The -XshowSettings launcher has a new security category. Settings from security properties, security providers and TLS related settings are displayed with this option. A security sub-category can be passed as an argument to the security category option. See the output from java -X:

   -XshowSettings:security

       show all security settings and continue
   -XshowSettings:security:*sub-category*
       show settings for the specified security sub-category and continue. Possible *sub-category* arguments for this option include:
       all: show all security settings and continue
       properties: show security properties and continue
       providers: show static security provider settings and continue
       tls: show TLS related security settings and continue

Third party security provider details will be reported if they are included in the application class path or module path and such providers are configured in the java.security file.

The Code Root Scan Phase during garbage collection finds references to Java objects in compiled code. To speed up this process, G1 maintains a remembered set for compiled code that contains references into the Java heap. That is, every region contains a set of compiled code that contains references into it.

Assuming that such references are few, previous code used a single thread per region to iterate over a particular region's references, which poses a scalability bottleneck if the distribution of these references is very unbalanced.

G1 now distributes this code root scan work across multiple threads within regions, removing this bottleneck.

During a young collection, ParallelGC partitions the old generation into 64kB stripes when scanning it for references into the young generation. These stripes are assigned to worker threads that do the scanning in parallel as work units.

Before this change, Parallel GC always scanned these stripes completely even if only a small part had been known to contain interesting references. Additionally, every worker thread processed the objects that start in that stripe by itself, including parts of objects that extend into other stripes. This behavior limited parallelism when processing large objects. A single large object, potentially containing thousands of references, had been scanned by a single thread only and in full. This would cause bad scaling due to memory sharing and cache misses in the subsequent long, work stealing phase.

With this change, Parallel GC workers limit work to their stripe and only process interesting parts of large object arrays. This reduces the work done by a single thread for a stripe, improves parallelism, and reduces the amount of work stealing. Parallel GC pauses are now on par with G1 in presence of large object arrays, reducing pause times by 4-5 times in some cases.

Fixed an issue on Debian Linux distros where jpackage could not always build an accurate list of required packages from shared libraries with symbolic links in their paths, causing installations to fail due to missing shared libraries.

Delete nonfunctional desktop integration functionality from Linux installers. The installers will stop depositing files in /usr/share/icons, /usr/share/mime, and /usr/share/applications subtrees.

The following root certificates have been added to the cacerts truststore:

+ GlobalSign

  + globalsignr46
    DN: CN=GlobalSign Root R46, O=GlobalSign nv-sa, C=BE

+ GlobalSign
  + globalsigne46
    DN: CN=GlobalSign Root E46, O=GlobalSign nv-sa, C=BE

Native executables and libraries on Linux have switched to using RPATH instead of RUNPATH in this release.

JDK native executables and libraries use embedded runtime search paths to locate other internal JDK native libraries. On Linux these can be defined as either RPATH or RUNPATH. The main difference is that the dynamic linker considers RPATH before the LD_LIBRARY_PATH environment variable, while RUNPATH is only considered after LD_LIBRARY_PATH.

By making the change to using RPATH, it is no longer possible to replace JDK internal native libraries using LD_LIBRARY_PATH.

The installation directory name of the Oracle JDK in RPM and DEB packages has changed from /usr/lib/jvm/jdk-${FEATURE}-oracle-${ARCH} to /usr/lib/jvm/jdk-${VERSION}-oracle-${ARCH}.

Every update release will be installed in a separate directory on Linux platform.

Installers will create a /usr/java/jdk-${FEATURE}-oracle-${ARCH} link pointing to the installation directory to allow programs to find the latest JDK version in the ${FEATURE} release train.

The XML Signature implementation has been updated to Santuario 3.0.3. Support for four new SHA-3 based RSA-MGF1 signature methods have been added: SHA3_224_RSA_MGF1, SHA3_256_RSA_MGF1, SHA3_384_RSA_MGF1, and SHA3_512_RSA_MGF1. While these new algorithm URIs are not defined in javax.xml.crypto.dsig.SignatureMethod in the JDK update releases, they may be represented as string literals in order to be functionally equivalent. SHA-3 hash algorithm support was delivered to JDK 9 via JEP 287. Releases earlier than that may use third party security providers.

There is an issue on Debian Linux distros where jpackage could not always build an accurate list of required packages from shared libraries with symbolic links in their paths, causing installations to fail due to missing shared libraries.

The java.awt.SystemTray API is used for notifications in a desktop taskbar and may include an icon representing an application. On Linux, the Gnome desktop's own icon support in the taskbar has not worked properly for several years due to a platform bug. This, in turn, has affected the JDK's API, which relies upon that.

Therefore, in accordance with the existing Java SE specification, java.awt.SystemTray.isSupported() will return false where ever the JDK determines the platform bug is likely to be present.

The impact of this is likely to be limited since applications always must check for that support anyway. Additionally, some distros have not supported the SystemTray for several years unless the end-user chooses to install non-bundled desktop extensions.

The following root certificates have been added to the cacerts truststore:

+ Certainly

  + certainlyrootr1
    DN: CN=Certainly Root R1, O=Certainly, C=US

+ Certainly
  + certainlyroote1
    DN: CN=Certainly Root E1, O=Certainly, C=US

JDK 21 introduced pattern matching in the Java language. However, javac allowed final in front of a record pattern, such as (case final R(...) ->), something which is not allowed by the Java Language Specification.

Programs that could be compiled erroneously in JDK21 with final will now fail to compile. This change fixes the issue in the compiler. Impacted users will need to remove the final keyword.

The java.net.ExtendedSocketOptions TCP_KEEPIDLE and TCP_KEEPINTERVAL are supported on Windows platforms starting from Windows 10 version 1709 and onwards. TCP_KEEPCOUNT is supported starting from Windows 10 version 1703 and onwards.

On Linux (kernel version 4.11 or newer, glibc version 2.28 or newer), BasicFileAttributes.creationTime now returns the creation or birth time of the file through the stx_btime element of struct statx. In previous JDK releases, the last modified time was returned. If the birth time is supported by the platform but not by the file system, then the epoch 1970-01-01T00:00:00Z is returned.

When the C1 compiler is the only compiler available to the VM, it applies loop predication to remove array access range checks from loop bodies. Due to a defect, this optimization was disabled, potentially leading to a performance regression.

This only affects the client VM or VM's running with the non-default command line flags -XX:+NeverActAsServerClassMachine or -XX:TieredStopAtLevel=[1,2,3].

The JDK 21 issue that could potentially lead to JVM crashes or incorrect execution when running the JVM with -XX:+UseZGC and non-default value of -XX:ObjectAlignmentInBytes has been resolved, and it is possible again to use this combination of JVM options.

The following root certificates have been added to the cacerts truststore:

+ DigiCert, Inc.

