JDK 15 Release Notes

This release upgrades Unicode support to 13.0, which includes the following:

• The java.lang.Character class supports Unicode Character Database of 13.0 level, which 13.0 adds 5,930 characters, for a total of 143,859 characters. These additions include 4 new scripts, for a total of 154 scripts, as well as 55 new emoji characters.
• The java.text.Bidi and java.text.Normalizer classes support 13.0 level of Unicode Standard Annexes, #9 and #15, respectively.
• The java.util.regex package supports Extended Grapheme Clusters based on 13.0 level of Unicode Standard Annex #29 For more detail about Unicode 13.0, refer to the Unicode Consortium's release note.

java.lang.CharSequence has been updated in this release to define a default isEmpty method that tests if a character sequence is empty. Testing for, and filtering out, empty Strings and other CharSequences is a common occurrence in code and CharSequence::isEmpty can be used as a method reference. Classes that implement java.lang.CharSequence and another interface that defines isEmpty method should be aware of this addition as they may need to be modified to override the isEmpty method.

JEP 371 introduces hidden classes in Java 15. Hidden classes have the following implications to existing code:

1. Class::getName traditionally returns a binary name, but for a hidden class it returns a string that contains an ASCII forward slash (/) and is therefore not a binary name. Programs that assume the returned string is a binary name might need to be updated to handle hidden classes. That said, the longstanding practice of Unsafe::defineAnonymousClass was to define classes whose names were not binary names, so some programs may already handle such names successfully.

2. Class::descriptorString and MethodType::descriptorString returns a string that contains a ASCII dot (.) for a hidden class and therefore is not a type descriptor conforming to JVMS 4.3. Programs that assume the returned string is a type descriptor that conforms to JVMS 4.3 might need to be updated to handle hidden classes.

3. Class::getNestMembers is changed to not throw an exception when it fails to validate a nest membership of any member listed in NestMembers attribute. Instead, Class::getNestMembers returns the nest host and the members listed in the host's NestMembers attribute that are successfully resolved and determined to have the same nest host as this class. (This means it may return fewer members that listed in NestMembers attribute.) Existing code that expects LinkageError when there is a bad nest membership might be impacted.

4. The nestmate test in the JVM is changed to throw only IllegalAccessError when the nest membership is invalid. Some historical understanding is necessary:

• In Java 8, every access control failure was signaled with IllegalAccessError (IAE). Moreover, if a given access check failed with IAE once, then the same check would fail with IAE every time.
• In Java 11, the introduction of nest mates (JEP 181) meant that an access control failure could be signaled either with IllegalAccessError or, if nest membership was invalid, LinkageError. Still, if a given access check failed with a specific exception, then the same check would always fail with the same exception.
• In Java 15, the introduction of Lookup::defineHiddenClass implies that the nest host of the lookup class must be determined eagerly, when the hidden class is defined as a nestmate of the lookup class. Both Lookup::defineHiddenClass and Class::getNestHost both determine the nest host of a class in a more resilient manner than the JVM did in Java 11; namely, the API simply treats a class as self-hosted if its purported nest membership is invalid. For consistency with the API, the JVM no longer throws LinkageError when a class's nest membership is invalid, and instead treats the class as self-hosted. This means that the JVM only throws IAE from access control (because a self-hosted class will not permit any other class to access its private members). This is the behavior expected by the vast majority of user code.

5. JVM TI GetClassSignature returns a string that contains a ASCII dot (.) for a hidden class. JVM TI agents might need updating for hidden classes if they assume the returned string from GetClassSignature is a type descriptor conforming to JVMS 4.3.

A new JDK-specific socket option SO_INCOMING_NAPI_ID has been added to jdk.net.ExtendedSocketOptions in this release. The socket option is Linux specific and allows applications to query the NAPI (New API) ID of the underlying device queue associated with its socket connection and take advantage of the Application Device Queue (ADQ) feature of high performance Network Interface Card (NIC) devices.

