Add pretty printing to yaz-json-parse

[yaz-moved-to-github.git] / doc / tools.xml

 <chapter id="tools"><title>Supporting Tools</title>
  
  <para>
   In support of the service API - primarily the ASN module, which
   provides the pro-grammatic interface to the Z39.50 APDUs, &yaz; contains
   a collection of tools that support the development of applications.
  </para>

  <sect1 id="tools.query"><title>Query Syntax Parsers</title>

   <para>
    Since the type-1 (RPN) query structure has no direct, useful string
    representation, every origin application needs to provide some form of
    mapping from a local query notation or representation to a
    <token>Z_RPNQuery</token> structure. Some programmers will prefer to
    construct the query manually, perhaps using
    <function>odr_malloc()</function> to simplify memory management.
    The &yaz; distribution includes three separate, query-generating tools
    that may be of use to you.
   </para>

   <sect2 id="PQF"><title>Prefix Query Format</title>

    <para>
     Since RPN or reverse polish notation is really just a fancy way of
     describing a suffix notation format (operator follows operands), it
     would seem that the confusion is total when we now introduce a prefix
     notation for RPN. The reason is one of simple laziness - it's somewhat
     simpler to interpret a prefix format, and this utility was designed
     for maximum simplicity, to provide a baseline representation for use
     in simple test applications and scripting environments (like Tcl). The
     demonstration client included with YAZ uses the PQF.
    </para>

    <note>
     <para>
      The PQF have been adopted by other parties developing Z39.50
      software. It is often referred to as Prefix Query Notation
      - PQN.
     </para>
    </note>
    <para>
     The PQF is defined by the pquery module in the YAZ library. 
     There are two sets of function that have similar behavior. First
     set operates on a PQF parser handle, second set doesn't. First set
     set of functions are more flexible than the second set. Second set
     is obsolete and is only provided to ensure backwards compatibility.
    </para>
    <para>
     First set of functions all operate on a PQF parser handle:
    </para>
    <synopsis>
     #include &lt;yaz/pquery.h&gt;

     YAZ_PQF_Parser yaz_pqf_create (void);

     void yaz_pqf_destroy (YAZ_PQF_Parser p);

     Z_RPNQuery *yaz_pqf_parse (YAZ_PQF_Parser p, ODR o, const char *qbuf);

     Z_AttributesPlusTerm *yaz_pqf_scan (YAZ_PQF_Parser p, ODR o,
                          Odr_oid **attributeSetId, const char *qbuf);


     int yaz_pqf_error (YAZ_PQF_Parser p, const char **msg, size_t *off);
    </synopsis>
    <para>
     A PQF parser is created and destructed by functions
     <function>yaz_pqf_create</function> and
     <function>yaz_pqf_destroy</function> respectively.
     Function <function>yaz_pqf_parse</function> parses query given
     by string <literal>qbuf</literal>. If parsing was successful,
     a Z39.50 RPN Query is returned which is created using ODR stream
     <literal>o</literal>. If parsing failed, a NULL pointer is
     returned.
     Function <function>yaz_pqf_scan</function> takes a scan query in 
     <literal>qbuf</literal>. If parsing was successful, the function
     returns attributes plus term pointer and modifies
     <literal>attributeSetId</literal> to hold attribute set for the
     scan request - both allocated using ODR stream <literal>o</literal>.
     If parsing failed, yaz_pqf_scan returns a NULL pointer.
     Error information for bad queries can be obtained by a call to
     <function>yaz_pqf_error</function> which returns an error code and
     modifies <literal>*msg</literal> to point to an error description,
     and modifies <literal>*off</literal> to the offset within last
     query were parsing failed.
    </para>
    <para>
     The second set of functions are declared as follows:
    </para>
    <synopsis>
     #include &lt;yaz/pquery.h&gt;

     Z_RPNQuery *p_query_rpn (ODR o, oid_proto proto, const char *qbuf);

     Z_AttributesPlusTerm *p_query_scan (ODR o, oid_proto proto,
                             Odr_oid **attributeSetP, const char *qbuf);

     int p_query_attset (const char *arg);
    </synopsis>
    <para>
     The function <function>p_query_rpn()</function> takes as arguments an
      &odr; stream (see section <link linkend="odr">The ODR Module</link>)
     to provide a memory source (the structure created is released on
     the next call to <function>odr_reset()</function> on the stream), a
     protocol identifier (one of the constants <token>PROTO_Z3950</token> and
     <token>PROTO_SR</token>), an attribute set reference, and
     finally a null-terminated string holding the query string.
    </para>
    <para>
     If the parse went well, <function>p_query_rpn()</function> returns a
     pointer to a <literal>Z_RPNQuery</literal> structure which can be
     placed directly into a <literal>Z_SearchRequest</literal>. 
     If parsing failed, due to syntax error, a NULL pointer is returned.
    </para>
    <para>
     The <literal>p_query_attset</literal> specifies which attribute set
     to use if the query doesn't specify one by the
     <literal>@attrset</literal> operator.
     The <literal>p_query_attset</literal> returns 0 if the argument is a
     valid attribute set specifier; otherwise the function returns -1.
    </para>

    <para>
     The grammar of the PQF is as follows:
    </para>

    <literallayout>
     query ::= top-set query-struct.

     top-set ::= [ '@attrset' string ]

     query-struct ::= attr-spec | simple | complex | '@term' term-type query

     attr-spec ::= '@attr' [ string ] string query-struct

     complex ::= operator query-struct query-struct.

     operator ::= '@and' | '@or' | '@not' | '@prox' proximity.

     simple ::= result-set | term.

     result-set ::= '@set' string.

     term ::= string.

     proximity ::= exclusion distance ordered relation which-code unit-code.

     exclusion ::= '1' | '0' | 'void'.

     distance ::= integer.

     ordered ::= '1' | '0'.

     relation ::= integer.

     which-code ::= 'known' | 'private' | integer.

     unit-code ::= integer.

     term-type ::= 'general' | 'numeric' | 'string' | 'oid' | 'datetime' | 'null'.
    </literallayout>

    <para>
     You will note that the syntax above is a fairly faithful
     representation of RPN, except for the Attribute, which has been
     moved a step away from the term, allowing you to associate one or more
     attributes with an entire query structure. The parser will
     automatically apply the given attributes to each term as required.
    </para>

    <para>
     The @attr operator is followed by an attribute specification 
     (<literal>attr-spec</literal> above). The specification consists
     of an optional attribute set, an attribute type-value pair and
     a sub-query. The attribute type-value pair is packed in one string:
     an attribute type, an equals sign, and an attribute value, like this:
     <literal>@attr 1=1003</literal>.
     The type is always an integer but the value may be either an
     integer or a string (if it doesn't start with a digit character).
     A string attribute-value is encoded as a Type-1 ``complex''
     attribute with the list of values containing the single string
     specified, and including no semantic indicators.
    </para>