  + digicertcseccrootg5
    DN: CN=CN=DigiCert CS ECC P384 Root G5, O="DigiCert, Inc.", C=US

+ DigiCert, Inc.
  + digicertcsrsarootg5
    DN: CN=DigiCert CS RSA4096 Root G5, O="DigiCert, Inc.", C=US

+ DigiCert, Inc.
  + digicerttlseccrootg5
    DN: DigiCert TLS ECC P384 Root G5, O="DigiCert, Inc.", C=US

+ DigiCert, Inc.
  + digicerttlsrsarootg5
    DN: DigiCert TLS RSA4096 Root G5, O="DigiCert, Inc.", C=US

The following root certificates have been added to the cacerts truststore:

+ eMudhra Technologies Limited

  + emsignrootcag1
    DN: CN=emSign Root CA - G1, O=eMudhra Technologies Limited, OU=emSign PKI, C=IN

+ eMudhra Technologies Limited
  + emsigneccrootcag3
    DN: CN=emSign ECC Root CA - G3, O=eMudhra Technologies Limited, OU=emSign PKI, C=IN

+ eMudhra Technologies Limited
  + emsignrootcag2
    DN: CN=emSign Root CA - G2, O=eMudhra Technologies Limited, OU=emSign PKI, C=IN

The following root certificate has been added to the cacerts truststore:

+ Telia Root CA v2

  + teliarootcav2
    DN: CN=Telia Root CA v2, O=Telia Finland Oyj, C=FI

The following root certificate has been added to the cacerts truststore:

+ Let's Encrypt

  + letsencryptisrgx2
    DN: CN=ISRG Root X2, O=Internet Security Research Group, C=US

NMT reports will now show peak values for all categories. Peak values contain the highest value for committed memory in a given NMT category over the lifetime of the JVM process.

If the committed memory for an NMT category is currently at peak, NMT prints "at peak"; otherwise, it prints the peak value.

For example:

- Compiler (reserved=230KB, committed=230KB)
                            (malloc=34KB #64) (peak=49KB #71)
                            (arena=196KB #4) (peak=6126KB #16)
							

This shows Compiler arena memory peaked at a bit more than 6MB, whereas it now hovers around 200 KB.

On Linux, if the JVM is started with +UseTransparentHugePages but the system does not support Transparent Huge Pages, a warning will now be printed to stdout:

UseTransparentHugePages disabled; transparent huge pages are not supported by the operating system.

A section containing the thread local lock stack has been added to hs_err report files. It only gets printed when the new lightweight locking mode is enabled (-XX:LockingMode=2).

An example is given here with details about the locked objects omitted:

Lock stack of current Java thread (top to bottom):

LockStack[1]: nsk.share.jdi.EventHandler 
...
LockStack[0]: java.util.Collections$SynchronizedRandomAccessList
...

It lists objects which are lightweight locked, through synchronized methods or statements, by the Java thread which is being analyzed. The object which has been locked most recently is printed first. Objects which are not lightweight locked are not displayed in this section.

The fix for JDK-8302017 updated the RSA signature verification algorithm for compliance with RFC 8017. However, this modification introduced a regression: signatures not strictly conforming to RFC 8017 may fail verification. This issue will be addressed in a forthcoming update. For further information, refer to JDK-8320597.

The following root certificate has been added to the cacerts truststore:

+ Certigna (Dhimyotis)

  + certignarootca
    DN: CN=Certigna Root CA, OU=0002 48146308100036, O=Dhimyotis, C=FR

The system property, jdk.jar.maxSignatureFileSize, allows applications to control the maximum size of signature files in a signed JAR. Its default value has been increased from 8000000 bytes (8 MB) to 16000000 bytes (16 MB).

Processes started by Runtime.exec and ProcessBuilder can be enabled to log the command, arguments, directory, stack trace, and process id. The exposure of this information should be reviewed before implementation. Logging of the information is enabled when the logging level of the System#getLogger(String) named java.lang.ProcessBuilder is System.Logger.Level.DEBUG or Logger.Level.TRACE. When enabled for Level.DEBUG, only the process id, directory, command, and stack trace are logged. When enabled for Level.TRACE, the command arguments are included with the process id, directory, command, and stack trace.

Calls to java.lang.System.exit() and Runtime.exit() are logged to the logger named java.lang.Runtime with a logging level of System.Logger.DEBUG. When the configuration of the logger allows, the caller can be identified from the stack trace included in the log.

The methods Math.clamp() and StrictMath.clamp() are added to conveniently clamp the numeric value between the specified minimum and maximum values. Four overloads are provided in both Math and StrictMath classes for int, long, float, and double types. A clamp(long value, int min, int max) overload can also be used to safely narrow a long value to int.

Two new methods indexOf(int ch, int beginIndex, int endIndex) and indexOf(String str, int beginIndex, int endIndex) are added to java.lang.String to support forward searches of character ch, and of String str, respectively, and limited to the specified range of indices.

Besides full control on the search range, they are safer to use than indexOf(int ch, int fromIndex) and indexOf(String str, int fromIndex), respectively, because they throw an exception on illegal search ranges.

Method indexOf(int ch, int beginIndex, int endIndex) is covered by JDK-8302590, and method indexOf(String str, int beginIndex, int endIndex) is covered by JDK-8303648.

The following six new methods are added to java.lang.Character for obtaining Emoji character properties, which are defined in the Unicode Emoji Technical Standard (UTS #51) :

Unlike the split() methods, these new splitWithDelimiters() methods in java.lang.String and java.util.regex.Pattern return an alternation of strings and matching delimiters, rather than just the strings.

The following methods have been added to the API:

• void close(): closes the client gracefully, waiting for submitted requests to complete.
• void shutdown(): initiates a graceful shutdown, then returns immediately without waiting for the client to terminate.
• void shutdownNow(): initiates an immediate shutdown, trying to interrupt active operations, and returns immediately without waiting for the client to terminate.
• boolean awaitTermination(Duration duration): waits for the client to terminate, within the given duration; returns true if the client is terminated, false otherwise.
• boolean isTerminated(): returns true if the client is terminated.

The instances returned by HttpClient.newHttpClient(), and the instances built from HttpClient.newBuilder(), provide a best effort implementation for these methods. They allow the reclamation of resources allocated by the HttpClient early, without waiting for its garbage collection.

Note that an HttpClient instance typically manages its own pools of connections, which it may then reuse when necessary. Connection pools are typically not shared between HttpClient instances. Creating a new client for each operation, though possible, will usually prevent reusing such connections.