The TreeMap class now provides overriding implementations of the putIfAbsent, computeIfAbsent, computeIfPresent, compute, and merge methods. The new implementations provide a performance improvement. However, if the function provided to the compute- or merge methods modifies the map, ConcurrentModificationException may be thrown, because the function that is provided to these methods is prohibited from modifying the map. If a ConcurrentModificationException occurs, the function must either be changed to avoid modifying the map, or the surrounding code should be rewritten to replace uses of the compute- and merge methods with conventional Map methods such as get and put.

See JDK-8227666 for further information.

JMX supports (explicit) remote network access through the configuration of two network ports (either from the command line or in a property file), by setting the following properties:

com.sun.management.jmxremote.port=<port#>

com.sun.management.jmxremote.rmi.port=<port#>

Note: If it is not specified, the second port will default to the first.

A third local port is also opened to accept (local) JMX connections. This port previously had its number selected at random, which could cause port collisions.

However, it is now possible to configure the third JMX port (local only) by using:

com.sun.management.jmxremote.local.port=<port#>

A new -r <port> option has been added to the jstatd command to specify the RMI connector port number. If a port number is not specified, a random available port is used.

A new integer option gz has been added to the GC.heap_dump diagnostic command. If it is specified, it will enable the gzip compression of the written heap dump. The supplied value is the compression level. It can range from 1 (fastest) to 9 (slowest, but best compression). The recommended level is 1.

Three new options have been added to the jhsdb command for the debugd mode:

1. --rmiport <port> is used to specify a RMI connector port number. If a port number is not specified, a random available port is used.
2. --registryport <port> is used to specify a RMI registry port number. This option overrides the system property sun.jvm.hotspot.rmi.port. If a port number is not specified, the system property is used. If the system property is not set, the default port 1099 is used.
3. --hostname <hostname> is used to specify a RMI connector host name. The value could be a hostname or an IPv4/IPv6 address. This option overrides the system property java.rmi.server.hostname. If a host name not specified, the system property is used. If the system property is not set, a system host name is used.

A new -revCheck option has been added to the jarsigner command to enable revocation checking of certificates.

The keytool and jarsigner tools have been updated to warn users about weak cryptographic algorithms being used before they are disabled. In this release, the tools issue warnings for the SHA-1 hash algorithm and 1024-bit RSA/DSA keys.

The SunJCE provider has been enhanced to support HmacSHA3-224, HmacSHA3-256, HmacSHA3-384, and HmacSHA3-512. Implementations for these algorithms are available under the Mac and KeyGenerator services. The Mac service generates the keyed-hash and the KeyGenerator service generates the key for the Mac.

Two new system properties have been added to customize the TLS signature schemes in JDK. jdk.tls.client.SignatureSchemes has been added for the TLS client side, and jdk.tls.server.SignatureSchemes has been added for the server side.

Each system property contains a comma-separated list of supported signature scheme names specifying the signature schemes that could be used for the TLS connections.

The names are described in the "Signature Schemes" section of the Java Security Standard Algorithm Names Specification.

The "certificate_authorities" extension is an optional extension introduced in TLS 1.3. It is used to indicate the certificate authorities (CAs) that an endpoint supports and should be used by the receiving endpoint to guide certificate selection.

With this JDK release, the "certificate_authorities" extension is supported for TLS 1.3 in both the client and the server sides. This extension is always present for client certificate selection, while it is optional for server certificate selection.

Applications can enable this extension for server certificate selection by setting the jdk.tls.client.enableCAExtension system property to true. The default value of the property is false.

Note that if the client trusts more CAs than the size limit of the extension (less than 2^16 bytes), the extension is not enabled. Also, some server implementations do not allow handshake messages to exceed 2^14 bytes. Consequently, there may be interoperability issues when jdk.tls.client.enableCAExtension is set to true and the client trusts more CAs than the server implementation limit.

The 'canonicalize' flag in the krb5.conf file is now supported by the JDK Kerberos implementation. When set to true, RFC 6806 name canonicalization is requested by clients in TGT requests to KDC services (AS protocol). Otherwise, and by default, it is not requested.

The new default behavior is different from JDK 14 and previous releases where name canonicalization was always requested by clients in TGT requests to KDC services (provided that support for RFC 6806 was not explicitly disabled with the sun.security.krb5.disableReferrals system or security properties).