    <para>
     Version 3 of the Z39.50 specification defines various encoding of terms.
     Use <literal>@term </literal> <replaceable>type</replaceable>
     <replaceable>string</replaceable>,
     where type is one of: <literal>general</literal>,
     <literal>numeric</literal> or <literal>string</literal>
     (for InternationalString).
     If no term type has been given, the <literal>general</literal> form
     is used.  This is the only encoding allowed in both versions 2 and 3
     of the Z39.50 standard.
    </para>
    
    <sect3 id="PQF-prox">
      <title>Using Proximity Operators with PQF</title>
      <note>
        <para>
          This is an advanced topic, describing how to construct
          queries that make very specific requirements on the
          relative location of their operands.
          You may wish to skip this section and go straight to
          <link linkend="pqf-examples">the example PQF queries</link>.
        </para>
        <para>
          <warning>
            <para>
              Most Z39.50 servers do not support proximity searching, or
              support only a small subset of the full functionality that
              can be expressed using the PQF proximity operator.  Be
              aware that the ability to <emphasis>express</emphasis> a
              query in PQF is no guarantee that any given server will
              be able to <emphasis>execute</emphasis> it.
            </para>
          </warning>
        </para>
      </note>
      <para>
        The proximity operator <literal>@prox</literal> is a special
        and more restrictive version of the conjunction operator
        <literal>@and</literal>.  Its semantics are described in 
        section 3.7.2 (Proximity) of Z39.50 the standard itself, which
        can be read on-line at
        <ulink url="&url.z39.50.proximity;"/>
      </para>
      <para>
        In PQF, the proximity operation is represented by a sequence
        of the form
        <screen>
@prox <replaceable>exclusion</replaceable> <replaceable>distance</replaceable> <replaceable>ordered</replaceable> <replaceable>relation</replaceable> <replaceable>which-code</replaceable> <replaceable>unit-code</replaceable>
        </screen>
        in which the meanings of the parameters are as described in in
        the standard, and they can take the following values:
        <itemizedlist>
          <listitem><formalpara><title>exclusion</title><para>
            0 = false (i.e. the proximity condition specified by the
            remaining parameters must be satisfied) or
            1 = true (the proximity condition specified by the
            remaining parameters must <emphasis>not</emphasis> be
            satisifed).
          </para></formalpara></listitem>
          <listitem><formalpara><title>distance</title><para>
            An integer specifying the difference between the locations
            of the operands: e.g. two adjacent words would have
            distance=1 since their locations differ by one unit.
          </para></formalpara></listitem>
          <listitem><formalpara><title>ordered</title><para>
            1 = ordered (the operands must occur in the order the
            query specifies them) or
            0 = unordered (they may appear in either order).
          </para></formalpara></listitem>
          <listitem><formalpara><title>relation</title><para>
            Recognised values are
            1 (lessThan),
            2 (lessThanOrEqual),
            3 (equal),
            4 (greaterThanOrEqual),
            5 (greaterThan) and
            6 (notEqual).
          </para></formalpara></listitem>
          <listitem><formalpara><title>which-code</title><para>
            <literal>known</literal>
            or
            <literal>k</literal>
            (the unit-code parameter is taken from the well-known list
            of alternatives described in below) or
            <literal>private</literal>
            or
            <literal>p</literal>
            (the unit-code paramater has semantics specific to an
            out-of-band agreement such as a profile).
          </para></formalpara></listitem>
          <listitem><formalpara><title>unit-code</title><para>
            If the which-code parameter is <literal>known</literal>
            then the recognised values are
            1 (character),
            2 (word),
            3 (sentence),
            4 (paragraph),
            5 (section),
            6 (chapter),
            7 (document),
            8 (element),
            9 (subelement),
            10 (elementType) and
            11 (byte).
            If which-code is <literal>private</literal> then the
            acceptable values are determined by the profile.
          </para></formalpara></listitem>
        </itemizedlist>
        (The numeric values of the relation and well-known unit-code
        parameters are taken straight from
        <ulink url="&url.z39.50.proximity.asn1;"
        >the ASN.1</ulink> of the proximity structure in the standard.)
      </para>
    </sect3>

    <sect3 id="pqf-examples"><title>PQF queries</title>

     <example id="example.pqf.simple.terms">
      <title>PQF queries using simple terms</title>
      <para>
       <screen>
        dylan

        "bob dylan"
       </screen>
      </para>
     </example>
     <example id="pqf.example.pqf.boolean.operators">
      <title>PQF boolean operators</title>
      <para>
       <screen>
        @or "dylan" "zimmerman"

        @and @or dylan zimmerman when

        @and when @or dylan zimmerman
       </screen>
      </para>
     </example>
     <example id="example.pqf.result.sets">
      <title>PQF references to result sets</title>
      <para>
       <screen>
        @set Result-1

        @and @set seta @set setb
       </screen>
      </para>
     </example>
     <example id="example.pqf.attributes">
      <title>Attributes for terms</title>
      <para>
       <screen>
        @attr 1=4 computer

        @attr 1=4 @attr 4=1 "self portrait"

        @attrset exp1 @attr 1=1 CategoryList

        @attr gils 1=2008 Copenhagen

        @attr 1=/book/title computer
       </screen>
      </para>
     </example>
     <example id="example.pqf.proximity">
      <title>PQF Proximity queries</title>
      <para>
       <screen>
        @prox 0 3 1 2 k 2 dylan zimmerman
       </screen>
       <note><para>
         Here the parameters 0, 3, 1, 2, k and 2 represent exclusion,
         distance, ordered, relation, which-code and unit-code, in that
         order.  So:
         <itemizedlist>
          <listitem><para>
            exclusion = 0: the proximity condition must hold
           </para></listitem>
          <listitem><para>
            distance = 3: the terms must be three units apart
           </para></listitem>
          <listitem><para>
            ordered = 1: they must occur in the order they are specified
           </para></listitem>
          <listitem><para>
            relation = 2: lessThanOrEqual (to the distance of 3 units)
           </para></listitem>
          <listitem><para>
            which-code is ``known'', so the standard unit-codes are used
           </para></listitem>
          <listitem><para>
            unit-code = 2: word.
           </para></listitem>
         </itemizedlist>
         So the whole proximity query means that the words
         <literal>dylan</literal> and <literal>zimmerman</literal> must
         both occur in the record, in that order, differing in position
         by three or fewer words (i.e. with two or fewer words between
         them.)  The query would find ``Bob Dylan, aka. Robert
         Zimmerman'', but not ``Bob Dylan, born as Robert Zimmerman''
         since the distance in this case is four.
        </para></note>
      </para>
     </example>
     <example id="example.pqf.search.term.type">
      <title>PQF specification of search term type</title>
      <para>
       <screen>
        @term string "a UTF-8 string, maybe?"
       </screen>
      </para>
     </example>
     <example id="example.pqf.mixed.queries">
      <title>PQF mixed queries</title>
      <para>
       <screen>
        @or @and bob dylan @set Result-1
        
