This document describes some WALA-specific aspects of our new Gradle build system, plus a few general Gradle features that may be of particular interest to WALA developers. However, it is not a Gradle tutorial.
- more comprehensive management of external dependencies
- faster builds using parallel task execution and user-scoped caching (both enabled by default)
- trustworthy dependencies for incremental builds
- composite builds for easier integration of WALA into larger projects
- automated Travis CI testing on macOS
The Gradle build system is not yet ready to replace Maven, due to a few known shortcomings. Paramount among these is that Gradle WALA builds do not yet package up Eclipse plug-ins / features in the proper manner. I have poked at this a bit, but I simply do not understand Eclipse and/or OSGi well enough to get the job done. I welcome help from anyone with the right knowledge!
Either Gradle or Maven can be used to build WALA from the command line. However, it was not possible to extend this dual-build-system support to Eclipse. Working with WALA in Eclipse requires doing things the Gradle way. Fortunately, this is rather seamless; I see no reason why an Eclipse-using WALA developer should need to avoid this switch.
As noted below, Eclipse
.classpath and .project files are now generated rather than being
repository-tracked source files. However, a few Maven-run tests
depend on having certain of these files present. One way to create
them is to import WALA into
Eclipse before starting your
Maven tests. If you prefer a non-interactive approach, you can
instead run ./gradlew prepareMavenBuild before starting your Maven
tests.
Gradle downloads many packages and supporting Java libraries as
needed. Your first Gradle build may take a long time. On a fast
workstation with a University-grade network and no local caches, my
initial run of ./gradlew assemble processTestResources took five
minutes. On a decent laptop with residential DSL and no local caches,
the same initial build took twenty minutes. Fortunately, user- and
project-level Gradle caches will make incremental rebuilds much
faster. Rerunning ./gradlew assemble processTestResources with a
warm cache in an already-built tree takes under three seconds.
Maven is the same, really. You may already have most of what Maven needs downloaded and cached locally, but your first Maven WALA build was probably slow as well. Recent Travis CI runs have showed Gradle and Maven builds completing in fifteen to twenty minutes, without significant variation between the two build systems.
To work with WALA inside Eclipse, first install Eclipse Buildship using either the Eclipse Marketplace or the Eclipse update manager. Buildship integrates Eclipse with Gradle, much like how M2Eclipse integrates Eclipse with Maven. Restart Eclipse after installing this feature.
Once you are running a Buildship-enabled Eclipse, use the “Existing Gradle Project” import wizard to import WALA into Eclipse. Select and import the topmost level of your WALA source tree. On the “Import Options” page of the import wizard, leave all settings at their defaults: the “Override workspace settings” option should be off, and the grayed-out “Gradle distribution” choice should be set to “Gradle wrapper”. You do not need to select each of WALA’s sub-projects; import only the top-level WALA source tree, and the rest will follow.
After the lengthy import process completes, use “Run → After Importing WALA Into Eclipse” to perform some post-import cleanup and configuration. Immediately after importing, you may see some errors in the Eclipse “Problems” view. These should all go away after running the “After Importing WALA Into Eclipse” step.
Note: a pristine WALA source tree is not pre-configured as a group of Eclipse projects. Using the standard Eclipse “Existing Projects into Workspace” import wizard will not work correctly. You must use the “Existing Gradle Project” import wizard instead.
You will find no .classpath or .project files anywhere in the
Gradle fork of WALA’s git repository.
Importing using the “Existing Gradle Project” wizard creates these
Eclipse project configuration files automatically based on the
underlying Gradle configuration.
Therefore, when working with Eclipse + Gradle, you should treat
.classpath and .project files as generated artifacts, not as
files to edit directly or through the Eclipse project configuration
GUI. For example, avoid using the Java Build Path settings dialog to
make changes that are stored in a .classpath file: the modified
.classpath file is not git-tracked, so your changes will eventually
be lost or overwritten.
The right way to change the contents of any of a .classpath or
.project file is to change the Gradle configuration such that the
generated .classpath and .project files will have the desired
contents, likely by using Gradle’s eclipse
plugin.
A few WALA sub-projects already use this: look for eclipse.project
in */build.gradle for examples.
WALA comes preconfigured as an openable IntelliJ IDEA project. Open
the top-level WALA directory as a project; it should have a
distinctive badge on its folder icon marking it as a directory
containing a recognized IntelliJ IDEA project. Do not open the
top-level WALA build.gradle in that directory as a project: this
will begin an “Import Project from Gradle” process that is not the
recommended way to bring WALA up in IntelliJ IDEA.
The first time you open the WALA project, IntelliJ IDEA will synchronize its project model with the Gradle build configuration, including downloading some large supporting libraries. After the lengthy import process completes, use “Run → Run… → After Opening Pristine Project” to perform some post-import configuration. This also can take tens of minutes, but is only necessary in a clean, never-previously-built tree. Immediately after importing, you may see some Java compilation errors in the IntelliJ IDEA “Messages” view. These should all go away after running the “After Opening Pristine Project” step. Be patient during the initial project open and also during the “After Opening Pristine Project” step, especially if you have a slow network connection.
If you prefer a non-interactive approach, instead of using “After
Opening Pristine Project” you can run ./gradlew prepareIntelliJIDEA
either before or after opening WALA in IntelliJ IDEA.