Text blocks have been added to the Java language. A text block is a multi-line string literal that avoids the need for most escape sequences, automatically formats the string in a predictable way, and gives the developer control over the format when desired.

In this release, in conjunction with the removal of the Solaris port in JEP 381, the JDK-specific socket option jdk.net.ExtendedSocketOptions.SO_FLOW_SLA, which is only relevant to sockets on Solaris, and its supporting classes SocketFlow and SocketFlow.Status, have been removed.

The RMI static stub compiler rmic has been removed. The rmic tool is obsolete and has been deprecated for removal since JDK 13.

The Nashorn JavaScript script engine, its APIs, and the jjs tool have been removed. The engine, the APIs, and the tool were deprecated for removal in Java 11 with the express intent to remove them in a future release.

The terminally deprecated constant javax.management.remote.rmi.RMIConnectorServer.CREDENTIAL_TYPE has been removed. A filter pattern can be specified instead by using RMIConnectorServer.CREDENTIALS_FILTER_PATTERN.

The VM option UseAdaptiveGCBoundary is obsolete. Use of this option will produce an obsolete option warning but will otherwise be ignored.

This option was previously disabled by default, and enabling it only had an effect when also using -XX:+UseParallelGC. Enabling it was intended to provide a performance benefit for some applications. However, it has been disabled by default for a long time because of crashes and performance regressions.

The following expired DocuSign root CA certificate has been removed from the cacerts keystore:

+ alias name "keynectisrootca [jdk]"

  Distinguished Name: CN=KEYNECTIS ROOT CA, OU=ROOT, O=KEYNECTIS, C=FR

The following expired Comodo root CA certificate has been removed from the cacerts keystore:

+ alias name "addtrustclass1ca [jdk]"

  Distinguished Name: CN=AddTrust Class 1 CA Root, OU=AddTrust TTP Network, O=AddTrust AB, C=SE

The legacy SunJSSE provider name, "com.sun.net.ssl.internal.ssl.Provider" has been removed and should no longer be used. The "SunJSSE" name should be used instead. For example, SSLContext.getInstance("TLS", "SunJSSE").

The implementation of the deprecated SSLSession.getPeerCertificateChain() method has been removed from the JDK in the SunJSSE provider and the HTTP client implementation. The default implementation of this method has been changed to throw UnsupportedOperationException.

SSLSession.getPeerCertificateChain() is a deprecated method and will be removed in a future release. To mitigate the removal compatibility impact, applications should use the SSLSession.getPeerCertificates() method instead. For service providers, please remove this method from the existing implementation, and do not support this method in any new implementation.

After an upgrade of the macOS SDK used to build the JDK, the behavior of the apple.awt.brushMetalLook and textured Swing properties has changed. When these properties are set, the title of the frame is still visible. It is recommended that the apple.awt.transparentTitleBar property be set to true to make the title of the frame invisible again. The apple.awt.fullWindowContent property can also be used.

Please note that Textured window support was implemented by using the NSTexturedBackgroundWindowMask value of NSWindowStyleMask. However, this was deprecated in macOS 10.12 along with NSWindowStyleMaskTexturedBackground, which was deprecated in macOS 10.14.

For additional information, refer to the following documentation:

• apple.awt.brushMetalLook: https://developer.apple.com/documentation/appkit/nstexturedbackgroundwindowmask?language=objc
• apple.awt.transparentTitleBar: https://developer.apple.com/documentation/appkit/nswindow/1419167-titlebarappearstransparent?language=objc
• apple.awt.fullWindowContent: https://developer.apple.com/documentation/appkit/nsfullsizecontentviewwindowmask

The RMI Activation mechanism has been deprecated and may be removed in a future version of the platform. RMI Activation is an obsolete part of RMI that has been optional since Java 8. It allows RMI server JVMs to be started ("activated") upon receipt of a request from a client, instead of requiring RMI server JVMs to be running continuously. Other parts of RMI are not deprecated. See JEP 385 for further information.

The VM option ForceNUMA is deprecated. Use of this option will produce a deprecation warning. This option will be removed in a future release.

This option has always been disabled by default. It exists to support testing of NUMA-related code paths when running on a single node / UMA platform.