        @attr 4=1 @and @attr 1=1 "bob dylan" @attr 1=4 "slow train coming"
        
        @and @attr 2=4 @attr gils 1=2038 -114 @attr 2=2 @attr gils 1=2039 -109
      </screen>
       <note>
        <para>
         The last of these examples is a spatial search: in
         <ulink url="http://www.gils.net/prof_v2.html#sec_7_4"
          >the GILS attribute set</ulink>,
         access point
         2038 indicates West Bounding Coordinate and
         2030 indicates East Bounding Coordinate,
         so the query is for areas extending from -114 degrees
         to no more than -109 degrees.
        </para>
       </note>
      </para>
     </example>
    </sect3>
   </sect2>
   <sect2 id="CCL"><title>CCL</title>

    <para>
     Not all users enjoy typing in prefix query structures and numerical
     attribute values, even in a minimalistic test client. In the library
     world, the more intuitive Common Command Language - CCL (ISO 8777)
     has enjoyed some popularity - especially before the widespread
     availability of graphical interfaces. It is still useful in
     applications where you for some reason or other need to provide a
     symbolic language for expressing boolean query structures.
    </para>

    <sect3 id="ccl.syntax">
     <title>CCL Syntax</title>

     <para>
      The CCL parser obeys the following grammar for the FIND argument.
      The syntax is annotated by in the lines prefixed by
      <literal>--</literal>.
     </para>

     <screen>
      CCL-Find ::= CCL-Find Op Elements
                | Elements.

      Op ::= "and" | "or" | "not"
      -- The above means that Elements are separated by boolean operators.

      Elements ::= '(' CCL-Find ')'
                | Set
                | Terms
                | Qualifiers Relation Terms
                | Qualifiers Relation '(' CCL-Find ')'
                | Qualifiers '=' string '-' string
      -- Elements is either a recursive definition, a result set reference, a
      -- list of terms, qualifiers followed by terms, qualifiers followed
      -- by a recursive definition or qualifiers in a range (lower - upper).

      Set ::= 'set' = string
      -- Reference to a result set

      Terms ::= Terms Prox Term
             | Term
      -- Proximity of terms.

      Term ::= Term string
            | string
      -- This basically means that a term may include a blank

      Qualifiers ::= Qualifiers ',' string
                  | string
      -- Qualifiers is a list of strings separated by comma

      Relation ::= '=' | '>=' | '&lt;=' | '&lt;>' | '>' | '&lt;'
      -- Relational operators. This really doesn't follow the ISO8777
      -- standard.

      Prox ::= '%' | '!'
      -- Proximity operator

     </screen>
     
     <example id="example.ccl.queries">
      <title>CCL queries</title>
      <para>
       The following queries are all valid:
      </para>
      
      <screen>
       dylan
       
       "bob dylan"
       
       dylan or zimmerman
       
       set=1
       
       (dylan and bob) or set=1
       
      </screen>
      <para>
       Assuming that the qualifiers <literal>ti</literal>,
       <literal>au</literal>
       and <literal>date</literal> are defined we may use:
      </para>
      
      <screen>
       ti=self portrait
       
       au=(bob dylan and slow train coming)

       date>1980 and (ti=((self portrait)))
       
      </screen>
     </example>
     
    </sect3>
    <sect3 id="ccl.qualifiers">
     <title>CCL Qualifiers</title>
     
     <para>
      Qualifiers are used to direct the search to a particular searchable
      index, such as title (ti) and author indexes (au). The CCL standard
      itself doesn't specify a particular set of qualifiers, but it does
      suggest a few short-hand notations. You can customize the CCL parser
      to support a particular set of qualifiers to reflect the current target
      profile. Traditionally, a qualifier would map to a particular
      use-attribute within the BIB-1 attribute set. It is also
      possible to set other attributes, such as the structure
      attribute.
     </para>

     <para>
      A  CCL profile is a set of predefined CCL qualifiers that may be
      read from a file or set in the CCL API.
      The YAZ client reads its CCL qualifiers from a file named
      <filename>default.bib</filename>. There are four types of
      lines in a CCL profile: qualifier specification,
      qualifier alias, comments and directives.
     </para>
     <sect4 id="ccl.qualifier.specification">
      <title>Qualifier specification</title>
      <para>
       A qualifier specification is of the form:
      </para>
      
      <para>
       <replaceable>qualifier-name</replaceable>  
       [<replaceable>attributeset</replaceable><literal>,</literal>]<replaceable>type</replaceable><literal>=</literal><replaceable>val</replaceable>
       [<replaceable>attributeset</replaceable><literal>,</literal>]<replaceable>type</replaceable><literal>=</literal><replaceable>val</replaceable> ...      
      </para>
      
      <para>
       where <replaceable>qualifier-name</replaceable> is the name of the
       qualifier to be used (eg. <literal>ti</literal>),
       <replaceable>type</replaceable> is attribute type in the attribute
       set (Bib-1 is used if no attribute set is given) and
       <replaceable>val</replaceable> is attribute value.
       The <replaceable>type</replaceable> can be specified as an
       integer or as it be specified either as a single-letter:
       <literal>u</literal> for use, 
       <literal>r</literal> for relation,<literal>p</literal> for position,
       <literal>s</literal> for structure,<literal>t</literal> for truncation
       or <literal>c</literal> for completeness.
       The attributes for the special qualifier name <literal>term</literal>
       are used when no CCL qualifier is given in a query.
       <table id="ccl.common.bib1.attributes">
        <title>Common Bib-1 attributes</title>
        <tgroup cols="2">
         <colspec colwidth="2*" colname="type"></colspec>
         <colspec colwidth="9*" colname="description"></colspec>
         <thead>
          <row>
           <entry>Type</entry>
           <entry>Description</entry>
          </row>
         </thead>
         <tbody>
          <row>
           <entry><literal>u=</literal><replaceable>value</replaceable></entry>
           <entry>
            Use attribute (1). Common use attributes are
            1 Personal-name, 4 Title, 7 ISBN, 8 ISSN, 30 Date,
            62 Subject, 1003 Author), 1016 Any. Specify value
            as an integer.
           </entry>
          </row>

