If you want to install from source you'll need to download the source code if you haven't already done so.
Running "make" and then "make install" will "install" a JNI glue shared library into a "built" subdirectory of the java build directory. The jar file is built into the "built" subdirectory too.
You can copy these two files into your java installation, or just use them in-place.
cd java javac -classpath built/xapian.jar:. docs/examples/SimpleIndex.java javac -classpath built/xapian.jar:. docs/examples/SimpleSearch.java
To run the examples, you need to give Java a special system-property named "java.library.path". The value of this property is the path of the directory where the libxapian_jni.so (or whatever extension is used on your platform) JNI library is located.
java -Djava.library.path=built -classpath built/xapian.jar:docs/examples \ SimpleIndex ./test.db index words like java java -Djava.library.path=built -classpath built/xapian.jar:docs/examples \ SimpleSearch ./test.db index words like java
Alternatively, you can avoid needing the -Djava.library.path setting by setting the LD_LIBRARY_PATH environment variable, or by installing the JNI library in the appropriate directory so your JVM finds it automatically (for example, on macOS you can copy it into /Library/Java/Extensions/ - you can also copy the .jar file there and avoid needing to specify it via -classpath).
The java bindings have been tested recently with OpenJDK versions 1.8.0_77, 1.7.0_03, and 1.6.0_38, but they should work with any java toolchain with suitable JNI support - please report success stories or any problems to the development mailing list: xapian-devel@lists.xapian.org
The Xapian C++ API is largely agnostic about character encoding, and uses the std::string type as an opaque container for a sequence of bytes. In places where the bytes represent text (for example, in the Stem, QueryParser and TermGenerator classes), UTF-8 encoding is used. In Java, the String class uses UTF-16 encoding, and can't hold arbitrary binary data.
The approach taken to this problem by these bindings (in Xapian 1.4.4 and later) is to map C++ std::string to/from Java byte arrays (byte[]) in places where the data is inherently binary (serialisation functions) or likely to be binary (document values).
This loses a bit of generality compared to the C++ API - for example, in C++ you can add a term with a binary data value but in Java it has to be a Unicode string. But users rarely actually need or want that generality, and losing it means that you can just work with Java String.
Document values work best when the values are compactly encoded, so a binary encoding is usually appropriate. However, if you really want to put a text value in a document value slot you can explicitly convert String to/from a byte array of UTF-8 data like so:
import java.nio.charset.StandardCharsets; //... doc.addValue(1, some_string.getBytes(StandardCharsets.UTF_8)); //... String value = new String(doc.getValue(1), StandardCharsets.UTF_8);
As well as terms, document data and user metadata are also required to be text at the moment when using these bindings.
Methods are renamed to match Java's naming conventions. So get_mset becomes getMSet, etc. Also get_description is wrapped as toString.
In Xapian 1.3.0 and later, these are wrapped as static constants Query.MatchAll and Query.MatchNothing.
If you want to be compatible with earlier versions, you can continue to use new Query("") instead of Query.MatchAll and new Query() instead of Query.MatchNothing.
Fix string passing to be zero-byte clean: https://trac.xapian.org/ticket/46
These were missing in the JNI bindings - it would be good to add them to SmokeTest.java:
- optional parameter "parameter" for Query ctor.
- changes to Enquire sorting API.
- new method ESet::back().
- Third (optional) argument to Document::add_posting().
- Xapian::Weight and standard subclasses.