Biased locking has been disabled by default in this release. In addition, the VM option UseBiasedLocking along with the VM options BiasedLockingStartupDelay, BiasedLockingBulkRebiasThreshold, BiasedLockingBulkRevokeThreshold, BiasedLockingDecayTime and UseOptoBiasInlining have been deprecated. The options will continue to work as intended but will generate a deprecation warning when they are used.

Biased locking might affect performance on applications that exhibit significant amounts of uncontended synchronization, such as applications that rely on older Java Collections APIs that synchronize on every access. Hashtable and Vector are examples of these APIs. Use -XX:+BiasedLocking on the command line to re-enable biased locking. Report any significant performance regressions to Oracle with biased locking disabled.

The SunEC crypto provider no longer advertises curves that are not implemented by using modern formulas and techniques. Arbitrary and named curves, listed at the bottom of this note, are disabled. Commonly used named curves, secp256r1, secp384r1, secp521r1, x25519, and x448, remain supported and enabled by SunEC because they use modern techniques. Applications that still require the disabled curves from the SunEC provider can re-enable them by setting the System property jdk.sunec.disableNative to false. For example: java -Djdk.sunec.disableNative=false ....

If this property is set to any other value, the curves will remain disabled. Exceptions thrown when the curves are disabled will contain the message Legacy SunEC curve disabled, followed by the name of the curve. Methods affected by the change are KeyPair.generateKeyPair(), KeyAgreement.generateSecret(), Signature.verify(), and Signature.sign(). These methods throw the same exception class they had before when the curve was not supported.

The following curves are disabled: secp112r1, secp112r2, secp128r1, secp128r2, secp160k1, secp160r1, secp160r2, secp192k1, secp192r1, secp224k1, secp224r1, secp256k1, sect113r1, sect113r2, sect131r1, sect131r2, sect163k1, sect163r1, sect163r2, sect193r1, sect193r2, sect233k1, sect233r1, sect239k1, sect283k1, sect283r1, sect409k1, sect409r1, sect571k1, sect571r1, X9.62 c2tnb191v1, X9.62 c2tnb191v2, X9.62 c2tnb191v3, X9.62 c2tnb239v1, X9.62 c2tnb239v2, X9.62 c2tnb239v3, X9.62 c2tnb359v1, X9.62 c2tnb431r1, X9.62 prime192v2, X9.62 prime192v3, X9.62 prime239v1, X9.62 prime239v2, X9.62 prime239v3, brainpoolP256r1 brainpoolP320r1, brainpoolP384r1, brainpoolP512r1

The ContentSigner and ContentSignerParameters classes in the com.sun.jarsigner package support alternative signers and have been deprecated with forRemoval=true. When the -altsigner or -altsignerpath options are specified, the jarsigner tool produces a warning that these options are deprecated and will be removed.

During the setup of new connections, java.net.http.HttpClient now uses the default set of protocols provided by the SSLContext when negotiating the TLS handshake. In the absence of any SSLParameters explicitly supplied to the HttpClient.builder, the HttpClient has been updated to no longer override any default-selected protocols in the SSLContext. As a result, the actual TLS version that is negotiated might differ from that of previous releases, or it might even succeed or fail to negotiate when it previously might not have.

jpackage cannot create packages on macOS that are suitable for notarization.

It has been found that some Windows GDI functions don't support all types of Java heap memory allocation schemes. This problem can cause repaint issues and printing bugs. It has been worked around by allocating temporary buffers off heap.

See: https://support.microsoft.com/en-us/help/4567569/gdi-apis-may-fail-when-large-pages-or-vad-spanning-is-used

When it is interrupted, the implementation of the java.awt.Robot.delay() method has been changed to complete with the interrupt status set.

If a thread is interrupted while waiting in the java.awt.Robot.delay() method, then this method returns immediately with the interrupt status set. If the interrupted status is already set, this method returns immediately with the interrupt status set.

The implementation of String.substring and related methods stripLeading and stripTrailing have changed in this release to avoid redundantly creating a new empty String. This may impact code that depends on unspecified behaviour and the identity of empty sub-strings.

Lookup::defineClass is specified to throw LinkageError if a linkage error occurs, but the implementation was not actually linking the class. In this release, the implementation has been changed to link the class before returning, so conforming to the specification. If Lookup::defineClass is called to define a class that fails linking, LinkageError will be thrown.