China National Standard body (CESI) has recently published GB18030-2022 which is an updated version of the GB18030 standard and brings GB18030 in sync with Unicode version 11.0. The Charset implementation for this new standard has now replaced the prior 2000 standard. However, this new standard has some incompatible changes from the prior implementation. For those who need to use the old mappings, a new system property jdk.charset.GB18030 is introduced. By setting its value to 2000, the previous JDK releases' mappings for the GB18030 Charset are used which are based on the 2000 standard.

The methods public StringBuilder repeat(int codePoint, int count) and public StringBuilder repeat(CharSequence cs, int count) have been added to java.lang.StringBuilder and java.lang.StringBuffer to simplify the appending of multiple copies of characters or strings. For example, sb.repeat('-', 80) will insert 80 hyphens into the value of the java.lang.StringBuilder sb object.

Emoji-related properties introduced in (JDK-8303018) can now be used as binary properties in the java.util.regex.Pattern class. One can match characters that have Emoji-related properties with the new \p{IsXXX} constructs. For example,

Pattern.compile("\\p{IsEmoji}").matcher("🉐").matches()

returns true.

The Sequenced Collection API introduces several new interfaces into the collections framework, providing enhancements to many existing collections classes. The new API facilitates access to elements at each end of a sequenced collection, and provides the ability to view and iterate such collections in reverse order. See JEP 431 for additional information.

The introduction of new collections interfaces, along with default methods, introduces some compatibility risk, including the possibility of both source and binary incompatibilities. The introduction of default methods in an interface hierarchy may cause conflicts with methods declared on existing classes or interfaces that extend collections interfaces - this could result in either source or binary incompatibilities. The introduction of new interfaces also introduces new types into the system, which can change the results of type inference, leading in turn to source incompatibilities.

For a discussion of potential incompatibilities and possible ways to mitigate them, please see the document JDK 21: Sequenced Collections Incompatibilities.

The Java Virtual Machine (JVM) now prints a warning to standard error when a JVM Tool Interface (JVM TI) agent or Java Agent is dynamically loaded into a running JVM. The warning is intended to prepare for a future release that disallows, by default, dynamic loading of agent code into a running JVM.

Agents are programs that run in the JVM process and make use of powerful JVM TI or java.lang.instrument APIs. These APIs are designed to support tooling such as profilers and debuggers. Agents are started via a command line option, for example -agentlib or -javaagent, or they can be started into a running VM using the JDK specific com.sun.tools.attach API or the jcmd command. Agents loaded into a running VM will now print a warning. There is no warning for agents that are loaded at startup via command line options.

The HotSpot VM option EnableDynamicAgentLoading controls dynamic loading of agents. This option has existed since JDK 9. The default, since JDK 9, is to allow dynamic loading of agents. Running with -XX:+EnableDynamicAgentLoading on the command line serves as an explicit "opt-in" that allows agent code to be loaded into a running VM and thus suppresses the warning. Running with -XX:-EnableDynamicAgentLoading disallows agent code from being loaded into a running VM and can be used to test possible future behavior.

In addition, the system property jdk.instrument.traceUsage can be used to trace uses of the java.lang.instrument API. Running with -Djdk.instrument.traceUsage or -Djdk.instrument.traceUsage=true causes usages of the API to print a trace message and stack trace. This can be used to identify agents that are dynamically loaded instead of being started on the command line with -javaagent.

More information on this change can be found in JEP 451.

Applications running with Generational ZGC should enjoy:

• Lower risks of allocations stalls,
• Lower required heap memory overhead, and
• Lower garbage collection CPU overhead.

Enable Generational ZGC with command line options -XX:+UseZGC -XX:+ZGenerational

For further details, see JEP 439.

A full garbage collection (GC) in the Garbage First (G1) collector now moves humongous objects to avoid Out-Of-Memory situations due to a lack of contiguous space in the Java heap when the application allocates humongous objects.

Previously, G1 failed to allocate those humongous objects, reporting an Out-Of-Memory exception in this situation.

This functionality is a last resort measure, causing a second full GC in the same pause after the previous full GC fails to clear out enough contiguous memory for the allocation.

A new view command has been added to the JFR tool and jcmd. The command can aggregate and display event data in a tabular form without the need to dump a recording file or open JDK Mission Control. There are 70 predefined views, such as hot-methods, gc-pauses, pinned-threads, allocation-by-site, gc, memory-leaks-by-class, and more. A list of available views can be found through using jcmd <pid> JFR.view or jfr view.

This feature delivers an enhanced syntax for properties related to certificate, CRL, and OCSP connect and read timeouts. The new syntax allows the timeout values to be specified either in seconds or milliseconds. This feature also delivers three new System properties related to connect and read timeouts.

New properties: The existing com.sun.security.ocsp.timeout property will now be paired with the new com.sun.security.ocsp.readtimeout property. The former property will be used to set timeouts for the transport-layer connection while the latter will be used to manage timeouts for reading the data. The new com.sun.security.cert.timeout and com.sun.security.cert.readtimeout properties will be used to control connect and read timeouts, respectively, when following an X.509 certificate's AuthorityInfoAccess extension. For the certificate fetching properties, the com.sun.security.enableAIAcaIssuers property must be set to true in order for fetching to occur and these property timeouts to be enabled.

Enhanced timeout syntax: The new syntax applies to the aforementioned properties, and also to the com.sun.security.crl.timeout and com.sun.security.crl.readtimeout properties as well. The allowed syntax is as follows:

• A decimal integer will be interpreted in seconds and ensures backward compatibility.
• A decimal integer ending in "s" (case-insensitive, no space) appended to it. This will also be interpreted in seconds.
• A decimal integer value with "ms" (case-insensitive, no space) appended to it. This will be interpreted as milliseconds. For example, a value of "2500ms" will be a 2.5 second timeout.
• Negative, non-numeric, or non-decimal (for example, hexadecimal values prepended by "0x") values will be interpreted as illegal and will default to the 15 second timeout.
• Whether the value is interpreted in seconds or milliseconds, a value of zero will disable the timeout.

A new standard signature algorithm named "HSS/LMS" has been introduced. The HSS/LMS algorithm is defined in RFC 8554: Leighton-Micali Hash-Based Signatures and NIST Special Publication 800-208. New KeyFactory and Signature implementations are available for the algorithm. The KeyFactory only operates on public keys and the Signature only covers the verification part.

The SunJCE provider is enhanced with additional PBES2 Cipher and Mac algorithms, such as those using SHA-512/224 and SHA-512/256 message digests. To be more specific, callers can now use the SunJCE provider for PBEWithHmacSHA512/224AndAES_128, PBEWithHmacSHA512/256AndAES_128, PBEWithHmacSHA512/224AndAES_256, and PBEWithHmacSHA512/256AndAES_256 Ciphers and PBEWithHmacSHA512/224, and PBEWithHmacSHA512/256 Mac.

