpath transitively. This is fragile because without a direct dependency on the relevant jar, a seemingly unrelated dependency upgrade could cause compilation to start failing.

Both of these tasks can be configured to ignore specific dependencies if this improves the signal-to-noise ratio. The following snippet illustrates the defaults that are baked into the plugin:

checkUnusedDependencies {
    ignore 'javax.annotation', 'javax.annotation-api'
}

checkImplicitDependencies {
    ignore 'org.slf4j', 'slf4j-api'
}

com.palantir.baseline-release-compatibility

This plugin adds the --release <number> flag to JavaCompile tasks (when the compiler supports it), so that published jars will only use methods available in the target JRE. Relying on sourceCompatibility = 1.8 and targetCompatibility = 1.8 is insufficient because you run the risk of using method that have been added in newer JREs, e.g. Optional#isEmpty.

This plugin may become redundant if this functionality is implemented upstream in Gradle.

com.palantir.baseline-testing

Configures some sensible defaults:

1. For debugging purposes:

   tasks.withType(Test) {
       jvmArgs '-XX:+HeapDumpOnOutOfMemoryError', '-XX:+CrashOnOutOfMemoryError'
   }

   This ensures that if one of your tests fails with an OutOfMemoryError (OOM), you'll get a large hprof file in the relevant subdirectory which can be analyzed with Eclipse Memory Analyzer Tool, Yourkit profiler, jvisualvm etc.

2. If Gradle detects you use JUnit 5 (i.e. you have a testImplementation 'org:junit.jupiter:junit-jupiter' dependency), it will automatically configure your Test tasks to run with useJUnitPlatform(), and configure all @Test methods to run in parallel by default. Many other languages take this stance by default - if some tests rely on static state then you can mark them as non-parallel.

   See more here: https://junit.org/junit5/docs/current/user-guide/#writing-tests-parallel-execution

The plugin also adds a checkJUnitDependencies to make the migration to JUnit5 safer. Specifically, it should prevent cases where the tests could silently not run due to misconfigured dependencies.

3. For repos that use 'snapshot' style testing, it's convenient to have a single command to accept the updated snapshots after a code change. This plugin ensures that if you run tests with ./gradlew test -Drecreate=true, the system property will be passed down to the running Java process (which can be detected with Boolean.getBoolean("recreate")).

com.palantir.baseline-fix-gradle-java

Fixes up all Java SourceSets by marking their deprecated configurations

• compile and runtime - as well as the compileOnly configuration as not resolvable (can't call resolve on them) and not consumable (can't be depended on from other projects).

See here for a more in-depth discussion on what these terms mean. By configuring them thusly, we are saying that these configurations now fulfil the "Bucket of dependencies" role described in that document, as they should.

This will become the default in Gradle 7 and leaving these as they currently are can cause both unnecessary confusion (users looking in compile instead of compileClasspath) and random crashes.