          <row>
           <entry><literal>r=</literal><replaceable>value</replaceable></entry>
           <entry>
            Relation attribute (2). Common values are
            1 &lt;, 2 &lt;=, 3 =, 4 &gt;=, 5 &gt;, 6 &lt;&gt;,
            100 phonetic, 101 stem, 102 relevance, 103 always matches.
           </entry>
          </row>

          <row>
           <entry><literal>p=</literal><replaceable>value</replaceable></entry>
           <entry>
            Position attribute (3). Values: 1 first in field, 2
            first in any subfield, 3 any position in field.
           </entry>
          </row>

          <row>
           <entry><literal>s=</literal><replaceable>value</replaceable></entry>
           <entry>
            Structure attribute (4). Values: 1 phrase, 2 word,
            3 key, 4 year, 5 date, 6 word list, 100 date (un),
            101 name (norm), 102 name (un), 103 structure, 104 urx,
            105 free-form-text, 106 document-text, 107 local-number,
            108 string, 109 numeric string.
           </entry>
          </row>

          <row>
           <entry><literal>t=</literal><replaceable>value</replaceable></entry>
           <entry>
            Truncation attribute (5). Values: 1 right, 2 left,
            3 left&amp; right, 100 none, 101 process #, 102 regular-1,
            103 regular-2, 104 CCL.
           </entry>
          </row>

          <row>
           <entry><literal>c=</literal><replaceable>value</replaceable></entry>
           <entry>
            Completeness attribute (6). Values: 1 incomplete subfield,
            2 complete subfield, 3 complete field.
           </entry>
          </row>

         </tbody>
         </tgroup>
        </table>
      </para>
      <para>
       Refer to <xref linkend="bib1"/> or the complete
       <ulink url="&url.z39.50.attset.bib1;">list of Bib-1 attributes</ulink>
      </para>
      <para>
       It is also possible to specify non-numeric attribute values, 
       which are used in combination with certain types.
       The special combinations are:
       
       <table id="ccl.special.attribute.combos">
        <title>Special attribute combos</title>
        <tgroup cols="2">
         <colspec colwidth="2*" colname="name"></colspec>
         <colspec colwidth="9*" colname="description"></colspec>
         <thead>
          <row>
           <entry>Name</entry>
           <entry>Description</entry>
          </row>
         </thead>
         <tbody>
          <row>
           <entry><literal>s=pw</literal></entry><entry>
            The structure is set to either word or phrase depending
            on the number of tokens in a term (phrase-word).
           </entry>
          </row>
          <row>
           <entry><literal>s=al</literal></entry><entry>
            Each token in the term is ANDed. (and-list).
            This does not set the structure at all.
           </entry>
          </row>
          
          <row><entry><literal>s=ol</literal></entry><entry>
            Each token in the term is ORed. (or-list).
            This does not set the structure at all.
           </entry>
          </row>
          
          <row><entry><literal>r=o</literal></entry><entry>
            Allows ranges and the operators greather-than, less-than, ...
            equals.
            This sets Bib-1 relation attribute accordingly (relation
            ordered). A query construct is only treated as a range if
            dash is used and that is surrounded by white-space. So
            <literal>-1980</literal> is treated as term 
            <literal>"-1980"</literal> not <literal>&lt;= 1980</literal>.
            If <literal>- 1980</literal> is used, however, that is
            treated as a range.
           </entry>
          </row>
          
          <row><entry><literal>r=r</literal></entry><entry>
            Similar to <literal>r=o</literal> but assumes that terms
            are non-negative (not prefixed with <literal>-</literal>).
            Thus, a dash will always be treated as a range.
            The construct <literal>1980-1990</literal> is
            treated as a range with <literal>r=r</literal> but as a
            single term <literal>"1980-1990"</literal> with
            <literal>r=o</literal>. The special attribute
            <literal>r=r</literal> is available in YAZ 2.0.24 or later.
           </entry>
          </row>
          
          <row><entry><literal>t=l</literal></entry><entry>
            Allows term to be left-truncated.
            If term is of the form <literal>?x</literal>, the resulting
            Type-1 term is <literal>x</literal> and truncation is left.
           </entry>
          </row>
          
          <row><entry><literal>t=r</literal></entry><entry>
            Allows term to be right-truncated.
            If term is of the form <literal>x?</literal>, the resulting
            Type-1 term is <literal>x</literal> and truncation is right.
           </entry>
          </row>
          
          <row><entry><literal>t=n</literal></entry><entry>
            If term is does not include <literal>?</literal>, the
            truncation attribute is set to none (100).
           </entry>
          </row>
          
          <row><entry><literal>t=b</literal></entry><entry>
            Allows term to be both left&amp;right truncated.
            If term is of the form <literal>?x?</literal>, the
            resulting term is <literal>x</literal> and trunctation is
            set to both left&amp;right.
           </entry>
          </row>

          <row><entry><literal>t=x</literal></entry><entry>
            Allows masking anywhere in a term, thus fully supporting
            # (mask one character) and ? (zero or more of any).
            If masking is used, trunction is set to 102 (regexp-1 in term)
            and the term is converted accordingly to a regular expression.
           </entry>
          </row>

          <row><entry><literal>t=z</literal></entry><entry>
            Allows masking anywhere in a term, thus fully supporting
            # (mask one character) and ? (zero or more of any).
            If masking is used, trunction is set to 104 (Z39.58 in term)
            and the term is converted accordingly to Z39.58 masking term -
            actually the same truncation as CCL itself.
           </entry>
          </row>

         </tbody>
        </tgroup>
       </table>
      </para>
      <example id="example.ccl.profile"><title>CCL profile</title>
       <para>
        Consider the following definition:
       </para>
       
       <screen>
        ti       u=4 s=1
        au       u=1 s=1
        term     s=105
        ranked   r=102
        date     u=30 r=o
      </screen>
       <para>
        <literal>ti</literal> and <literal>au</literal> both set 
        structure attribute to phrase (s=1).
        <literal>ti</literal>
        sets the use-attribute to 4. <literal>au</literal> sets the
        use-attribute to 1.
        When no qualifiers are used in the query the structure-attribute is
        set to free-form-text (105) (rule for <literal>term</literal>).
        The <literal>date</literal> sets the relation attribute to
        the relation used in the CCL query and sets the use attribute
        to 30 (Bib-1 Date).
       </para>
       <para>
        You can combine attributes. To Search for "ranked title" you
        can do 
        <screen>
         ti,ranked=knuth computer
        </screen>
        which will set relation=ranked, use=title, structure=phrase.
       </para>
       <para>
        Query
        <screen>
         date > 1980
        </screen>
        is a valid query. But
        <screen>
         ti > 1980
        </screen>
        is invalid.
       </para>
      </example>
     </sect4>
     <sect4 id="ccl.qualifier.alias">
      <title>Qualifier alias</title>
      <para>
       A qualifier alias is of the form:
      </para>
      <para>
       <replaceable>q</replaceable>  
       <replaceable>q1</replaceable> <replaceable>q2</replaceable> ..
      </para>
      <para>
       which declares <replaceable>q</replaceable> to
       be an alias for <replaceable>q1</replaceable>, 
       <replaceable>q2</replaceable>... such that the CCL
       query <replaceable>q=x</replaceable> is equivalent to
       <replaceable>q1=x or q2=x or ...</replaceable>.
      </para>
     </sect4>

