To run each example use: java --enable-preview --source 22 <FileName.java>

• 423 - Region Pinning for G1
• 447 - Statements before super() (Preview)
• 454 - Foreign Function & Memory API
• 456 - Unnamed Variables & Patterns
• 457 - Class-File API (Preview)
• 458 - Launch Multi-File Source-Code Programs
• 459 - String Templates (Second Preview)
• 460 - Vector API (Seventh Incubator)
• 461 - Stream Gatherers (Preview)
• 462 - Structured Concurrency (Second Preview)
• 463 - Implicity Declared Classes and Instance Main Methods (Second Preview)
• 464 - Scoped Values (Second Preview)

• Region Pinning for G1
  • goal: "reduce latency by implementing region pinning in G1, so that garbage collection need not be disable during JNI critial regions"
  • avoid worst case cenarios where the application is stoped for minutes, unnecessary out-of-memory conditions due to thread starvation, premature VM shutdown.
• Statements before super()
  • allow statements that do not reference the instance being created to appear before an explicit constructor invocation
  • will allow to use statements that use, transform or verify values before call super
  • the code before the super statement lives in pre-construction context:
    • can perform any statements that don't use any member of the instance being created (or its hierarchy or outter/inner classes that depends on the instance)
  • statements:
    • cannot access any instance member of the class or its parent class
    • can access static members
    • if the class is an inner class, it can access members of its enclosing class (like Outer.this.field++)
      • the outer class instance already exists
• Foreign Function & Memory API
  • promotion to standard
  • API to allow Java to interoperate with code and data outside of JVM;
  • will replace JNI, allowing efficiently invoking foreign functions and safely accessing foreign memory;
  • goals:
    • productivity: replace native methods and the JNI with a concise, readable, and pure-Java API;
    • performance: provide access to foreign functions and memory with overhead comparable or better than JNI and sun.misc.Unsafe;
    • broad platform support: enable the discovery and invocation of native libraries on every platform where the JVM runs;
    • uniformity: provide ways to operate on structured and unstructured data, of unlimited size, in multiple kinds of memory (e.g., native memory, persistent memory, and managed heap memory).
    • soundness: guarantee no use-after-free (dangling pointers) bugs, even when memory is allocated and deallocated across multiple threads.
    • integrity: allow programs to perform unsafe operations with native code and data, but warn users about such operations by default.
  • FFM API defines classes and interfaces (in package java.lang.foreign) to:
    • control the allocation and deallocation of foreign memory: MemorySegment, Arena, SegmentAllocator;
    • manipulate and access structured foreign memory: MemoryLayout, VarHandle;
    • call foreign functions: Linker, SymbolLookup, FunctionDescriptor, MethodHandle.
  • Memory segments and arenas:
    • memory segment is a abstraction backed by a contiguous region of memoery (off or on-heap);
    • can be:
      • native segment allocated from a off-heap memory;
      • mapped segment wrapped around a region of mapped off-heap memory;
      • array of buffer segment wrapped around a region of on-heap memory.
    • memory segment has spatial em temporal bounds:
      • spatial bound: guarantee that an access beyond the memory size won't be allowed;
      • temporal bound: guarantee that an access won't be allowed to a memory segment that its backing region was already deallocated.
    • arenas:
      • memory segment is created from an arena
      • arena controls the lifecycle of native memory segments
      • every memory segment allocated from the same arena will have the same temporal bounds
      • types:
        • global arena:
          • arena with unbounded lifetime, it is always alive
          • MemorySegment data = Arena.global().allocate(100); // allocates 100 bytes
        • automatic arena:
          • provides a bounded lifetime with non-deterministic lifetime
          • a segment allocated by a automatic arena can be accessed until JVM's GC detects that the memory segment is unreachable, then it is deallocated
          • MemorySegment data = Arena.ofAuto().allocate(100); // allocates 100 bytes
          • if it is allocated in a block, after that block the memory will be deallocated:

            • void processData() {
                MemorySegment data = Arena.ofAuto().allocate(1000);
                // process data in memory
              } // from here the allocated memory will be released (as long as there isn't a memory leak)
          • it's lifetime is non-deterministic because the JVM will deallocate at some point
        • confined arena:
          • provides a bounded lifetime with deterministic lifetime
          • can be explicity closed by the code
          • can only be accessed by one thread
          • we can use in a try-with-resources:

            • try (Arena confinedArena = Arena.ofConfined()) {
                  MemorySegment input = confinedArena.allocate(100);
                  MemorySegment output = confinedArena.allocate(100);
              } // will be deallocated from here
        • shared arena:
          • is a confined arena for multi-threading
          • any thread can access and close the memory segment
    • allocators:
      • is an abstraction to define operations to allocate and initialize memory segments
      • memory allocation can be bottleneck when using off-heap memory, allocators help to minimize this
      • we can allocate a large memory and then use allocator to distribute through the application
      • we can create an allocator using an already allocated memory segment
        • the segments allocated by the allocator will have the same bounded lifetime

