This is to announce GNU Libidn for Java 2026.10.0.

GNU Libidn for Java is a pure-Java implementation of the Stringprep,
Punycode and IDNA2003 specifications.  Its purpose is to encode and
decode internationalized domain name strings.  Until now it was the
Java port inside GNU Libidn.  It is now a project of its own, with the
git history of its sources preserved.  It is an independent
implementation, not a binding to the C library.

Be aware that IDNA2003 and Stringprep are superseded, except Punycode.
IDNA2008 (RFC 5891) replaces IDNA2003, and PRECIS (RFC 8264) replaces
Stringprep.  Use this library only for the historic IDNA2003 behaviour
or for the Stringprep profiles.

This is the first release of the new project.  Releases are now
numbered by year and month, so this release follows 1.44.  It fixes
several bugs in the Stringprep table checks, Punycode, IDNA and NFKC.
Some of these bugs date back to release 1.27.  All users are
encouraged to upgrade.

There have been 41 commits by 1 person since the split from GNU
Libidn.

See the NEWS below for a brief summary.

Thanks to everyone who has contributed, and to Simon for the support
with the split!  The following people contributed changes to this
release:

  Guus der Kinderen (41)

Kind regards,

  Guus

==================================================================

Here is the project home page:
  https://codeberg.org/libidn/libidn-java

The library is available from Maven Central:

  <dependency>
    <groupId>org.gnu.inet</groupId>
    <artifactId>libidn</artifactId>
    <version>2026.10.0</version>
  </dependency>

Here are the binary, sources and javadoc jars:
  https://repo1.maven.org/maven2/org/gnu/inet/libidn/2026.10.0/

The jars are reproducible.  Build them from the tagged sources with
'./mvnw verify' and compare them with the jars on Maven Central.

This release is based on the libidn-java git repository, available as

  git clone https://codeberg.org/libidn/libidn-java.git

with commit a013b3538486787b3c985fa619dfca0ec0acde67 tagged as
v2026.10.0.

NEWS

* Noteworthy changes in release 2026.10.0 (2026-10-12)

** Stringprep table checks no longer miss characters.
Prohibited characters, unassigned code points and bidi violations could
go undetected, or the check threw ArrayIndexOutOfBoundsException.  This
affected releases 1.27 to 1.44.

** The error message for bidi requirement 2 of RFC 3454 names the rule.
It is in the new constant StringprepException.BIDI_BOTH_L_AND_RAL.  The
old constant BIDI_BOTHRAL is deprecated, and has the same value.

** The table checks are simpler and at least as fast.

** Punycode decoding accepts uppercase digits and rejects non-digits.
IDNA.toUnicode no longer throws an unchecked exception for such input.

** IDNA.toUnicode recognizes the ACE prefix in any capitalization.

** IDNA.toASCII accepts a domain name that ends with a dot.
Empty labels are reported with the new constant
IDNAException.EMPTY_LABEL instead of TOO_LONG.

** NFKC no longer composes a Hangul syllable with U+11A7 or U+11C3.

** Punycode.encode reports an overflow for very long input.

** RangeSet.Builder.addRanges(Collection) adds the given ranges.

** IDNAException has the message of the exception it wraps.
Before, the message started with the class name of that exception.

** The micro-benchmarks use JMH instead of Caliper.

** NFKC no longer composes a character that a combining mark blocks.
This happened after a starter had composed across combining marks.

** Stringprep is fast on long input.
The time used to grow with the square of the length of the input.  A
string with a long run of combining marks could take minutes.

** The message constants of the exceptions are final.

** The generated tables are no longer public.
The classes CombiningClass, Composition, DecompositionKeys,
DecompositionMappings and RFC3454 are now package-private.  Code
outside the library could change the contents of the NFKC tables, and
so the results of NFKC and Stringprep.

** The license information is complete.
The pom lists all three licenses of the library: the LGPL version 3 or
later, the GPL version 2 or later, and the Apache License 2.0.  The
binary and sources jars contain the license files, and the copyright
notices of RFC 3454 and the Unicode data that the tables are made from.

** The Java port of GNU Libidn is now a project of its own.
It was split off from GNU Libidn, with the history of its sources
preserved.  Maven is now the only build system, and the library targets
Java 8.  It is still published on Maven Central as org.gnu.inet:libidn,
now together with its sources and javadoc jars.  The jars are
reproducible.

** Releases are numbered by year and month.  See the README.

** The Stringprep and NFKC tables are generated at build time.
They are generated from the RFC 3454 and Unicode 3.2.0 data files, which
are now part of the source tree.  The generated tables are the same as
in earlier releases.

** The Javadoc of IDNA.toASCII(String, boolean, boolean) is corrected.
It said that unassigned code points are never allowed and that the STD3
ASCII rules are always enforced.  Both depend on the parameters.  The
method converts a single label, without splitting it at dots.

** The public API has Javadoc comments where they were missing.

** More tests.
The tests now include IDNA and NFKC test vectors, and tests for the STD3
ASCII rules, the AllowUnassigned flag and the bidi rules of RFC 3454.

Reply via email to