JacoDB
is a pure Java library that allows you to get information about Java bytecode outside the JVM process and to store it in a database. While Java Reflection
makes it possible to inspect code at runtime, JacoDB
does the same for bytecode stored in a file system.
JacoDB
uses ASM framework for reading and parsing java bytecode.
Information about classes, hierarchies, annotations, methods, fields, and their usages is stored in SQLite database — either in-memory or persistent. Persisted data can be reused between restarts. Accessing the persistent storage from multiple processes simultaneously is not supported.
API has two levels: the one representing in filesystem (bytecode and classes) and the one appearing at runtime (types).
- bytecode and classes — represent data from
class
files: class with methods, fields, etc. - types — represent types that can be nullable, parameterized, etc.
Both levels are connected to JcClasspath
. You can't modify classes retrieved from pure bytecode. types may be constructed manually by generics substitution.
Java
class Example {
public static MethodNode findNormalDistribution() throws Exception {
File commonsMath32 = new File("commons-math3-3.2.jar");
File commonsMath36 = new File("commons-math3-3.6.1.jar");
File buildDir = new File("my-project/build/classes/java/main");
JcDatabase database = JacoDB.async(
new JcSettings()
.useProcessJavaRuntime()
.persistent("/tmp/compilation-db/" + System.currentTimeMillis()) // persist data
).get();
// Let's load these three bytecode locations
database.asyncLoad(Arrays.asList(commonsMath32, commonsMath36, buildDir));
// This method just refreshes the libraries inside the database. If there are any changes in libs then
// the database updates data with the new results.
database.asyncLoad(Collections.singletonList(buildDir));
// Let's assume that we want to get bytecode info only for `commons-math3` version 3.2.
JcClassOrInterface jcClass = database.asyncClasspath(Arrays.asList(commonsMath32, buildDir))
.get().findClassOrNull("org.apache.commons.math3.distribution.NormalDistribution");
System.out.println(jcClass.getDeclaredMethods().size());
System.out.println(jcClass.getAnnotations().size());
System.out.println(JcClasses.getConstructors(jcClass).size());
// At this point the database read the method bytecode and return the result.
return jcClass.getDeclaredMethods().get(0).body();
}
}
Kotlin
suspend fun findNormalDistribution(): Any {
val commonsMath32 = File("commons-math3-3.2.jar")
val commonsMath36 = File("commons-math3-3.6.1.jar")
val buildDir = File("my-project/build/classes/java/main")
val database = jacodb {
useProcessJavaRuntime()
persistent("/tmp/compilation-db/${System.currentTimeMillis()}") // persist data
}
// Let's load these three bytecode locations
database.load(listOf(commonsMath32, commonsMath36, buildDir))
// This method just refreshes the libraries inside the database. If there are any changes in libs then
// the database updates data with the new results.
database.load(listOf(buildDir))
// Let's assume that we want to get bytecode info only for `commons-math3` version 3.2.
val jcClass = database.classpath(listOf(commonsMath32, buildDir))
.findClass("org.apache.commons.math3.distribution.NormalDistribution")
println(jcClass.declaredMethods.size)
println(jcClass.constructors.size)
println(jcClass.annotations.size)
// At this point the database read the method bytecode and return the result.
return jcClass.methods[0].body()
}
Note: the body
method returns null
if the to-be-processed JAR-file was changed or removed. Class could be in incomplete environment (i.e super class, interface, return type or parameter of method is not found in classpath) then api will throw NoClassInClasspathException
at runtime.
The database can watch for file system changes in the background and refresh the JAR-files explicitly:
Java
public static void watchFileSystem() throws Exception {
JcDatabase database = JacoDB.async(new JcSettings()
.watchFileSystem()
.useProcessJavaRuntime()
.loadByteCode(Arrays.asList(lib1, buildDir))
.persistent("", false)).get();
}
// A user rebuilds the buildDir folder.
// The database re-reads the rebuilt directory in the background.
Kotlin
val database = jacodb {
watchFileSystem()
useProcessJavaRuntime()
loadByteCode(listOf(lib1, buildDir))
persistent("")
}
// A user rebuilds the buildDir folder.