The SunPKCS11 security provider now supports Password-Based Cryptography algorithms for Cipher, Mac, and SecretKeyFactory service types. You will find the list of algorithms in JDK-8308719. As a result of this enhancement, SunPKCS11 can now be used for privacy and integrity in PKCS #12 key stores.

A new system property named org.jcp.xml.dsig.secureValidation has been added. It can be used to enable or disable the XML Signature secure validation mode. The system property should be set to "true" to enable, or "false" to disable. Any other value for the system property is treated as "false". If the system property is set, it supersedes the XMLCryptoContext property value.

By default, the secure validation mode is enabled. Disabling the secure validation mode should be done at your own risk.

The XML Signature implementation has been updated to Santuario 3.0.2. The main, new feature is support for EdDSA. One difference is that the JDK still supports the here() function by default. However, we recommend avoiding the use of the here() function in new signatures and replacing existing signatures that use the here() function. Future versions of the JDK will likely disable, and eventually remove, support for this function, as it cannot be supported using the standard Java XPath API. Users can now disable the here() function by setting the security property jdk.xml.dsig.hereFunctionSupported to "false".

String templates allow text and expressions to be composed without using the + operator. The result is often a string, but can also be an object of another type. Each string template has a template processor that validates the text and expressions before composing them, achieving greater safety than basic 'string interpolation' features in other languages.

Unnamed classes and instance main methods enable students to write streamlined declarations for single-class programs and then seamlessly expand their programs later to use more advanced features as their skills grow. Unnamed classes allow the user to provide class content without the full ceremony of the class declaration. The instance main method feature allows the user to drop the formality of public static void main(String[] args) and simply declare void main().

The new lint option, this-escape, has been added to javac to warn about calls to overridable methods in the constructor body which might result in access to a partially constructed object from subclasses.

The new warning can be suppressed using SuppressWarnings("this-escape").

Prior to JDK 21, the javac compiler was overwriting some output files during compilation. This can occur, for example, on case-insensitive file systems.

Starting from JDK 21 a new compiler option: -Xlint:output-file-clash has been added to the javac compiler. This new option should provide a way for users experiencing this problem to convert what is currently a runtime error into a compile-time warning (or error with -Werror). This new compiler option enables output file clash detection. The term "output file" covers class files, source files, and native header files.

API documentation generated by JavaDoc now supports searching for headings of sections within the documentation.

The JShell tool for interactive exploration of Java code has been enhanced with a new predefined script, TOOLING. The TOOLING script provides direct access to the JDK's command line tools, such as javac, javadoc, and javap, from within JShell.

Similar to the existing predefined DEFAULT and PRINTING scripts, the TOOLING script can be loaded when JShell starts by running: jshell TOOLING. Alternatively, it can be loaded within a JShell session by using: /open TOOLING. With the TOOLING script loaded, JDK tools can be run by passing a name and arguments to the method run(String name, String... args). The method tools() prints the names of available tools.

The TOOLING script defines convenience methods for the most commonly used tools, such as javac(String... args). Here is an example of running the javap tool that disassembles and prints an overview of a class or interface:

jshell> interface Empty {}

jshell> javap(Empty.class)

The -XshowSettings launcher option has been enhanced to print the tzdata version configured with the JDK. The tzdata version is displayed as part of the locale showSettings option.

Example output using -X:showSettings:locale:

.....

Locale settings:
    default locale = English
    default display locale = English
    default format locale = English
    tzdata version = 2023c
    .....

The following changes have been made with regard to the JAXP configuration files:

• Added the jaxp.properties file to the JDK at $JAVA_HOME/conf/jaxp.properties as the default JAXP configuration file. Property settings in the file reflect the current, built-in defaults for the JDK.

• Added a new System Property, java.xml.config.file, for specifying the location of a custom configuration file. If it is set and the named file exists, the property settings contained in the file override those in the default JAXP configuration file. For more details, see the Configuration section of the module specification.

• Deprecated the stax.properties file that was defined in the StAX API and used by the StAX factories. It had been made redundant after StAX's integration into JAXP since the function has been fully covered by the JAXP configuration file. It is recommended that applications migrate to the JAXP configuration file as the stax.properties file is deprecated and may no longer be supported in the future.

java.io.File has historically cached canonical paths, and part paths, to help the performance of the File::getCanonicalFile and File::getCanonicalPath when running with a SecurityManager set. The cache had correctness issues in environments with symbolic links and has been disabled by default since JDK 12. The cache has been removed in this release, along with the system properties sun.io.useCanonCaches and sun.io.useCanonPrefixCache. Setting these properties no longer has any effect.

The method java.lang.ThreadGroup.allowThreadSuspension(boolean) has been removed in this release. The method was used for low memory handling in JDK 1.1 but was never fully specified. It was deprecated and changed to "do nothing" in JDK 1.2 (1998).

The system property java.compiler has been removed from the list of standard system properties.

Running with this system property set on the command line will now print a warning to say that the system property is obsolete; it has no other effect. In previous releases, running with -Djava.compiler or -Djava.compiler=NONE on the command line selected interpreter only execution mode. If needed, the -Xint option can be used to run in interpreter only mode.

The java.lang.Compiler class has been removed. This under-specified API dates from JDK 1.0 and the "Classic VM" used in early JDK releases. Its implementation in the HotSpot VM does nothing but print a warning that it is not supported. The class has been deprecated and marked for removal since Java SE 9.

The "JAR Index" feature has been dropped from the JAR file specification. JAR Index was a legacy optimization in early JDK releases to allow downloading of JAR files to be postponed when loading applets or other classes over the network. The feature has been disabled since JDK 18, meaning the META-INF/INDEX.LIST entry in a JAR file is ignored at run-time.

The system property jdk.net.URLClassPath.enableJarIndex, introduced in JDK 18 to re-enable the feature, has been removed. Setting this property no longer has any effect.

As part of the change, the jar tool will now output a warning if the -i or --generate-index options are used.

The class javax.management.remote.rmi.RMIIIOPServerImpl has been removed. The IIOP transport was removed from the JMX Remote API in JDK 9. This class has been deprecated and its constructor changed to throw UnsupportedOperationException since Java SE 9.

The G1 Hot Card Cache has been removed. Performance testing has shown that after improvements to the concurrent refinement control, it does not contribute to performance.

Removal reduces the memory footprint of the G1 garbage collector by around 0.2% of the Java heap size.

The associated configuration options G1ConcRSLogCacheSize and G1ConcRSHotCardLimit have been obsoleted. A warning will be issued at startup about these options if they are used.

Some older, user-defined class loaders would workaround a deadlock issue by releasing the class loader lock during the loading process. To prevent these loaders from encountering a java.lang.LinkageError: attempted duplicate class definition while loading the same class by parallel threads, the HotSpot Virtual Machine introduced a workaround in JDK 6 that serialized the load attempts, causing the subsequent attempts to wait for the first to complete.

