<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.0 Transitional//EN">
<HTML>
<HEAD>
<TITLE>Xapian: Quickstart</TITLE>
</HEAD>
<BODY BGCOLOR="white">
<H1>Quickstart</H1>
<P>
The document contains a quick introduction to the basic concepts, and then
a walk-through development of a simple application using the Xapian
library, together with commentary on how the application could be taken
further. It deliberately avoids going into a lot of detail - see the
<a href="index.html">rest of the documentation</a> for more detail.
</P>
<HR>
<H2>Requirements</H2>
<P>
Before following the steps outlined in this document, you will need to have
the Xapian library installed on your system.
For instructions on obtaining and installing Xapian, read the
<A HREF="install.html">Installation</A> document.
</P>
<HR>
<H2>Databases</H2>
<P>
An information retrieval system using Xapian typically has two parts. The
first part is the <EM>indexer</EM>, which takes documents in various
formats, processes them so that they can be efficiently searched, and
stores the processed documents in an appropriate data structure (the
<EM>database</EM>). The second part is the <EM>searcher</EM>, which takes
queries and reads the database to return a list of the documents relevant
to each query.
</P>
<P>
The database is the data structure which ties the indexer and searcher
together, and is fundamental to the retrieval process. Given how
fundamental it is, it is unsurprising that different applications put
different demands on the database. For example, some applications may be
happy to deal with searching a static collection of data, but need to do
this extremely fast (for example, a web search engine which builds new
databases from scratch nightly or even weekly). Other applications may
require that new data can be added to the system incrementally, but don't
require extremely high performance searching (perhaps an email system,
which is only being searched occasionally). There are many other
constraints which may be placed on an information retrieval system: for
example, it may be required to have small database sizes, even at the
expense of getting poorer results from the system.
</P>
<P>
To provide the required flexibility, Xapian has the ability to use one of
many available database <EM>backends</EM>, each of which satisfies a
different set of constraints, and stores its data in a different way.
Currently, these must be compiled into the whole system, and selected at
runtime, but the ability to dynamically load modules for each of these
backends is likely to be added in future, and would require little design
modification.
</P>
<!--
<P>
If you are in a real hurry, you could probably skip the rest of this
section, but it is helpful to understand roughly what information Xapian
stores in a database and how it is structured, and the following
subsections detail this.
</P>
<H3>The contents of a database</H3>
<P>
FIXME: to be written.
Documents, terms, data, keys.
What can be accessed fast, what can't.
How each piece of data might be stored.
</P>
<H3><A NAME="flint_databases">Flint databases</A></H3>
<P>
FIXME: to be written.
</P>
-->
<HR>
<H2><A NAME="indexer">An example indexer</A></H2>
<P>
We now present sample code for an indexer. This is deliberately
simplified to make it easier to follow. You can also read it in <A
HREF="quickstartindex.cc.html">an HTML formatted version</A>.
</P>
<P>
The "indexer" presented here is simply a small program which
takes a path to a database and a set of parameters defining a document on
the command line, and stores that document as a new entry in the database.
</P>
<H3>Include header files</H3>
<P>
The first requirement in any program using the Xapian library is to
include the Xapian header file, "<CODE>xapian.h</CODE>":
<PRE> #include <xapian.h></PRE>
</P>
<P>
We're going to use C++ iostreams for output, so we need to include
the <CODE>iostream</CODE> header, and we'll also import everything
from namespace <CODE>std</CODE> for convenience:
<PRE> #include <iostream>
using namespace std;</PRE>
</P>
<P>
Our example only has a single function, <CODE>main()</CODE>, so next we
define that:
<PRE> int main(int argc, char **argv)</PRE>
</P>
<H3>Options parsing</H3>
<P>
For this example we do very simple options parsing. We are going to
use the core functionality of Xapian of searching for specific terms in the
database, and we are not going to use any of the extra facilities, such as
the keys which may be associated with each document. We are also going to
store a simple string as the data associated with each document.