// The database re-reads the rebuilt directory in the background.
type can be represented as one of
- primitives
- classes
- arrays
- bounded and unbounded wildcards
It represents runtime behavior according to parameter substitution in the given generic type:
Java
public static class A<T> {
T x = null;
}
public static class B extends A<String> {
}
public static void typesSubstitution() {
JcClassType b = (JcClassType)classpath.findTypeOrNull("org.jacodb.examples.JavaReadMeExamples.B");
JcType xType = b.getFields()
.stream()
.filter(it -> "x".equals(it.getName()))
.findFirst().get().getFieldType();
JcClassType stringType = (JcClassType) classpath.findTypeOrNull("java.lang.String");
System.out.println(xType.equals(stringType)); // will print `true`
}
Kotlin
open class A<T>(val x: T)
class B(x: String): A<String>(x)
suspend fun typesSubstitution() {
val db = jacodb {
loadByteCode(listOf(File("all-classpath")))
}
val classpath = db.classpath(listOf(File("all-classpath")))
val b = classpath.findClass<B>().toType()
println(b.fields.first { it.name == "x"}.fieldType == classpath.findClass<String>().toType()) // will print `true`
}
The instances of JcClassOrInterface
, JcMethod
, and JcClasspath
are thread-safe and immutable.
JcClasspath
represents an independent snapshot of classes, which cannot be modified since it is created. Removing or modifying library files does not affect JcClasspath
instance structure. The JcClasspath#close
method releases all snapshots and cleans up the persisted data if some libraries are outdated.
Java
public static void refresh() throws Exception {
JcDatabase database = JacoDB.async(
new JcSettings()
.watchFileSystem()
.useProcessJavaRuntime()
.loadByteCode(Arrays.asList(lib1, buildDir))
.persistent("...")
).get();
JcClasspath cp = database.asyncClasspath(Collections.singletonList(buildDir)).get();
database.asyncRefresh().get(); // does not affect cp classes
JcClasspath cp1 = database.asyncClasspath(Collections.singletonList(buildDir)).get(); // will use new version of compiled results in buildDir
}
Kotlin
val database = jacodb {
watchFileSystem()
useProcessJavaRuntime()
loadByteCode(listOf(lib1, buildDir))
persistent("")
}
val cp = database.classpath(listOf(buildDir))
database.refresh() // does not affect cp classes
val cp1 = database.classpath(listOf(buildDir)) // will use new version of compiled results in buildDir
If there is a request for a JcClasspath
instance containing the libraries, which haven't been indexed yet, the indexing process is triggered and the new instance of the JcClasspath
set is returned.
Java
public static void autoProcessing() throws Exception {
JcDatabase database = JacoDB.async(
new JcSettings()
.loadByteCode(Arrays.asList(lib1))
.persistent("...")
).get();
JcClasspath cp = database.asyncClasspath(Collections.singletonList(buildDir)).get(); // database will automatically process buildDir
}
Kotlin
val database = jacodb {
loadByteCode(listOf(lib1))
persistent("")
}
val cp = database.classpath(listOf(buildDir)) // database will automatically process buildDir
JacoDB
is thread-safe. If one requests JcClasspath
instance while loading JAR-files from another thread,
JcClasspath
can represent only a consistent state of the JAR-files being loaded. It is the completely loaded
JAR-file that appears in JcClasspath
. Please note: there is no guarantee that all the JAR-files, submitted for loading, will be actually loaded.
Java
class Example {
public static void main(String[] args) {
val db = JacoDB.async(new JcSettings()).get();
new Thread(() -> db.asyncLoad(Arrays.asList(lib1, lib2)).get()).start();
new Thread(() -> {
// maybe created when lib2 or both are not loaded into database
// but buildDir will be loaded anyway
var cp = db.asyncClasspath(buildDir).get();
}).start();
}
}
Kotlin
val db = jacodb {
persistent("")
}
thread(start = true) {
runBlocking {
db.load(listOf(lib1, lib2))
}
}
thread(start = true) {
runBlocking {
// maybe created when lib2 or both are not loaded into database
// but buildDir will be loaded anyway
val cp = db.classpath(listOf(buildDir))
}
}
Bytecode loading consists of two steps:
- retrieving information about the class names from the JAR-files or build directories
- reading classes bytecode from the JAR-files or build directories and processing it (persisting data, setting up
JcFeature
implementations, etc.)
JacoDB
or JcClasspath
instances are returned right after the first step is performed. You retrieve the final representation of classes during the second step. It is possible that the .class
files undergo changes at some moment between the first step and the second, and classes representation is affected accordingly.
Here are Benchmarks results and Soot migration.