     <sect4 id="ccl.comments">
      <title>Comments</title>
      <para>
       Lines with white space or lines that begin with
       character <literal>#</literal> are treated as comments.
      </para>
     </sect4>

     <sect4 id="ccl.directives">
      <title>Directives</title>
      <para>
       Directive specifications takes the form
      </para>
      <para><literal>@</literal><replaceable>directive</replaceable> <replaceable>value</replaceable>
      </para>
      <table id="ccl.directives.table">
       <title>CCL directives</title>
       <tgroup cols="3">
        <colspec colwidth="2*" colname="name"></colspec>
        <colspec colwidth="8*" colname="description"></colspec>
        <colspec colwidth="1*" colname="default"></colspec>
        <thead>
         <row>
          <entry>Name</entry>
          <entry>Description</entry>
          <entry>Default</entry>
         </row>
        </thead>
        <tbody>
         <row>
          <entry>truncation</entry>
          <entry>Truncation character</entry>
          <entry><literal>?</literal></entry>
         </row>
         <row>
          <entry>field</entry>
          <entry>Specifies how multiple fields are to be
           combined. There are two modes: <literal>or</literal>:
           multiple qualifier fields are ORed,
           <literal>merge</literal>: attributes for the qualifier
           fields are merged and assigned to one term.
           </entry>
          <entry><literal>merge</literal></entry>
         </row>
         <row>
          <entry>case</entry>
          <entry>Specificies if CCL operatores and qualifiers should be
           compared with case sensitivity or not. Specify 0 for
           case sensitive; 1 for case insensitive.</entry>
          <entry><literal>0</literal></entry>
         </row>

         <row>
          <entry>and</entry>
          <entry>Specifies token for CCL operator AND.</entry>
          <entry><literal>and</literal></entry>
         </row>

         <row>
          <entry>or</entry>
          <entry>Specifies token for CCL operator OR.</entry>
          <entry><literal>or</literal></entry>
         </row>

         <row>
          <entry>not</entry>
          <entry>Specifies token for CCL operator NOT.</entry>
          <entry><literal>not</literal></entry>
         </row>

         <row>
          <entry>set</entry>
          <entry>Specifies token for CCL operator SET.</entry>
          <entry><literal>set</literal></entry>
         </row>
        </tbody>
        </tgroup>
      </table>
     </sect4>
    </sect3>
    <sect3 id="ccl.api">
     <title>CCL API</title>
     <para>
      All public definitions can be found in the header file
      <filename>ccl.h</filename>. A profile identifier is of type
      <literal>CCL_bibset</literal>. A profile must be created with the call
      to the function <function>ccl_qual_mk</function> which returns a profile
      handle of type <literal>CCL_bibset</literal>.
     </para>

     <para>
      To read a file containing qualifier definitions the function
      <function>ccl_qual_file</function> may be convenient. This function
      takes an already opened <literal>FILE</literal> handle pointer as
      argument along with a <literal>CCL_bibset</literal> handle.
     </para>

     <para>
      To parse a simple string with a FIND query use the function
     </para>
     <screen>
struct ccl_rpn_node *ccl_find_str (CCL_bibset bibset, const char *str,
                                   int *error, int *pos);
     </screen>
     <para>
      which takes the CCL profile (<literal>bibset</literal>) and query
      (<literal>str</literal>) as input. Upon successful completion the RPN
      tree is returned. If an error occur, such as a syntax error, the integer
      pointed to by <literal>error</literal> holds the error code and
      <literal>pos</literal> holds the offset inside query string in which
      the parsing failed.
     </para>

     <para>
      An English representation of the error may be obtained by calling
      the <literal>ccl_err_msg</literal> function. The error codes are
      listed in <filename>ccl.h</filename>.
     </para>

     <para>
      To convert the CCL RPN tree (type
      <literal>struct ccl_rpn_node *</literal>)
      to the Z_RPNQuery of YAZ the function <function>ccl_rpn_query</function>
      must be used. This function which is part of YAZ is implemented in
      <filename>yaz-ccl.c</filename>.
      After calling this function the CCL RPN tree is probably no longer
      needed. The <literal>ccl_rpn_delete</literal> destroys the CCL RPN tree.
     </para>

     <para>
      A CCL profile may be destroyed by calling the
      <function>ccl_qual_rm</function> function.
     </para>

     <para>
      The token names for the CCL operators may be changed by setting the
      globals (all type <literal>char *</literal>)
      <literal>ccl_token_and</literal>, <literal>ccl_token_or</literal>,
      <literal>ccl_token_not</literal> and <literal>ccl_token_set</literal>.
      An operator may have aliases, i.e. there may be more than one name for
      the operator. To do this, separate each alias with a space character.
     </para>
    </sect3>
   </sect2>
   <sect2 id="cql"><title>CQL</title>
    <para>
     <ulink url="&url.cql;">CQL</ulink>
      - Common Query Language - was defined for the
     <ulink url="&url.sru;">SRU</ulink> protocol.
     In many ways CQL has a similar syntax to CCL.
     The objective of CQL is different. Where CCL aims to be
     an end-user language, CQL is <emphasis>the</emphasis> protocol
     query language for SRU.
    </para>
    <tip>
     <para>
      If you are new to CQL, read the 
      <ulink url="&url.cql.intro;">Gentle Introduction</ulink>.
     </para>
    </tip>
    <para>
     The CQL parser in &yaz; provides the following:
     <itemizedlist>
      <listitem>
       <para>
        It parses and validates a CQL query.
       </para>
      </listitem>
      <listitem>
       <para>
        It generates a C structure that allows you to convert
        a CQL query to some other query language, such as SQL.
       </para>
      </listitem>
      <listitem>
       <para>
        The parser converts a valid CQL query to PQF, thus providing a
        way to use CQL for both SRU servers and Z39.50 targets at the
        same time.
       </para>
      </listitem>
      <listitem>
       <para>
        The parser converts CQL to
        <ulink url="&url.xcql;">XCQL</ulink>.
        XCQL is an XML representation of CQL.
        XCQL is part of the SRU specification. However, since SRU
        supports CQL only, we don't expect XCQL to be widely used.
        Furthermore, CQL has the advantage over XCQL that it is
        easy to read.
       </para>
      </listitem>
     </itemizedlist>
    </para>
    <sect3 id="cql.parsing"><title>CQL parsing</title>
     <para>
      A CQL parser is represented by the <literal>CQL_parser</literal>
      handle. Its contents should be considered &yaz; internal (private).
      <synopsis>
#include &lt;yaz/cql.h&gt;

typedef struct cql_parser *CQL_parser;