      • MemorySegment segment = Arena.ofAuto().allocate(1024 * 1024 * 1024); // 1 GB
        SegmentAllocator allocator = SegmentAllocator.slicingAllocator(segment);
        for (int i = 0; i < 10; i++) {
            MemorySegment msi = allocator.allocateFrom(ValueLayout.JAVA_INT, i);
        }
  • Memory layout:
    • Value layout:
      • abstraction to facilitate the memory value layout usage
        • eg.: to work with int, we can use ValueLayout.JAVA_INT to read 4 bytes using the platform endianness to correctly extract an int from a memory segment
        • MemorySegment int = Arena.global().allocateFrom(ValueLayout.JAVA_INT, 42)
        • MemorySegment intArray = Arena.global().allocateFrom(ValueLayout.JAVA_INT, 0, 1, 2, 3, 4, 5)
        • MemorySegment text = Arena.global().allocateFrom("Hello World!")
      • memory segments have simple dereference methods to read values from and write values to memory segments, these methods accept a value layout
      • example how to write value from 1 to 10 in a memory segment:

        • MemorySegment segment = Arena.ofAuto().allocate(
              ValueLayout.JAVA_INT.byteSize() * 10, // memory size
              ValueLayout.JAVA_INT.byteAlignment() // memory alignment
          );
          for (int i = 0; i < 10; i++) {
              int value = i + 1;
              // the memory offset is calculated from: value layout byte size * index
              segment.setAtIndex(ValueLayout.JAVA_INT, i, value);
          }
    • Structured access:
      • the API provides ways to declare any memory layout and also factory method to help calculate the memory size for us
      • eg.: declare a C struct in Java:
        • C code to declare an array of 10 points:

        struct Point { int x; int y } pts[10];
        • we can manually declare the memory layout:

        MemorySegment segment = Arena.ofAuto().allocate(
            2 * ValueLayout.JAVA_INT.byteSize() * 10,
            ValueLayout.JAVA_INT.byteAlignment()
        );
        • we can use memory layout to describe the content of a memory segment and use VarHandle to write the values in the struct:

        SequenceLayout ptsLayout = MemoryLayout.sequenceLayout(
            10, // size
            MemoryLayout.structLayout( // declare a C struct
                ValueLayout.JAVA_INT.withName("x"),
                ValueLayout.JAVA_INT.withName("y")
            )
        );
        MemorySegment segment = Arena.ofAuto().allocate(ptsLayout);
  • Foreign functions:
    • native library loaded from a lookup doesn'n belong to any class loader (unlike JNI), allowing the native library to be reloaded by any other class
      • JNI requires the native library to be loaded by only one class loader
    • FFM API doesn't provides any function for native code to access the Java environment (unlike JNI)
      • but we can pass a Java code as function pointer to native function
    • FFM API uses MethodHandle to interoperate between Java code and native function
    • steps need to call a foreign function:
      • find the address of a given symbol in a loaded native library
      • link the Java code to a foreign function
    • Symbol lookup:
      • we can use SymbolLookup to lookup a function address:
        • SymbolLookup::libraryLookup(String, Arena): creates a library lookup, loads the libreary and associates with the given Arena object;
        • SymbolLookup::loaderLookup(): creates a loader lookup that locates all the symbols in all the native libraries that have been loaded by classes in the current class loader (System::loadLibrary and System::load);
        • Linker::defaultLookup(): creates a default lookup that locates all the symbols that are commonly used in the native platform (OS).
      • we can use SymbolLookup::find(String) to find a function by its name
        • if it is present then a memory segment, which points to funtion's entry point, is returned
    • Linking Java code to foreign function:
      • interface Linker is the main point to allow Java code to interoperate with native code
      • calls:
        • downcall: call from Java code to native code;
        • upcall: call from native code back to Java code (linker a function pointer).
      • the native linker conforms to the Application Binary Interface (BNI) of the native platform
        • ABNI specifies the calling convension, the size, alignment, endianness of scalar types and others details
        • Linker linker = Linker.nativeLinker()
      • downcall:
        • we use Linker::downcallHandle to link Java code to a foreign function
        • Linker::downcallHandle(MemorySegment, FunctionDescriptor) returns a MethodHandle to be used to invoke the native function
          • we must provide a FunctionDescriptor to define the native function signature (return type and the parameters types)
            • we can create one using FunctionDescriptor.of or FunctionDescriptor.ofVoid passing the memory layout to define the signature
            • developer must be aware of the current native platform that will be used (size of scalar types used in C functions)
              • eg.: on Linux and macOS, a long is JAVA_LONG; on Windows, a long is JAVA_INT
            • we can use Linker::canonicalLayouts() to see the association between scalar C types and Java's ValueLayout
          • JVM guarantee that the functions arguments used in MethodHandle will match the FunctionDescriptor used to downcall the function
          • this way our types will be verified in the Java level
        • if the native function returns a by-value struct, we must provide an additional SegmentAllocator argument that will be used to allocate the memory to hold the struct returned
        • if the native function allocates a memory and returns a pointer to it, a zero-length memory segment is return to Java code
          • because the JVM cannot guarantee the allocated memory size off-heap
          • the client must call MemorySegment::reinterpret to tell the JVM the memory's size (the user may not know)
          • this is an unsafe operation (could crash the JVM or leave the memory in a corruption stat)
      • upcall:
        • we can pass Java code as a function pointer to be called by the foreign function
        • we use Linker::upcallStub to "externalize" a Java code
          • we must provide a MethodHandle that points to the Java method, a FunctionDescriptor with the method signature
• Unnamed Variables and Patterns
  • promotion to standard
  • no change from JDK 2
• Class-File API
  • provide standard API for parsing, generating and transforming Java class file
• Launch Multi-File Source-Code Programs
  • enhance the Java launcher's source-file mode to be able to run a program made by multiple Java files
  • the launcher will compile the given Java file and any other Java file that is referenced by the program
  • the referenced class will only be compiled in memory when the class is used
    • any compiler error in the referenced class will be thrown after the program started the execution
  • we can also used pre-compiled classes or module path:
    • java --class-path '*' MyProgram.java
    • java -p . MyProgram.java
  • limitations:
    • annotation processing is disabled (--proc:none)
    • is not possible to run a source-code program whose Java files span multiple modules
• String Templates
  • minor change from JDK 21
  • changed the type of template expressions
• Stream Gatherers
  • enhance the Stream API to support custom intermediate operations
  • will allow stream pipelines to transform data more easily than the existing built-in intermediate operations
  • some built-in intermediate operations: mapping, filtering, reduction, sorting
  • the goal is to provide an extension point (like the one implemented in Stream::collect(Collector))
  • Stream::gather(Gatherer) is an intermediate stream operation
    • it processes the elements of a stream by applying a user-defined entity called a gatherer
    • a gatherer represents a transform of the elements of a stream
    • can transform elements: one-to-one, one-to-many- many-to-one, many-to-many
    • it can keep track previously seen elements in order to compute some transformation of later elements
    • a gatherer will only be evaluated in parallel if it provides a combiner function
  • gatherer is defined by four functions:
    • initializer (optional): an object that maintains private state while processing the stream, the type is Supplier.
    • integrator: integrates a new element from the input stream, also can inspect the private state object, emit elements to the output stream, terminate the processing (by returning false) and so on.
    • combiner (optional): used to evaluate the gatherer in parallel or sequentially (when the operation cannot be parallelized).
    • finisher: invoked when there are no more input elements to consume, can inspect private state object, emit additional output elements.
  • Stream::gather performs the equivalent of the following steps:
    • create a java.util.stream.Gatherer.Downstream object passes the result (object of gatherer's output type) to the next stage in the pipeline;
    • obtain the gatherer's private state object from initializer method get();
    • obtain the gatherer's integrator to process the stream by invoking the method integrator();
    • while there are more inputs elements, invoke the integrator method integrate passing state object, next element and downstream object, terminate if returned false;
    • obtain the gatherer's finisher and invoke it passing the state and downstream object.
  • there are built-in gatherers provided in java.util.stream.Gatherers:
    • fold: stateful many-to-one gatherer;
    • mapConcurrent: stateful one-to-one gatherer which invokes a supplied function for each element concurrently;
    • scan: stateful one-to-one gatherer which applies a supplied function to the current state and the current element to produce the next element to downstream;
    • windowFixed: stateful many-to-many gatherer which groups elements into lists of a supplied size and emit the window to downstream;
    • windowSliding: like the windowFixed but applying sliding in the stream elements (drop the first element from the previous window and added the current elemenet).
    • peek: stateless one-to-one gatherer which applies a function to each element in the stream;
  • is possible to composing gatherers with andThen(Gatherer):
    • stream.gather(a).gather(b).collect(toList()) is equivalent to stream.gather(a.andThen(b)).collect(toList())
• Structured Concurrency
  • no change from JDK 20/21
  • re-preview for additional feedback
• Implicity Declared Classes and Instance Main Methods
  • minor change from JDK 21
  • changed the concept name from unnamed class to implicitly declared class
    • "source file without an enclosing class declaration is said to implicitly declare a class with a name chosen by the host system"
  • changed the procedure for selecting a main method to invoke
    • first it looks for a method main(String[]), if not found then it looks for a method main()
• Scoped Values
  • no change from JDK 20/21
  • re-preview for additional feedback

• JDK 22 - JEP Dashboard
• Gathering the Streams
• Presentations:
  • Foreign Function & Memory API - A (Quick) Peek Under the Hood