</P><P>
Thus, our command line syntax is:
<UL><LI>
<B>Parameter 1</B> - the (possibly relative) path to the database.
</LI><LI>
<B>Parameter 2</B> - the string to be stored as the document data.
</LI><LI>
<B>Parameters 3 onward</B> - the terms to be stored in the database. The
terms will be assumed to occur at successive positions in the document.
</LI></UL>
</P><P>
The validity of a command line can therefore be checked very simply by
ensuring that there are at least 3 parameters:
<PRE>
if (argc < 4) {
cout << "usage: " << argv[0] <<
" <path to database> <document data> <document terms>" << endl;
exit(1);
}
</PRE>
</P>
<H3>Catching exceptions</H3>
<P>
When an error occurs in Xapian it is reported by means of the C++ exception
mechanism. All errors in Xapian are derived classes of
<CODE>Xapian::Error</CODE>, so simple error handling can be performed by
enclosing all the code in a try-catch block to catch any
<CODE>Xapian::Error</CODE> exceptions. A (hopefully) helpful message can be
extracted from the <CODE>Xapian::Error</CODE> object by calling its
<CODE>get_msg()</CODE> method, which returns a human readable string.
</P>
<P>
Note that all calls to the Xapian library should be performed inside a
try-catch block, since otherwise errors will result in uncaught exceptions;
this usually results in the execution aborting.
</P>
<P>
Note also that Xapian::Error is a virtual base class, and thus can't be copied:
you must therefore catch exceptions by reference, as in the following example
code:
</P>
<PRE>
try {
<B>[code which accesses Xapian]</B>
} catch (const Xapian::Error & error) {
cout << "Exception: " << error.get_msg() << endl;
}
</PRE>
<H3>Opening the database</H3>
<P>
In Xapian, a database is opened for writing by creating a
Xapian::WritableDatabase object.
</P>
<P>
If you pass Xapian::DB_CREATE_OR_OPEN and there isn't an existing database
in the specified directory, Xapian will try to create a new empty database
there. If there is already database in the specified directory, it will be
opened.
</P>
<P>
If an error occurs when trying to open a database, or to create a new database,
an exception, usually of type <CODE>Xapian::DatabaseOpeningError</CODE> or
<CODE>Xapian::DatabaseCreateError</CODE>, will be thrown.
</P>
<P>
The code to open a database for writing is, then:
</P>
<PRE>
Xapian::WritableDatabase database(argv[1], Xapian::DB_CREATE_OR_OPEN);
</PRE>
<H3>Preparing the new document</H3>
<P>
Now that we have the database open, we need to prepare a document to
put in it. This is done by creating a Xapian::Document object, filling
this with data, and then giving it to the database.
</P>
<P>
The first step, then, is to create the document:
</P>
<PRE>
Xapian::Document newdocument;
</PRE>
<P>
Each <code>Xapian::Document</code> has a "cargo" known as the <i>document data</i>.
This data is opaque to Xapian - the meaning of it is entirely user-defined.
Typically it contains information to allow results to be displayed by the
application, for example a URL for the indexed document and
some text which is to be displayed when returning the document as search
result.
</P>
<P>
For our example, we shall simply store the second parameter given on the
command line in the data field:
</P>
<PRE>
newdocument.set_data(string(argv[2]));
</PRE>
<P>
The next step is to put the terms which are to be used when searching
for the document into the Xapian::Document object.
</P>
<P>
We shall use the <CODE>add_posting()</CODE> method, which adds an
occurrence of a term to the struct. The first parameter is the
"<EM>termname</EM>", which is a string defining the term. This
string can be anything, as long as the same string is always used to refer
to the same term. The string will often be the (possibly stemmed) text
of the term, but might be in a compressed, or even hashed, form.
In general, there is no upper limit to the length of a termname, but some
database methods may impose their own limits.