CQL_parser cql_parser_create(void);
void cql_parser_destroy(CQL_parser cp);
      </synopsis>
     A parser is created by <function>cql_parser_create</function> and
     is destroyed by <function>cql_parser_destroy</function>.
     </para>
     <para>
      To parse a CQL query string, the following function
      is provided:
      <synopsis>
int cql_parser_string(CQL_parser cp, const char *str);
      </synopsis>
      A CQL query is parsed by the <function>cql_parser_string</function>
      which takes a query <parameter>str</parameter>.
      If the query was valid (no syntax errors), then zero is returned;
      otherwise -1 is returned to indicate a syntax error.
     </para>
     <para>
      <synopsis>
int cql_parser_stream(CQL_parser cp,
                      int (*getbyte)(void *client_data),
                      void (*ungetbyte)(int b, void *client_data),
                      void *client_data);

int cql_parser_stdio(CQL_parser cp, FILE *f);
      </synopsis>
      The functions <function>cql_parser_stream</function> and
      <function>cql_parser_stdio</function> parses a CQL query
      - just like <function>cql_parser_string</function>.
      The only difference is that the CQL query can be
      fed to the parser in different ways.
      The <function>cql_parser_stream</function> uses a generic
      byte stream as input. The <function>cql_parser_stdio</function>
      uses a <literal>FILE</literal> handle which is opened for reading.
     </para>
    </sect3>
    
    <sect3 id="cql.tree"><title>CQL tree</title>
     <para>
      The the query string is valid, the CQL parser
      generates a tree representing the structure of the
      CQL query.
     </para>
     <para>
      <synopsis>
struct cql_node *cql_parser_result(CQL_parser cp);
      </synopsis>
      <function>cql_parser_result</function> returns the
      a pointer to the root node of the resulting tree.
     </para>
     <para>
      Each node in a CQL tree is represented by a 
      <literal>struct cql_node</literal>.
      It is defined as follows:
      <synopsis>
#define CQL_NODE_ST 1
#define CQL_NODE_BOOL 2
struct cql_node {
    int which;
    union {
        struct {
            char *index;
            char *index_uri;
            char *term;
            char *relation;
            char *relation_uri;
            struct cql_node *modifiers;
        } st;
        struct {
            char *value;
            struct cql_node *left;
            struct cql_node *right;
            struct cql_node *modifiers;
        } boolean;
    } u;
};
      </synopsis>
      There are two node types: search term (ST) and boolean (BOOL).
      A modifier is treated as a search term too.
     </para>
     <para>
      The search term node has five members:
      <itemizedlist>
       <listitem>
        <para>
         <literal>index</literal>: index for search term.
         If an index is unspecified for a search term,
         <literal>index</literal> will be NULL.
        </para>
       </listitem>
       <listitem>
        <para>
         <literal>index_uri</literal>: index URi for search term
         or NULL if none could be resolved for the index.
        </para>
       </listitem>
       <listitem>
        <para>
         <literal>term</literal>: the search term itself.
        </para>
       </listitem>
       <listitem>
        <para>
         <literal>relation</literal>: relation for search term.
        </para>
       </listitem>
       <listitem>
        <para>
         <literal>relation_uri</literal>: relation URI for search term.
        </para>
       </listitem>
       <listitem>
        <para>
         <literal>modifiers</literal>: relation modifiers for search
         term. The <literal>modifiers</literal> list itself of cql_nodes
         each of type <literal>ST</literal>.
        </para>
       </listitem>
      </itemizedlist>
     </para>

     <para>
      The boolean node represents both <literal>and</literal>,
      <literal>or</literal>, not as well as
      proximity.
      <itemizedlist>
       <listitem>
        <para>
         <literal>left</literal> and <literal>right</literal>: left
         - and right operand respectively.
        </para>
       </listitem>
       <listitem>
        <para>
         <literal>modifiers</literal>: proximity arguments.
        </para>
       </listitem>
      </itemizedlist>
     </para>

    </sect3>
    <sect3 id="cql.to.pqf"><title>CQL to PQF conversion</title>
     <para>
      Conversion to PQF (and Z39.50 RPN) is tricky by the fact
      that the resulting RPN depends on the Z39.50 target
      capabilities (combinations of supported attributes). 
      In addition, the CQL and SRU operates on index prefixes
      (URI or strings), whereas the RPN uses Object Identifiers
      for attribute sets.
     </para>
     <para>
      The CQL library of &yaz; defines a <literal>cql_transform_t</literal>
      type. It represents a particular mapping between CQL and RPN.
      This handle is created and destroyed by the functions:
     <synopsis>
cql_transform_t cql_transform_open_FILE (FILE *f);
cql_transform_t cql_transform_open_fname(const char *fname);
void cql_transform_close(cql_transform_t ct);
      </synopsis>
      The first two functions create a tranformation handle from
      either an already open FILE or from a filename respectively.
     </para>
     <para>
      The handle is destroyed by <function>cql_transform_close</function> 
      in which case no further reference of the handle is allowed.
     </para>
     <para>
      When a <literal>cql_transform_t</literal> handle has been created
      you can convert to RPN.
      <synopsis>
int cql_transform_buf(cql_transform_t ct,
                      struct cql_node *cn, char *out, int max);
      </synopsis>
      This function converts the CQL tree <literal>cn</literal> 
      using handle <literal>ct</literal>.
      For the resulting PQF, you supply a buffer <literal>out</literal>
      which must be able to hold at at least <literal>max</literal>
      characters.
     </para>
     <para>
      If conversion failed, <function>cql_transform_buf</function>
      returns a non-zero SRU error code; otherwise zero is returned
      (conversion successful).  The meanings of the numeric error
      codes are listed in the SRU specifications at
      <ulink url="&url.sru.diagnostics.list;"/>
     </para>
     <para>
      If conversion fails, more information can be obtained by calling
      <synopsis>
int cql_transform_error(cql_transform_t ct, char **addinfop);
      </synopsis>
      This function returns the most recently returned numeric
      error-code and sets the string-pointer at
      <literal>*addinfop</literal> to point to a string containing
      additional information about the error that occurred: for
      example, if the error code is 15 (``Illegal or unsupported context
      set''), the additional information is the name of the requested
      context set that was not recognised.
     </para>
     <para>
      The SRU error-codes may be translated into brief human-readable
      error messages using
      <synopsis>
const char *cql_strerror(int code);
      </synopsis>
     </para>
     <para>
      If you wish to be able to produce a PQF result in a different
      way, there are two alternatives.
      <synopsis>
void cql_transform_pr(cql_transform_t ct,
                      struct cql_node *cn,
                      void (*pr)(const char *buf, void *client_data),
                      void *client_data);