The need for class loaders to work this way was removed in JDK 7 when parallel-capable class loaders were introduced, but the workaround remained in the VM. The workaround was deprecated in JDK 20 and the option -XX:+EnableWaitForParallelLoad was introduced for users who relied on this legacy behavior. The default for this option was off.

In JDK 21, the option -XX:+EnableWaitForParallelLoad, and the code to support it, has been removed.

See CSR JDK-8304056 for more details.

The option MetaspaceReclaimPolicy existed to fine-tune the memory reclamation behavior of metaspace after class unloading. In practice, this had limited effect and was rarely used.

The option has therefore been obsoleted. It now produces an obsolete warning and is ignored.

The following root certificate from SECOM Trust System has been removed from the cacerts keystore:

+ alias name "secomscrootca1 [jdk]"

  Distinguished Name: OU=Security Communication RootCA1, O=SECOM Trust.net, C=JP

The jarsigner options -altsigner and -altsignerpath have been removed, along with the underlying ContentSigner API in the com.sun.jarsigner package. The mechanism was deprecated in JDK 9 and marked for removal in JDK 15.

Implementation support for AWT/Swing using GTK2 on Linux is now deprecated for removal.

With the announcement of the GTK4 release in December 2020, the GTK 2 toolkit is reaching its end of life. GTK2 support is therefore expected to be removed some time after no JDK supported platform needs it.

GTK3 is the current default and Swing applications which opt-in to using GTK2 on Linux by setting the System Property -Djdk.gtk.version=2 will now see the following warning printed:

WARNING: the GTK 2 library is deprecated and its support will be removed in a future release.

com.sun.nio.file.SensitivityWatchEventModifier has been deprecated and is marked for removal in a future release. The constants in this enum were used with the polling based WatchService implementation on macOS to set the interval when polling files for changes. The polling based WatchService has been changed to ignore these modifiers when registering files to be watched.

Users now see a warning message if they specify either COMPAT or JRE locale data with the java.locale.providers system property and call some locale-sensitive operations. COMPAT was provided for migration to the CLDR locale data at the time of JDK 9, where it became the default locale data (JEP 252). JDK 21 retains the legacy locale data of JDK 8 for compatibility, but some of the newer functionalities are not applied. The legacy locale data will be removed in a future release. Users are encouraged to migrate to the CLDR locale data.

The JMX Subject Delegation feature is deprecated and marked for removal in a future release. This feature is enabled by the method javax.management.remote.JMXConnector.getMBeanServerConnection(javax.security.auth.Subject) which is deprecated for removal.

If a client application needs to perform operations as, or on behalf of, multiple identities, it will need to make multiple calls to JMXConnectorFactory.connect() and to the getMBeanServerConnection() method on the returned JMXConnector.

Since JDK 13, executing commands in a sub-process uses the so-called POSIX_SPAWN launching mechanism (that is, -Djdk.lang.Process.launchMechanism=POSIX_SPAWN) by default on Linux. In cases where the parent JVM process terminates abnormally before the handshake between the JVM and the newly created jspawnhelper process has completed, jspawnhelper can hang indefinitely in JDK 13 to JDK 20. This issue is fixed in JDK 21. The issue was especially harmful if the parent process had open sockets, because in that case, the forked jspawnhelper process will inherit them and keep all the corresponding ports open, effectively preventing other processes from binding to them.

This misbehavior has been observed with applications which frequently fork child processes in environments with tight memory constraints. In such cases, the OS can kill the JVM in the middle of the forking process leading to the described issue. Restarting the JVM process after such a crash will be impossible if the new process tries to bind to the same ports as the initial application because they will be blocked by the hanging jspawnhelper child process.

The root cause of this issue is jspawnhelper's omission to close its writing end of the pipe, which is used for the handshake with the parent JVM. It was fixed by closing the writing end of the communication pipe before attempting to read data from the parent process. This way, jspawnhelper will reliably read an EOF event from the communication pipe and terminate once the parent process dies prematurely.

A second variant of this issue could happen because the handshaking code in the JDK didn't handle interrupts to write(2) correctly. This could lead to incomplete messages being sent to the jspawnhelper child process. The result is a deadlock between the parent thread and the child process which manifests itself in a jspawnhelper process being blocked while reading from a pipe and the following stack trace in the corresponding parent Java process:

java.lang.Thread.State: RUNNABLE

  at java.lang.ProcessImpl.forkAndExec(java.base@17.0.7/Native Method)
  at java.lang.ProcessImpl.<init>(java.base@17.0.7/ProcessImpl.java:314)
  at java.lang.ProcessImpl.start(java.base@17.0.7/ProcessImpl.java:244)
  at java.lang.ProcessBuilder.start(java.base@17.0.7/ProcessBuilder.java:1110)
  at java.lang.ProcessBuilder.start(java.base@17.0.7/ProcessBuilder.java:1073)

The computation of the time between java.time.Instants using ChronoUnit.MILLIS.between(t1, t2), ChronoUnit.MICROS.between(t1, t2), t1.until(t2, MILLIS), or t1.until(t2, MICROS) has been corrected. The implementation computing the number of units between Instants, as of JDK 18, did not propagate carry and borrow between seconds and nanoseconds when computing milliseconds and microseconds.

The JDK RPM installer will remove incorrectly constructed entries of "java" and "javac" groups registered by older Oracle JDK RPM installers from the alternatives before registering new "java" and "javac" entries.

An incorrectly constructed entry of the "java" group contains commands that are supposed to belong to the "javac" group.

An incorrectly constructed entry of the "javac" group contains commands that are supposed to belong to the "java" group.

All incorrectly constructed entries belonging to Oracle JDK RPM packages will be removed from the alternatives to avoid corruption of the alternatives internal data.

The removal has a potential side effect for users who have installed multiple JDK versions that are not updated to the latest release. Commands from a removed "java" or "javac" group are now unavailable for system Java switch, which potentially changes the current system Java without a warning. For example, if there is an out-of-date JDK RPM from an 11+ release, say 11.0.17, with an incorrectly constructed single "java" group installed and 8u381 RPM with this patch is installed, it will remove an entry from the "java" group belonging to the 11.0.17 RPM and thus will switch the current system Java from 11.0.17 to 8u381. The side effect will only happen when you install a lower JDK family with the fix, such as 8u381, and there is an out-of-date JDK from a higher family, such as 11.0.17, installed on the system. In that case, 8u381 will replace the older 11.0.17 as the latest. The remedy for the user is to install the latest JDK 11.