</P>
<P>
The second parameter is the position at which the term occurs within the
document. These positions start at 1. This information is used for
some search features such as phrase matching or passage retrieval, but
is not essential to the search.
</P>
<P>
We add postings for terms with the termname given as each of the remaining
command line parameters:
</P>
<PRE>
for (int i = 3; i < argc; ++i) {
newdocument.add_posting(argv[i], i - 2);
}
</PRE>
<H3>Adding the document to the database</H3>
<P>
Finally, we can add the document to the database. This simply involves
calling <CODE>Xapian::WritableDatabase::add_document()</CODE>, and passing it
the <CODE>Xapian::Document</CODE> object:
</P>
<PRE>
database.add_document(newdocument);
</PRE>
<P>
The operation of adding a document is atomic: either the document will be
added, or an exception will be thrown and the document will not be in the
new database.
</P>
<P>
<CODE>add_document()</CODE> returns a value of type <CODE>Xapian::docid</CODE>.
This is the document ID of the newly added document, which is simply a
handle which can be used to access the document in future.
</P>
<P>
Note that this use of <CODE>add_document()</CODE> is actually fairly
inefficient: if we had a large database, it would be desirable to group
as many document additions together as possible, by encapsulating
them within a session. For details of this, and of the transaction
facility for performing sets of database modifications atomically, see
the <A HREF="overview.html">API Overview</A>.
</P>
<HR>
<H2><A NAME="searcher">An example searcher</A></H2>
<P>
Now we show the code for a simple searcher, which will search the
database built by the indexer above. Again, you can read <A
HREF="quickstartsearch.cc.html">an HTML formatted version</A>.
</P>
<P>
The "searcher" presented here is, like the "indexer",
simply a small command line driven program. It takes a path to a database
and some search terms, performs a probabilistic search for documents
represented by those terms and displays a ranked list of matching documents.
</P>
<H3>Setting up</H3>
<P>
Just like "quickstartindex", we have a single-function example.
So we include the Xapian header file, and begin:
</P>
<PRE>
#include <xapian.h>
int main(int argc, char **argv)
{
</PRE>
<H3>Options parsing</H3>
<P>
Again, we are going to use no special options, and have a very simple
command line syntax:
<UL><LI>
<B>Parameter 1</B> - the (possibly relative) path to the database.
</LI><LI>
<B>Parameters 2 onward</B> - the terms to be searched for in the database.
</LI></UL>
</P><P>
The validity of a command line can therefore be checked very simply by
ensuring that there are at least 2 parameters:
</P>
<PRE>
if (argc < 3) {
cout << "usage: " << argv[0] <<
" <path to database> <search terms>" << endl;
exit(1);
}
</PRE>
</P>
<H3>Catching exceptions</H3>
<P>
Again, this is performed just as it was for the simple indexer.
</P>
<PRE>
try {
<B>[code which accesses Xapian]</B>
} catch (const Xapian::Error & error) {
cout << "Exception: " << error.get_msg() << endl;
}
</PRE>
<H3>Specifying the databases</H3>
<P>
Xapian has the ability to search over many databases simultaneously,
possibly even with the databases distributed across a network of machines.
Each database can be in its own format, so, for example, we might have a
system searching across two remote databases and a flint database.
</P>
<P>
To open a single database, we create a Xapian::Database object, passing
the path to the database we want to open:
</P>
<PRE>
Xapian::Database db(argv[1]);
</PRE>
<P>
You can also search multiple database by adding them together using
<CODE>Xapian::Database::add_database</CODE>:
</P>
<PRE>
Xapian::Database databases;
databases.add_database(Xapian::Database(argv[1]));
databases.add_database(Xapian::Database(argv[2]));
</PRE>
<H3>Starting an enquire session</H3>
<P>
All searches across databases by Xapian are performed within the context of
an "<EM>Enquire</EM>" session. This session is represented by a
<CODE>Xapian::Enquire</CODE> object, and is across a specified collection of
databases. To change the database collection, it is necessary to open a
new enquire session, by creating a new <CODE>Xapian::Enquire</CODE> object.