Each time you open the WALA project, IntelliJ IDEA may report “Non-managed pom.xml file found” in its event log. This arises because WALA supports both Gradle and Maven, but WALA in IntelliJ IDEA needs only the Gradle configuration. You can safely ignore this notification, permanently disable it using the offered “Disable notification” link, or even disable the IntelliJ IDEA Maven plugin entirely if you have no other need for it.
IntelliJ IDEA automatically derives its project models from the Gradle build configuration, including all information about both internal and external build dependencies. However, this synchronization only goes in one direction: from Gradle to IntelliJ IDEA, not from IntelliJ IDEA back into Gradle. If you manipulate the project structure using the IntelliJ IDEA’s user interface, your changes will likely be overwritten the next time IntelliJ IDEA scans the Gradle build configuration.
This particularly applies to settings found in the “Modules” and “Libraries” sections of the “Project Structure” dialog. The right way to change module and library settings is to change the Gradle configuration such that the derived IntelliJ IDEA model is what you want it to be.
You do not need to install Gradle separately. WALA includes its own
copy of Gradle, available as the gradlew script in the top-level
WALA directory. Use this script for all command-line Gradle actions.
For example, to compile all of WALA’s main (non-test) code and gather
it into jar archives, run ./gradlew assemble.
In general, most Gradle-generated artifacts will appear somewhere
under */build. For example the jar archives created by the
assemble task can be found as */build/libs/*.jar. Note, however,
that Eclipse-generated artifacts will still appear in the same places
as before, such as */bin and */target.
Gradle has excellent understanding of task and file dependencies. You
can trust it to perform incremental rebuilds rather than always
rebuilding from scratch. If you are used to always running mvn clean compile instead of mvn compile, or mvn clean install instead of
mvn install, I recommend that you drop clean as a reflexive extra
step and trust Gradle to do incremental builds correctly.
Some useful Gradle tasks include:
-
assemble: build WALA’s main (non-test) code -
build: build all WALA code and run all automated tests -
javadoc: build all Javadoc documentation -
publishToMavenLocal: install WALA’s jar files under~/.m2 -
clean: remove all Gradle-generated artifacts
When you run ./gradlew in the top-level WALA directory, any tasks
you list will be built in all sub-projects. For example, ./gradlew assemble builds all non-test WALA jars in all sub-projects. If you
want to build tasks only in specific sub-projects, you have two options:
-
Give the fully-qualified name of the sub-project task. For example, to assemble only the Dalvik jar, you could run
./gradlew :com.ibm.wala.dalvik:assemble. -
Run Gradle from within some sub-project directory. For example, to assemble only the Dalvik jar, you could
cd com.ibm.wala.dalvikand then run../gradlew assemble. Note the proper relative path to the top-level Gradle script:../gradleinstead of./gradlew.
Any build task can be
abbreviated
by shortening each camel-case-delimited word in its name. For
example, the processTestResources task can probably be abbreviated
as procTeRes or even pTR.
Among Gradle’s command-line flags, I have found the following particularly useful:
-
--continue: keep building non-dependent sub-tasks even after an initial failure. Especially useful in conjunction with thebuildortesttasks to see multiple test failures rather than giving up after the first failure. -
-t,--continuous: keep Gradle process running and re-execute the given tasks whenever input files change. Similar to Eclipse’s behavior of updating the build whenever you change and save a file. -
--tests=...: run only the selected tests. Use in conjunction with thebuildortesttasks for faster turnaround if you are focusing on getting just one or a few failing tests to pass. -
--scan: upload a detailed report of the build process to a Gradle-hosted server for further exploration and analysis. The only security here is the obscurity of the generated URL for the build report. If you are not concerned about potentially making your build details public, then--scanis a good way to gain insights into why Gradle did what it did, and how long each piece took.
Gradle’s composite
builds
allow a Gradle-managed project to recursively include other
Gradle-managed projects, with Gradle managing the entire build process
in a coherent, integrated manner. Thus, if you use Gradle to build
your WALA-based project, you can easily have it use WALA from your
own, private WALA tree instead of from ~/.m2 or the public Maven
repository.
This is especially useful if you frequently find yourself switching
between multiple different personal or experimental WALA builds. By
avoiding ~/.m2, each WALA-based project can be its own composite
build, with its own WALA subtree, and no project interferes with any
other.
I use a Travis CI build matrix to perform automated testing in three configurations:
- Gradle build on Ubuntu 14 (Trusty Tahr)
- Maven build on Ubuntu 14 (Trusty Tahr)
- Gradle build on macOS 10.12 (Sierra)
Until we are ready to completely replace Maven with Gradle, it is
important that both keep working. Therefore, I use Travis CI to build
and test WALA on Ubuntu using both Gradle and Maven. Every new pull
request must be validated in both of these configurations before I
will accept it onto the gradle-and-buildship branch.
The official WALA repository has no macOS CI testing. However, macOS is the main development platform for at least one WALA maintainer, so it is great to have Travis CI helping us keep that platform working. I will not accept pull requests that introduce regressions into Gradle macOS builds. However, I am not using Travis CI to test Maven macOS builds. Initial attempts using the official WALA master sources failed. As it is my goal to replace Maven entirely, investigating Maven+macOS failures further is not a priority.