In this release, the default port number for a datagram packet has been changed to 0. Previously, this value was -1, which was undocumented. The port can be retrieved by using DatagramPacket::getPort.

Previously, DatagramChannel::disconnect threw an IOException while DatagramSocket::disconnect did not. As a result, the DatagramChannel::socket adapter, which calls DatagramChannel::disconnect, catches the thrown IOException and rethrows it as an Error. However, this was undocumented behavior and not user-friendly.

The DatagramChannel::socket adapter has been changed to throw an UncheckedIOException, and the specification of DatagramSocket::disconnect has been updated to document that an implementation may now throw an UncheckedIOException. This ensures consistency in behavior between DatagramSocket, DatagramChannel, and DatagramChannel::socket adapter.

In this release, the behavior of InetAddress.getAllByName has been modified when the alternative hosts file name service is selected .

The JDK allows specifying an alternative host's file name service by using the jdk.net.hosts.file system property. The implementation of the alternative name service has been changed to take into account the values of the java.net.preferIPv4Stack and java.net.preferIPv6Addresses system properties. This affects the results returned by InetAddress.getAllByName when the host's file name service is selected. For details about java.net.preferIPv4Stack and java.net.preferIPv6Addresses, see Networking Properties in the API documentation.

In this release, some of the one-way byte-to-char mappings have been aligned with the preferred mappings provided by the Unicode Consortium.

DecimalFormat/DecimalFormatSymbols classes are now capable of dealing with grouping separators for currency values. For example, the monetary grouping separator for the German language used in Austria (the de-AT locale) is '.', whereas the monetary grouping separator in other German locales is ' '.

java.time.temporal.ValueRange.of() methods are now correctly throwing an InvalidArgumentException on given invalid arguments. For example, of(5, 2, 10) which is invalid because the minimum is greater than the smallest maximum, now throws the exception.

java.time.format.DateTimeFormatter.localizedBy(Locale) method now honors the default locale values, such as Chronologyand/or DecimalStyle of the specified locale argument.

For example, in previous JDK releases:

jshell> DateTimeFormatter.ofLocalizedDate(FormatStyle.FULL)
    .localizedBy(Locale.forLanguageTag("fa"))
    .format(LocalDate.now())
$3 ==> "جمعه 1 مهٔ 2020"

the numbers are in Arabic (Western) numerals.

In JDK 15:

jshell> DateTimeFormatter.ofLocalizedDate(FormatStyle.FULL)
    .localizedBy(Locale.forLanguageTag("fa"))
    .format(LocalDate.now())
$3 ==> "جمعه ۱ مهٔ ۲۰۲۰"

the numbers are in Extended Arabic-Indic numerals because it is the default numbering system for the Farsi locale.

InflaterOutputStream(OutputStream out, Inflater infl, int bufLen) allows for specifying the decompressor and buffer size to be used.

InflaterOutputStream.write(byte[] b, int off, int len) was writing data using a max buffer size of 512 bytes.

Starting with JDK 15 the buffer size specified via InflaterOutputStream(OutputStream out, Inflater infl, int bufLen) will be used in calls to InflaterOutputStream.write(byte[] b, int off, int len). If the buffer size is not specified when invoking the InflaterOutputStreamconstructor, it will default to 512 bytes.

The Java regular expression engine supports the case insensitive mode. When this mode is turned on, the engine is supposed to match the input text without regard to the case of the characters it consists of.

However, the current implementation of matching against some named character classes (those that are encoded with p{name} or P{name} constructs) fails to respect the case insensitive mode.

This fix makes these character classes behave consistently with respect to case sensitivity. When the regular expression engine operates in the case insensitive mode, the named character classes will match the input characters without regard to their case: lower case, upper case, or title case.

English time zone names provided by the CLDR locale provider are now correctly synthesized following the CLDR spec, rather than substituted from the COMPAT provider. For example, SHORT style names are no longer synthesized abbreviations of LONG style names, but instead produce GMT offset formats.