int cql_transform_FILE(cql_transform_t ct,
                       struct cql_node *cn, FILE *f);
      </synopsis>
      The former function produces output to a user-defined
      output stream. The latter writes the result to an already
      open <literal>FILE</literal>.
     </para>
    </sect3>
    <sect3 id="cql.to.rpn">
     <title>Specification of CQL to RPN mappings</title>
     <para>
      The file supplied to functions 
      <function>cql_transform_open_FILE</function>,
      <function>cql_transform_open_fname</function> follows
      a structure found in many Unix utilities.
      It consists of mapping specifications - one per line.
      Lines starting with <literal>#</literal> are ignored (comments).
     </para>
     <para>
      Each line is of the form
      <literallayout>
       <replaceable>CQL pattern</replaceable><literal> = </literal> <replaceable> RPN equivalent</replaceable>
      </literallayout>
     </para>
     <para>
      An RPN pattern is a simple attribute list. Each attribute pair
      takes the form:
      <literallayout>
       [<replaceable>set</replaceable>] <replaceable>type</replaceable><literal>=</literal><replaceable>value</replaceable>
      </literallayout>
      The attribute <replaceable>set</replaceable> is optional.
      The <replaceable>type</replaceable> is the attribute type,
      <replaceable>value</replaceable> the attribute value.
     </para>
     <para>
      The character <literal>*</literal> (asterisk) has special meaning
      when used in the RPN pattern.
      Each occurrence of <literal>*</literal> is substituted with the
      CQL matching name (index, relation, qualifier etc).
      This facility can be used to copy a CQL name verbatim to the RPN result.
     </para>
     <para>
      The following CQL patterns are recognized:
      <variablelist>
       <varlistentry><term>
         <literal>index.</literal><replaceable>set</replaceable><literal>.</literal><replaceable>name</replaceable>
        </term>
        <listitem>
         <para>
          This pattern is invoked when a CQL index, such as 
          dc.title is converted. <replaceable>set</replaceable>
          and <replaceable>name</replaceable> are the context set and index
          name respectively.
          Typically, the RPN specifies an equivalent use attribute.
         </para>
         <para>
          For terms not bound by an index the pattern
          <literal>index.cql.serverChoice</literal> is used.
          Here, the prefix <literal>cql</literal> is defined as
          <literal>http://www.loc.gov/zing/cql/cql-indexes/v1.0/</literal>.
          If this pattern is not defined, the mapping will fail.
         </para>
         <para>
          The pattern, 
          <literal>index.</literal><replaceable>set</replaceable><literal>.*</literal>
          is used when no other index pattern is matched.
        </para>
        </listitem>
       </varlistentry>
       <varlistentry><term>
         <literal>qualifier.</literal><replaceable>set</replaceable><literal>.</literal><replaceable>name</replaceable>
         (DEPRECATED)
        </term>
        <listitem>
         <para>
          For backwards compatibility, this is recognised as a synonym of
          <literal>index.</literal><replaceable>set</replaceable><literal>.</literal><replaceable>name</replaceable>
         </para>
        </listitem>
       </varlistentry>
       <varlistentry><term>
         <literal>relation.</literal><replaceable>relation</replaceable>
        </term>
        <listitem>
         <para>
          This pattern specifies how a CQL relation is mapped to RPN.
          <replaceable>pattern</replaceable> is name of relation
          operator. Since <literal>=</literal> is used as
          separator between CQL pattern and RPN, CQL relations
          including <literal>=</literal> cannot be
          used directly. To avoid a conflict, the names
          <literal>ge</literal>,
          <literal>eq</literal>,
          <literal>le</literal>,
          must be used for CQL operators, greater-than-or-equal,
          equal, less-than-or-equal respectively.
          The RPN pattern is supposed to include a relation attribute.
         </para>
         <para>
          For terms not bound by a relation, the pattern
          <literal>relation.scr</literal> is used. If the pattern
          is not defined, the mapping will fail.
         </para>
         <para>
          The special pattern, <literal>relation.*</literal> is used
          when no other relation pattern is matched.
         </para>
        </listitem>
       </varlistentry>

       <varlistentry><term>
         <literal>relationModifier.</literal><replaceable>mod</replaceable>
        </term>
        <listitem>
         <para>
          This pattern specifies how a CQL relation modifier is mapped to RPN.
          The RPN pattern is usually a relation attribute.
         </para>
        </listitem>
       </varlistentry>

       <varlistentry><term>
         <literal>structure.</literal><replaceable>type</replaceable>
        </term>
        <listitem>
         <para>
          This pattern specifies how a CQL structure is mapped to RPN.
          Note that this CQL pattern is somewhat to similar to
          CQL pattern <literal>relation</literal>. 
          The <replaceable>type</replaceable> is a CQL relation.
         </para>
         <para>
          The pattern, <literal>structure.*</literal> is used
          when no other structure pattern is matched.
          Usually, the RPN equivalent specifies a structure attribute.
         </para>
        </listitem>
       </varlistentry>

       <varlistentry><term>
         <literal>position.</literal><replaceable>type</replaceable>
        </term>
        <listitem>
         <para>
          This pattern specifies how the anchor (position) of
          CQL is mapped to RPN.
          The <replaceable>type</replaceable> is one
          of <literal>first</literal>, <literal>any</literal>,
          <literal>last</literal>, <literal>firstAndLast</literal>.
         </para>
         <para>
          The pattern, <literal>position.*</literal> is used
          when no other position pattern is matched.
         </para>
        </listitem>
       </varlistentry>

       <varlistentry><term>
         <literal>set.</literal><replaceable>prefix</replaceable>
        </term>
        <listitem>
         <para>
          This specification defines a CQL context set for a given prefix.
          The value on the right hand side is the URI for the set - 
          <emphasis>not</emphasis> RPN. All prefixes used in
          index patterns must be defined this way.
         </para>
        </listitem>
       </varlistentry>

