uber / NullAway

NullAway: Fast Annotation-Based Null Checking for Java NullAway is a tool to help eliminate s (NPEs) in your Java code. To use NullAway, first add annotations in your code wherever a field, method parameter, or return value may be . Given these annotations, NullAway performs a series of type-based, local checks to ensure that any pointer that gets dereferenced in your code cannot be . NullAway is similar to the type-based nullability checking in the Kotlin and Swift languages, and the Checker Framework and Eradicate null checkers for Java. NullAway is *fast*. It is built as a plugin to Error Prone and can run on every single build of your code. In our measurements, the build-time overhead of running NullAway is usually less than 10%. NullAway is also *practical*: it does not prevent all possible NPEs in your code, but it catches most of the NPEs we have observed in production while imposing a reasonable annotation burden, giving a great "bang for your buck." Installation Overview NullAway requires that you build your code with JDK 17 or higher and Error Prone, version 2.36.0 or higher. See the Error Prone documentation for instructions on getting started with Error Prone and integration with your build system. The instructions below assume you are using Gradle; see the docs for discussion of other build systems. If you are building with JSpecify mode enabled, we recommend building with the most recent JDK available; see the wiki docs on JSpecify support for more details. Gradle Java (non-Android) To integrate NullAway into your non-Android Java project, add the following to your file: Let's walk through this script step by step. The section pulls in the Gradle Error Prone plugin for Error Prone integration. In , the first line loads NullAway, and the line loads the JSpecify library which provides suitable nullability annotations, e.g., . NullAway allows for any annotation to be used, so, e.g., from the AndroidX annotations Library or JetBrains annotations is also fine. The second line sets the version of Error Prone is used. Finally, in the section, we pass some configuration options to NullAway. First sets NullAway issues to the error level (it's equivalent to the standard Error Prone argument); by default NullAway emits warnings. Then, (equivalent to the standard Error Prone argument) tells NullAway that source code in packages under the namespace should be checked for null dereferences and proper usage of annotations, and that class files in these packages should be assumed to have correct usage of (see the docs for more detail). NullAway requires exactly one of the or configuration arguments to run, in order to distinguish between annotated and unannotated code. See the configuration docs for more details and other useful configuration options. For even simpler configuration of NullAway options, use the Gradle NullAway plugin. Finally, we show how to disable NullAway on test code, if desired. We recommend addressing all the issues that Error Prone reports, particularly those reported as errors (rather than warnings). But, if you'd like to try out NullAway without running other Error Prone checks, you can use (equivalent to passing to the compiler, before the NullAway-specific arguments). Android Versions 3.0.0 and later of the Gradle Error Prone Plugin no longer support Android. So if you're using a recent version of this plugin, you'll need to add some further configuration to run Error Prone and NullAway. Our sample app file shows one way to do this, but your Android project may require tweaks. Alternately, 2.x versions of the Gradle Error Prone Plugin still support Android and may still work with your project. Beyond that, compared to the Java configuration, the JSpecify dependency can be removed; you can use the annotation from the AndroidX annotation library instead. Annotation Processors / Generated Code Some annotation processors like Dagger and AutoValue generate code into the same package namespace as your own code. This can cause problems when setting NullAway to the level as suggested above, since errors in this generated code will block the build. Currently the best solution to this problem is to completely disable Error Prone on generated code, using the option added in Error Prone 2.1.3 (documented here, use in Gradle). To use, figure out which directory contains the generated code, and add that directory to the excluded path regex. **Note for Dagger users**: Dagger versions older than 2.12 can have bad interactions with NullAway; see here. Please update to Dagger 2.12 to fix the problem. JSpecify Mode / Guava As of version 33.4.1, Guava uses JSpecify annotations for most of its packages, and hence NullAway will treat those packages as annotated by default. This treatment may lead to some false positives (due to handling of type variables), which can be mitigated by running NullAway in its (under-development) JSpecify mode. See the wiki docs on JSpecify support for more details. Lombok Unlike other annotation processors above, Lombok modifies the in-memory AST of the code it processes, which is the source of numerous incompatibilities with Error Prone and, consequently, NullAway. We do not particularly recommend using NullAway with Lombok. However, NullAway encodes some knowledge of common Lombok annotations and we do try for best-effort compatibility. In particular, common usages like and classes should be supported. In order for NullAway to successfully detect Lombok generated code within the in-memory Java AST, the following configuration option must be passed to Lombok as part of an applicable file: This causes Lombok to add to the methods/classes it generates. NullAway will ignore (i.e. not check) the implementation of this generated code, treating it as unannotated. Code Example Let's see how NullAway works on a simple code example: This code is buggy: when is called, the subsequent call to will fail with an NPE. You can see this error…