Locale data based on Unicode Consortium's CLDR has been upgraded to their version 37. For the detailed locale data changes, please refer to the Unicode Consortium's CLDR release notes:

• http://cldr.unicode.org/index/downloads/cldr-37

A number of flags controlling inlining in the C1 and C2 compilers have been split up into separate flags. The C2 compiler keeps the flags with the old names, and the C1 compiler gets the new flags.

Old flags now only controlling C2

• MaxInlineLevel
• MaxRecursiveInlineLevel
• MaxInlineSize
• MaxTrivialSize
• InlineSmallCode
• FreqInlineSize

New flags for C1 that replace the old ones

• C1MaxInlineLevel
• C1MaxRecursiveInlineLevel
• C1MaxInlineSize
• C1MaxTrivialSize

Deprecation

If the old flags are used in a JDK build without the C2 compiler, a deprecation warning will be printed.

The Z Garbage Collector (ZGC) is now ready for use in production and no longer marked as an experimental feature. ZGC is enabled by using the -XX:+UseZGC command-line option (using -XX:+UnlockExperimentalVMOptions is no longer needed).

See JEP 377 for more details.

The default heap region size calculation has been changed to return larger regions by default. The calculation still aims to have 2048 regions, but two aspects have changed:

• Only the maximum heap size is considered. The old calculation also took the initial heap size into consideration, but this can give unexpected behavior when no heap size is set.
• The region size is rounded up to the nearest power of 2 instead of down. This will return larger region sizes in cases where the maximum heap size is not a power of 2.

These changes improve startup and runtime performance.

-XX:+UseNUMAInterleaving has no effect on Windows in this release. It was found that GDI APIs used by java2d don't support the memory reservation scheme used for NUMA interleaving. The JVM detects this problem and both warns about this and turns off NUMA interleaving. See: https://support.microsoft.com/en-us/help/4567569/gdi-apis-may-fail-when-large-pages-or-vad-spanning-is-used

-XX:+UseLargePages has no effect on Windows in this release. It was found that GDI APIs used by java2d don't support large pages. The JVM detects this problem and both warns about this and reverts to using small pages. See: https://support.microsoft.com/en-us/help/4567569/gdi-apis-may-fail-when-large-pages-or-vad-spanning-is-used

The way that field layout is computed has been changed, with more aggressive optimizations to avoid unused gaps in instances. These new optimizations can be disabled by using a new VM option -XX:-UseEmptySlotsInSupers.

For a limited time, it is possible to continue to use the old code to compute field layout with a new VM option -XX:-UseNewFieldLayout. However, this option has been deprecated in JDK 15 and the old code will be removed in a future release.

The default of the flag ShowCodeDetailsInExceptionMessages was changed to 'true'. The helpful NullPointerException messages of JEP 358 are now printed by default. The messages contain snippets of the code where the NullPointerException was raised.

App deployers should double check the output of their web applications and similar usage scenarios. The NullPointerException message could be included in application error messages or be displayed by other means in the app. This could give remote attackers valuable hints about a potential vulnerable state of the software components being used.

An example message is 'Cannot read field "c" because "a.b" is null'. The attacker knows that field b of a contains null which might be unintended and offer an opportunity for an attack. For more details of what the message can contain see the above mentioned JEP 358.

Signature.getParameters() and SignatureSpi.engineGetParameters() may return null if the underlying provider does not support returning the parameters as AlgorithmParameters. For further details, see the Signature and SignatureSpi method descriptions.

The SunPKCS11 security provider can now be initialized with NSS when FIPS-enabled external modules are configured in the Security Modules Database (NSSDB). Before this change, when such a library was configured for NSS in non-FIPS mode, the SunPKCS11 provider would throw a RuntimeException with the message "FIPS flag set for non-internal module".

This change allows the JDK to work properly with recent NSS releases in GNU/Linux operating systems when the system-wide FIPS policy is turned on.

In JDK 11 and later, javax.net.ssl.SSLEngine by default used client mode when handshaking. As a result, the set of default enabled protocols may differ to what is expected. SSLEngine would usually be used in server mode. From this JDK release onwards, SSLEngine will default to server mode. The javax.net.ssl.SSLEngine.setUseClientMode​(boolean mode) method may be used to configure the mode.