The macOS KeychainStore implementation now exposes certificates with proper trust in the user domain, admin domain, or both. Before, only the user domain was considered. Furthermore, if there exists a "deny" entry for a particular purpose in a certificate's trust settings in either domain, the certificate will not be part of the macOS KeychainStore.

The SunJCE implementation for Cipher objects using the ChaCha20 and ChaCha20-Poly1305 algorithms will now allow key/nonce reuse when in DECRYPT_MODE. This change aligns these algorithms with the current SunJCE AES-GCM decrypt mode behavior as it pertains to key/nonce reuse. All ENCRYPT_MODE key/nonce reuse prohibitions remain unchanged from their current behavior.

The Java Language Specification does not allow extra semicolons to appear between import statements, yet the compiler was allowing them; JDK-8027682 fixed this.

As a result, a program like this, which previously would have compiled successfully:

import java.util.Map;;;;

import java.util.Set;
class Test { }

will now generate an error:

Test.java:1: error: extraneous semicolon

import java.util.Map;;;;
                     ^

For backward compatibility, when compiling source versions prior to 21, a warning is generated instead of an error.

A JDK enhancement has improved validation of the ZIP64 Extra Fields contained within zip files and jar files. Files which do not satisfy these new validation checks may result in ZipException : Invalid CEN header (invalid zip64 extra data field size).

The following third party tools have released patches to better adhere to the ZIP File Format Specification:

• Apache Commons Compress fix for Empty CEN Zip64 Extra Headers fixed in Commons Compress release 1.11
• Apache Ant fix for Empty CEN Zip64 Extra Headers fixed in Ant 1.10.14
• BND issue with writing invalid Extra Headers fixed in BND 5.3
• The maven-bundle-plugin 5.1.5 includes the BND 5.3 patch.

If these improved validation checks cause issues for deployed zip or jar files, check how the file was created and whether patches are available from the generating software to resolve the issue. The new validation checks can be disabled by adding -Djdk.util.zip.disableZip64ExtraFieldValidation=true to the runtime launcher arguments.

Further modification of validations on ZIP64 Extra Fields contained within zip and jar files will be made in the upcoming JDK release. See JDK-8313765.

JDK-8132995 introduced an unintended regression when using instances returned by java.util.regex.Matcher.toMatchResult().

This happens on java.util.regex.Patterns containing lookaheads and lookbehinds that, in turn, contain groups. If these are located outside the match, it results in throwing StringIndexOutOfBoundsException when accessing these groups. See JDK-8312976 for an example.

Running the JVM with -XX:+UseZGC and non-default value of -XX:ObjectAlignmentInBytes may lead to JVM crashes or incorrect execution. The issue is caused by an incorrect JIT compiler optimization of the java.lang.Object.clone() method for this configuration. If using ZGC with a non-default value of ObjectAlignmentInBytes is desired, JIT compilation of java.lang.Object.clone() can be disabled using the command-line options -XX:+UnlockDiagnosticVMOptions -XX:DisableIntrinsic=_clone.

The JVM can hang under an uncommon condition that involves the JVM running out of heap memory, the GC just starting a relocation phase to reclaim memory, and a JVM thread-local Handshake asking to relocate an object.

Applications using early releases of JDK 21 may experience long pause times during stop-the-world (STW) garbage collection. This affects various phases of the collection pauses including Class Unloading, Root Scanning, and CodeCache Unloading in all STW garbage collectors (Serial, Parallel, and G1 GC).

These performance problems have been addressed in JDK 21.0.4.

If KeyFactory.generatePrivate or KeyFactory.generatePublic is called on a PKCS8EncodedKeySpec or X509EncodedKeySpec object that has extra bytes at the end of its ASN.1 encoding, an InvalidKeySpecException will be thrown. Before this code change, these extra bytes were ignored.

Some Swing components, such as JLabels and JButtons, which display application text, will try to interpret that text as HTML, principally to enable styled text. The HTML processing of the text for these components will no longer recognize the <object> tag which allows for subclasses of java.awt.Component to be rendered on the component. To re-enable this, applications must specify -Dswing.html.object=true.

The behavior of the method java.io.File.listRoots() on Microsoft Windows has changed in this release so that the returned array includes a File object for all available disk drives. This differs from the behavior in JDK 10 to JDK 20, where this method filtered out disk drives that were not accessible or did not have media present. This change avoids performance issues observed in the previous releases and also ensures that the method is consistent with the root directories in the iteration returned by FileSystem.getDefault().getRootDirectories().

ObjectInputStream::readObject() now throws a StreamCorruptedException instead of a NegativeArraySizeException when reading an array with a negative array size from a corrupted object input stream. Collection classes with a custom readObject() method which previously threw a NegativeArraySizeException when the number of their elements read from the deserialization stream was negative will now throw a StreamCorruptedException instead.

The Thread.sleep(millis, nanos) method is now able to perform sub-millisecond sleeps on POSIX platforms. Before this change, a non-zero nanos argument would round up to a full millisecond. While the precision is improved on most POSIX systems, the actual sleep duration is still subject to the precision and accuracy of system facilities.

Maintainers of applications that do network multicasting or use the java.net.NetworkInterface API should note that the names that the JDK assigns to network interfaces on Windows have changed in this release. The JDK historically synthesized names for network interfaces on Windows. This has changed to use the names assigned by the Windows operating system. For example, the JDK may have historically assigned a name such as “eth0” for an Ethernet interface and “lo” for the loopback. The equivalent names that Windows assigns may be names such as “ethernet_32768” and “loopback_0".

This change may impact code that does a lookup of network interfaces with the NetworkInterace.getByName(String name) method. It also may also be surprising to code that enumerates all network interfaces with the NetworkInterfaces.networkInterfaces() or NetworkInterface.getNetworkInterfaces() methods as the names of the network interfaces will look different to previous releases. Depending on configuration, it is possible that enumerating all network interfaces will enumerate network interfaces that weren’t previously enumerated because they didn’t have an Internet Protocol address assigned. The display name returned by NetworkInterface::getDisplayName has not changed so this should facilitate the identification of network interfaces when using Windows native tools.

FileChannel.transferFrom has changed in this release to support transferring bytes from a source channel to a file at a file position that is beyond the file's current size. FileChannel.transferFrom was previously specified to not transfer any bytes when called with a file position greater than the file's current size.

If the system property file.encoding is set on the command line to the name of a charset that is not in the java.base module, then the JDK will ignore it and default to UTF-8. In JDK 17 and older, the JDK would attempt to locate the charset, even though it was never supported or documented to change the value of this system property in these releases. Since JDK 18, it is possible to set the system property on the command line to the value UTF-8 (the default) or COMPAT. Setting it to any other value is not defined.

The BCP 47 Unicode extension for the strength and normalization collation settings are now supported in the java.text.Collator. If the locale passed to the getInstance(Locale) factory method contains ks and/or kk collation settings, the created Collator instance is set to have the strength and the decomposition modes corresponding to the specified strength and normalization settings.