<PRE>
Xapian::Enquire enquire(databases);
</PRE>
</P>
<P>
An enquire session is also the context within which all other database
reading operations, such as query expansion and reading the data associated
with a document, are performed.
</P>
<H3>Preparing to search</H3>
<P>
We are going to use all command line parameters from the second onward
as terms to search for in the database. For convenience, we shall store
them in an STL vector. This is probably the point at which we would want
to apply a stemming algorithm, or any other desired normalisation and
conversion operation, to the terms.
<PRE>
vector<string> queryterms;
for (int optpos = 2; optpos < argc; optpos++) {
queryterms.push_back(argv[optpos]);
}
</PRE>
</P>
<P>
Queries are represented within Xapian by <CODE>Xapian::Query</CODE> objects, so
the next step is to construct one from our query terms.
Conveniently there is a constructor which will take our vector
of terms and create an <CODE>Xapian::Query</CODE> object from it.
<PRE>
Xapian::Query query(Xapian::Query::OP_OR, queryterms.begin(), queryterms.end());
</PRE>
</P>
<P>
You will notice that we had to specify an operation to be performed on
the terms (the <CODE>Xapian::Query::OP_OR</CODE> parameter).
Queries in Xapian are actually
fairly complex things: a full range of boolean operations can be applied to
queries to restrict the result set, and probabilistic weightings are then
applied to order the results by relevance. By specifying the OR operation,
we are not performing any boolean restriction, and are performing a
traditional pure probabilistic search.
</P>
<P>
We now print a message out to confirm to the user what the query being
performed is. This is done with the <CODE>Xapian::Query::get_description()</CODE>
method, which is mainly included for debugging purposes, and displays
a string representation of the query.
</P>
<PRE>
cout << "Performing query `" <<
query.get_description() << "'" << endl;
</PRE>
<H3>Performing the search</H3>
<P>
Now, we are ready to perform the search. The first step of this is to
give the query object to the enquire session. Note that the query is
copied at this operation, and that changing the Xapian::Query object after
setting the query with it has no effect.
</P>
<PRE>
enquire.set_query(query);
</PRE>
<P>
Next, we ask for the results of the search. There is no need to tell
Xapian to perform the search: it will do this automatically. We use
the <CODE>get_mset()</CODE> method to get the results, which are returned
in an <CODE>Xapian::MSet</CODE> object. (MSet for Match Set)
</P>
<P>
<CODE>get_mset()</CODE> can take many parameters, such as a set of
relevant documents to use, and various options to modify the search,
but we give it the minimum; which is the first document to return (starting
at 0 for the top ranked document), and the maximum number of documents
to return (we specify 10 here):
<PRE>
Xapian::MSet matches = enquire.get_mset(0, 10);
</PRE>
</P>
<H3>Displaying the results of the search</H3>
<P>
Finally, we display the results of the search. The results are stored in
in the <CODE>Xapian::MSet</CODE> object, which provides the features required
to be an STL-compatible container, so first we display how many items are in
the MSet:
<PRE>
cout << matches.size() << " results found" << endl;
</PRE>
</P>
<P>
Now we display some information about each of the items in the
<CODE>Xapian::MSet</CODE>. We access these items using an
<CODE>Xapian::MSetIterator</CODE>:
<UL><LI>
First, we display the document ID, accessed by <CODE>*i</CODE>.
This is not usually very useful information to give to users, but it is
at least a unique handle on each document.
</LI><LI>
Next, we display a "percentage" score for the document. Readers
familiar with Information Retrieval will not be surprised to hear that this
is not really a percentage: it is just a value from 0 to 100, such that a
more relevant document has a higher value. We get this using
<CODE>i.get_percent()</CODE>.