Swedish collation rules have been modified to reflect the modern sorting for the language. Collation in Swedish now distinguishes 'v' and 'w' as well as sorting alphabetically. For example, {"wb", "va", "vc"} is sorted as {"va", "vc", "wb"} with this change, whereas previously it was sorted as {"va", "wb", "vc"}. In order to specify the old collation, use the co Unicode identifier in the locale. Refer to Support variant collations for more detail.

java.text.Collator now supports multiple collations for a locale. The type of collation may be specified with the Unicode collation identifier if the runtime provides an implementation. For example, the Collator instance created with the locale sv-u-co-trad, traditional collation in the Swedish language, may sort strings, treating v and w the same.

The implementation of java.util.Formatter for double and float conversions to decimal ('e', 'E', 'f', 'g', 'G') is now aligned with the one in Double.toString(double), which was changed in JDK 19.

As a consequence, in some rare circumstances, the outcomes may slightly differ from the ones in earlier releases.

One example is with double 2e23 and format "%.16e". With this change, the outcome is 2.0000000000000000e+23, while earlier releases produce 1.9999999999999998e+23. Any smaller precision in the format (e.g., "%.15e") on this value will produce outcomes that are equal to each other, though.

Another example is with double 9.9e-324 and format "%.2g". The new outcome is 9.9e-324, but earlier releases generate "1.0e-323".

java.util.zip.ZipFile has been updated to provide additional validation of ZIP64 extra fields when opening a ZIP file. This validation may be disabled by setting the system property jdk.util.zip.disableZip64ExtraFieldValidation to true.

Locale data based on the Unicode Consortium's CLDR has been upgraded to version 43. The JDK locale data now employs coverageLevels.txt, including the 'basic' and above level locale data, in addition to the data already existing in prior JDK releases for compatibility. For detailed locale data changes, please refer to the Unicode Consortium's CLDR release notes.

Maintainers of JVM Tool Interface (JVM TI) agents should note that JVM TI now specifies that the ThreadStart and ThreadEnd events are not sent for virtual threads. This is different from when virtual threads were a preview feature in Java SE 19 and Java SE 20. When the feature was in preview, these events were sent for virtual threads even when the can_support_virtual_threads capability was not enabled. Agents that wish to be notified when virtual threads start or terminate need to add the can_support_virtual_threads capability and enable the VirtualThreadStart and VirtualThreadEnd events.

Starting with the July 2023 CPU, on operating systems where ASLR (Address Space Layout Randomization) is enabled, the CDS archive will be placed at a random address picked by the operating system.

This change may have a minor performance impact: (a) Start-up time may increase because the JVM needs to patch pointers inside the CDS archive; (b) Memory usage may increase because the memory used by the CDS archive is no longer shareable across processes. We expect the impact to be small because such increases should be only a small fraction of the overall application usage.

In the unlikely event that you must disable ASLR for CDS, you can use the JVM flags -XX:+UnlockDiagnosticVMOptions -XX:ArchiveRelocationMode=0. The usage of such flags is not recommended.

In prior releases, JNI_GetCreatedJavaVMs:

jint JNI_GetCreatedJavaVMs(JavaVM **vmBuf, jsize bufLen, jsize *nVMs);

could return a JavaVM, via the vmBuf array, that was still in the process of being initialized and may not be ready for use. This has now changed so that it will only return fully initialized VMs. It is important that the programmer checks that the returned number of VMs, in nVMs, is greater than zero, before trying to use any vmBuf entries.

The installation directory of Oracle JDK Debian packages has changed from /usr/lib/jvm/jdk-${FEATURE} to /usr/lib/jvm/jdk-${FEATURE}-oracle-${ARCH}.

The Oracle JDK Debian package registers jexec as an interpreter for launching .jar files from the command line.

The Oracle JDK Debian package configures storage for Java Preferences API in the /etc/.java/.systemPrefs directory.

The Oracle JDK Debian package registers JDK commands with the update-alternatives command and supplies the /usr/lib/jvm/.jdk-${FEATURE}-oracle-${ARCH}.jinfo file for the update-java-alternatives command.

The following root certificate has been added to the cacerts truststore:

+ Certigna (Dhimyotis)

  + certignaca
    DN: CN=Certigna, O=Dhimyotis, C=FR

The following root certificates have been added to the cacerts truststore:

+ Microsoft Corporation

  + microsoftecc2017
    DN: CN=Microsoft ECC Root Certificate Authority 2017, O=Microsoft Corporation, C=US

+ Microsoft Corporation
  + microsoftrsa2017
    DN: CN=Microsoft RSA Root Certificate Authority 2017, O=Microsoft Corporation, C=US

The following root certificate has been added to the cacerts truststore:

+ TWCA

  + twcaglobalrootca
    DN: CN=TWCA Global Root CA, OU=Root CA, O=TAIWAN-CA, C=TW

The following root certificates have been added to the cacerts truststore:

+ Google Trust Services LLC

 + gtsrootcar1
  DN: CN=GTS Root R1, O=Google Trust Services LLC, C=US

+ Google Trust Services LLC
 + gtsrootcar2
  DN: CN=GTS Root R2, O=Google Trust Services LLC, C=US

+ Google Trust Services LLC
 + gtsrootecccar3
  DN: CN=GTS Root R3, O=Google Trust Services LLC, C=US

+ Google Trust Services LLC
 + gtsrootecccar4
  DN: CN=GTS Root R4, O=Google Trust Services LLC, C=US

Added the final keyword to the static java.security.cert.CertStore::getDefaultType(), javax.net.ssl.KeyManagerFactory::getDefaultAlgorithm() and javax.net.ssl.TrustManagerFactory::getDefaultAlgorithm() methods.

This reverts changes made in JDK 19 and JDK 20.

The keytool -genseckey and -importpass commands have been updated to warn users when weak password-based encryption algorithms are specified by the -keyalg option.

A new system property, jdk.jar.maxSignatureFileSize, has been added to allow applications to control the maximum size of signature files in a signed JAR. The value of the system property is the desired size in bytes. The default value is 8000000 bytes.

When initializing the DH KeyPairGenerator implementation of the SunPKCS11 provider with the keysize argument, it looks up the default DH parameters, including the default private exponent length used by other JDK providers, to initialize the underlying native PKCS11 implementation. If the KeyPairGenerator implementation is initialized with the DHParameterSpec object having a negative private exponent length, this invalid negative value will also be overridden with a default value matching the DH prime size.

The JDK implementation of TLS 1.2 now uses a default Diffie Hellman keysize of 2048 bits when a TLS_DHE cipher suite is negotiated and either the client or server does not support FFDHE, which can negotiate a stronger keysize. The JDK TLS implementation supports FFDHE and it is enabled by default.