</LI><LI>
Last, we display the data associated with each returned document, which
was specified by the user at database generation time. To do this, we
first use <CODE>i.get_document()</CODE> to get an <CODE>Xapian::Document</CODE>
object representing the returned document; then we use the
<CODE>get_data()</CODE> method of this object to get
access to the data stored in this document.
</LI></UL>
<PRE>
Xapian::MSetIterator i;
for (i = matches.begin(); i != matches.end(); ++i) {
cout << "Document ID " << *i << "\t";
cout << i.get_percent() << "% ";
Xapian::Document doc = i.get_document();
cout << "[" << doc.get_data() << "]" << endl;
}
</PRE>
</P>
<HR>
<H2>Compiling</H2>
Now that we have the code written, all we need to do is compile it!
<H3>Finding the Xapian library</H3>
<P>
A small utility, "xapian-config", is installed along with Xapian
to assist you in finding the installed Xapian library, and in generating
the flags to pass to the compiler and linker to compile.
</P><P>
After a successful compilation, this utility should be in your path, so
you can simply run
<BLOCKQUOTE><CODE>xapian-config --cxxflags</CODE></BLOCKQUOTE>
to determine the flags to pass to the compiler, and
<BLOCKQUOTE><CODE>xapian-config --libs</CODE></BLOCKQUOTE>
to determine the flags to pass to the linker.
These flags are returned on the utility's standard output (so you could use
backtick notation to include them on your command line).
</P><P>
If your project uses the GNU autoconf tool, you may also use the
<CODE>XO_LIB_XAPIAN</CODE> macro, which is included as part of Xapian,
and will check for an installation of Xapian and set (and
<CODE>AC_SUBST</CODE>) the <CODE>XAPIAN_CXXFLAGS</CODE> and
<CODE>XAPIAN_LIBS</CODE> variables to
be the flags to pass to the compiler and linker, respectively.
</P><P>
If you don't use GNU autoconf, don't worry about this.
</P>
<H3>Compiling the quickstart examples</H3>
Once you know the compilation flags, compilation is a simple matter of
invoking the compiler! For our example, we could compile the two
utilities (quickstartindex and quickstartsearch) with the commands:
<PRE>
c++ quickstartindex.cc `xapian-config --libs --cxxflags` -o quickstartindex
c++ quickstartsearch.cc `xapian-config --libs --cxxflags` -o quickstartsearch
</PRE>
<HR>
<H2>Running the examples</H2>
<P>
Once we have compiled the above examples, we can build up a simple
database as follows. Note that we must first create a directory for
the database files to live in; although Xapian will create new empty
database files if they do not yet exist, it will not create a new
directory for them.
<PRE>
$ mkdir proverbs
$ ./quickstartindex proverbs \
> "people who live in glass houses should not throw stones" \
> people live glass house stone
$ ./quickstartindex proverbs \
> "Don't look a gift horse in the mouth" \
> look gift horse mouth
</PRE>
</P>
<P>
Now, we should have a database with a couple of documents in it. Looking
in the database directory, you should see something like:
<PRE>
$ ls proverbs/
<i>[some files]</i>
</PRE>
</P>
<P>
Given the small amount of data in the database, you may be concerned that
the total size of these files is somewhat over 50k. Be reassured that the
database is block structured, here consisting of largely empty
blocks, and will behave much better for large databases.
</P>
<P>
We can now perform searches over the database using the quickstartsearch
program.
<PRE>
$ ./quickstartsearch proverbs look
Performing query `look'
1 results found
Document ID 2 50% [Don't look a gift horse in the mouth]
</PRE>
</P>
<!-- FOOTER $Author: olly $ $Date: 2008-03-05 06:28:18 +0000 (Wed, 05 Mar 2008) $ $Id: quickstart.html 10165 2008-03-05 06:28:18Z olly $ -->
</BODY>
</HTML>