As a workaround, users can revert to the previous size by setting the jdk.tls.ephemeralDHKeySize system property to 1024 (at their own risk).

This change does not affect TLS 1.3 as the minimum DH group size is already 2048 bits.

Annotation processing by javac is enabled by default, including when no annotation processing configuration options are present. Implicit annotation processing by default may be disabled in a future release, possibly as early as JDK 22. To alert javac users of this possibility, in JDK 21 javac prints a note if implicit annotation processing is being used. The text of the note is:

Annotation processing is enabled because one or more processors were found on the class path. A future release of javac may disable annotation processing unless at least one processor is specified by name (-processor), or a search path is specified (--processor-path, --processor-module-path), or annotation processing is enabled explicitly (-proc:only, -proc:full).
Use -Xlint:-options to suppress this message.
Use -proc:none to disable annotation processing.

Good build hygiene includes explicitly configuring annotation processing. To ease the transition to a different default policy in the future, the new-in-JDK-21 -proc:full javac option requests the current default behavior of looking for annotation processors on the class path.

A new compiler lint flag, output-file-clash, enables detection of output file clashes. An output file clash is when the compiler intends to write two different output files, but due to the behavior of the operating system, these files end up being written to the same underlying file.

This usually happens due to case-insensitive file systems. For example, a class like this would cause two class files to be written to the same file, Test$Inner.class:

public class Test {

    class Inner {
    }
    class INNER {
    }
}

However, this problem can also happen when the file system "normalizes" file names. For example, on macOS, compiling this class will generate such a clash:

public class Test {

    interface Cafe\u0301 {
    }
    interface Caf\u00e9 {
    }
}

The reason is that \u0301 is the Unicode character "Combining Acute Accent" which means "add an accent over the previous character". MacOS normalizes the letter e followed by a \u0301 into a Unicode \u00e9, that is, é. However, the Java language treats these the two names, Cafe\u0301 and Caf\u00e9, as distinct.

Compiling the example above on macOS with -Xlint:output-file-clash will now generate a warning like this:

    warning: [output-file-clash] output file written more than once: /home/test/Test$Café.class

Prior to JDK 21, the javac compiler was omitting some "potentially ambiguous overload" warnings enabled by the -Xlint:overloads option.

If the -Xlint:overloads option is enabled, the compiler warns when the methods in a class create a potential ambiguity for method invocations containing an implicit lambda expression parameter like x -> { ... }. An ambiguity can occur if two or more methods could match such a method call, like when one method takes a Consumer<Integer> parameter where the other takes an IntConsumer. For example, the javac compiler should issue a warning for code such as:

    interface I {

        void foo(Consumer<Integer> c);
        void foo(IntConsumer c);
    }

Prior to JDK 21, the warning was only issued for a class if one of the methods was declared in the class. The javac compiler now also warns when neither method is declared in the class. That is, both methods are inherited from supertypes. For example, for code like:

    interface I {

        void foo(Consumer<Integer> c);
    }

    interface J {
        void foo(IntConsumer c);
    }

    interface K extends I, J {}

Prior to JDK 21, the javac compiler did not always mark method parameters as synthetic or mandated when applicable. Starting with JDK 21, the javac compiler emits the MethodParameters attribute in the class file when applicable. This attribute stores information on whether or not parameters are synthetic or mandated. This change applies to all release and target versions supported by javac since all currently supported releases and targets have the MethodParameters attribute defined.

This change is justified by JLS § 13.1, in particular:

A binary representation for a class or interface must also contain all of the following:


[...]

11. A construct emitted by a Java compiler must be marked as synthetic if it does not
    correspond to a construct declared explicitly or implicitly in source code, unless
    the emitted construct is a class initialization method (JVMS §2.9).
12. A construct emitted by a Java compiler must be marked as mandated if it corresponds
    to a formal parameter declared implicitly in source code (§8.8.1, §8.8.9, §8.9.3,
    §15.9.5.1).

Prior to JDK 21, the javac compiler was not allowing annotations with no @Target annotation to be applied to type parameter declarations.

This is a bug in the javac compiler in versions prior to JDK21. JLS 9.6.4.1 specifies that annotations without an @Target annotation are applicable in 'all declaration contexts', which includes type parameter declarations.

Starting from JDK21, the javac compiler will accept code like:

import java.lang.annotation.Retention;

import java.lang.annotation.RetentionPolicy;

@Retention(RetentionPolicy.RUNTIME)
@interface Anno {}

class C<@Anno T> {}

This change affects compilations targeting -source/--release 14 and higher.

This JDK release relaxes the specification of java.awt.Robot to account for possible platform and desktop environment access restrictions or limitations.

NMT reports will now show peak values for all categories. Peak values contain the highest value for committed memory in a given NMT category over the lifetime of the JVM process.

If the committed memory for an NMT category is currently at peak, NMT prints "at peak"; otherwise, it prints the peak value.

For example:

- Compiler (reserved=230KB, committed=230KB)
                            (malloc=34KB #64) (peak=49KB #71)
                            (arena=196KB #4) (peak=6126KB #16)
							

This shows Compiler arena memory peaked at a bit more than 6MB, whereas it now hovers around 200 KB.

Although we have stated the goal to have Oracle JDK and OpenJDK binaries be as close to each other as possible, there remain several differences between the two options.

The current differences are:

• Oracle JDK offers "installers" (msi, rpm, deb, etc.) which not only place the JDK binaries in your system but also contain update rules and in some cases handle some common configurations like set common environmental variables (such as, JAVA_HOME in Windows) and establish file associations (such as, use java to launch .jar files). OpenJDK is offered only as compressed archive (tar.gz or .zip).
• Usage Logging is only available in Oracle JDK.
• Oracle JDK requires that third-party cryptographic providers be signed with a Java Cryptography Extension (JCE) Code Signing Certificate. OpenJDK continues allowing the use of unsigned third-party crypto providers.
• The output of java -version is different. Oracle JDK returns java and includes the Oracle-specific identifier. OpenJDK returns OpenJDK and does not include the Oracle-specific identifier.
• Oracle JDK 17 and later are released under the Oracle No-Fee Terms and Conditions License. OpenJDK is released under GPLv2wCP. License files included with each will therefore be different.
• Oracle JDK distributes FreeType under the FreeType license and OpenJDK does so under GPLv2. The contents of \legal\java.desktop\freetype.md is therefore different.
• Oracle JDK has Java cup and steam icons and OpenJDK has Duke icons.
• Oracle JDK source code includes "ORACLE PROPRIETARY/CONFIDENTIAL. Use is subject to license terms." Source code distributed with OpenJDK refers to the GPL license terms instead.