       <varlistentry><term>
         <literal>set</literal>
        </term>
        <listitem>
         <para>
          This specification defines a default CQL context set for index names.
          The value on the right hand side is the URI for the set.
         </para>
        </listitem>
       </varlistentry>

      </variablelist>
     </para>
     <example id="example.cql.to.rpn.mapping">
      <title>CQL to RPN mapping file</title>
      <para>
       This simple file defines two context sets, three indexes and three
       relations, a position pattern and a default structure.
      </para>
      <programlisting><![CDATA[
       set.cql  = http://www.loc.gov/zing/cql/context-sets/cql/v1.1/
       set.dc   = http://www.loc.gov/zing/cql/dc-indexes/v1.0/

       index.cql.serverChoice = 1=1016
       index.dc.title         = 1=4
       index.dc.subject       = 1=21
  
       relation.<             = 2=1
       relation.eq            = 2=3
       relation.scr           = 2=3

       position.any           = 3=3 6=1

       structure.*            = 4=1
]]>
      </programlisting>
      <para>
       With the mappings above, the CQL query
       <screen>
        computer
       </screen>
       is converted to the PQF:
       <screen>
        @attr 1=1016 @attr 2=3 @attr 4=1 @attr 3=3 @attr 6=1 "computer"
       </screen>
       by rules <literal>index.cql.serverChoice</literal>,
       <literal>relation.scr</literal>, <literal>structure.*</literal>,
       <literal>position.any</literal>.
      </para>
      <para>
       CQL query
       <screen>
        computer^
       </screen>
       is rejected, since <literal>position.right</literal> is
       undefined.
      </para>
      <para>
       CQL query
       <screen>
        >my = "http://www.loc.gov/zing/cql/dc-indexes/v1.0/" my.title = x
       </screen>
       is converted to
       <screen>
        @attr 1=4 @attr 2=3 @attr 4=1 @attr 3=3 @attr 6=1 "x"
       </screen>
      </para>
     </example>
     <example id="example.cql.to.rpn.string">
      <title>CQL to RPN string attributes</title>
      <para>
       In this example we allow any index to be passed to RPN as
       a use attribute.
      </para>
      <programlisting><![CDATA[
       # Identifiers for prefixes used in this file. (index.*)
       set.cql  = info:srw/cql-context-set/1/cql-v1.1
       set.rpn  = http://bogus/rpn
       set      = http://bogus/rpn

       # The default index when none is specified by the query
       index.cql.serverChoice     = 1=any

       index.rpn.*                = 1=*
       relation.eq                = 2=3
       structure.*                = 4=1
       position.any               = 3=3
]]>
      </programlisting>
      <para>
       The <literal>http://bogus/rpn</literal> context set is also the default
       so we can make queries such as
       <screen>
        title = a
       </screen>
       which is converted to
       <screen>
        @attr 2=3 @attr 4=1 @attr 3=3 @attr 1=title "a"
       </screen>
      </para>
     </example>
     <example id="example.cql.to.rpn.bathprofile">
      <title>CQL to RPN using Bath Profile</title>
      <para>
       The file <filename>etc/pqf.properties</filename> has mappings from
       the Bath Profile and Dublin Core to RPN.
       If YAZ is installed as a package it's usually located
       in <filename>/usr/share/yaz/etc</filename> and part of the
       development package, such as <literal>libyaz-dev</literal>.
      </para>
     </example>
    </sect3>
    <sect3 id="cql.xcql"><title>CQL to XCQL conversion</title>
     <para>
      Conversion from CQL to XCQL is trivial and does not
      require a mapping to be defined.
      There three functions to choose from depending on the
      way you wish to store the resulting output (XML buffer
      containing XCQL).
      <synopsis>
int cql_to_xml_buf(struct cql_node *cn, char *out, int max);
void cql_to_xml(struct cql_node *cn, 
                void (*pr)(const char *buf, void *client_data),
                void *client_data);
void cql_to_xml_stdio(struct cql_node *cn, FILE *f);
      </synopsis>
      Function <function>cql_to_xml_buf</function> converts
      to XCQL and stores result in a user supplied buffer of a given
      max size.
     </para>
     <para>
      <function>cql_to_xml</function> writes the result in
      a user defined output stream.
      <function>cql_to_xml_stdio</function> writes to a
      a file.
     </para>
    </sect3>
   </sect2>
  </sect1>
  <sect1 id="tools.oid"><title>Object Identifiers</title>

   <para>
    The basic YAZ representation of an OID is an array of integers,
    terminated with the value -1. This integer is of type 
    <literal>Odr_oid</literal>.
   </para>
   <para>
    Fundamental OID operations and the type <literal>Odr_oid</literal>
    are defined in <filename>yaz/oid_util.h</filename>.
   </para>
   <para>
    An OID can either be declared as a automatic variable or it can
    allocated using the memory utilities or ODR/NMEM. It's
    guaranteed that an OID can fit in <literal>OID_SIZE</literal> integers.
   </para>
   <example id="tools.oid.bib1.1"><title>Create OID on stack</title>
    <para>
     We can create an OID for the Bib-1 attribute set with:
     <screen>
      Odr_oid bib1[OID_SIZE];
      bib1[0] = 1;
      bib1[1] = 2;
      bib1[2] = 840;
      bib1[3] = 10003;
      bib1[4] = 3;
      bib1[5] = 1;
      bib1[6] = -1;
     </screen>
    </para>
   </example>
   <para>
    And OID may also be filled from a string-based representation using
    dots (.). This is achieved by function
    <screen>
     int oid_dotstring_to_oid(const char *name, Odr_oid *oid);
    </screen>
    This functions returns 0 if name could be converted; -1 otherwise.
   </para>
   <example id="tools.oid.bib1.2"><title>Using oid_oiddotstring_to_oid</title>
    <para>
     We can fill the Bib-1 attribute set OID easier with:
     <screen>
      Odr_oid bib1[OID_SIZE];
      oid_oiddotstring_to_oid("1.2.840.10003.3.1", bib1);
     </screen>
   </para>
   </example>
   <para>
    We can also allocate an OID dynamically on a ODR stream with:
   <screen>
    Odr_oid *odr_getoidbystr(ODR o, const char *str);
   </screen>
    This creates an OID from string-based representation using dots.
    This function take an &odr; stream as parameter. This stream is used to
    allocate memory for the data elements, which is released on a
    subsequent call to <function>odr_reset()</function> on that stream.
   </para>

   <example id="tools.oid.bib1.3"><title>Using odr_getoidbystr</title>
    <para>
     We can create a OID for the Bib-1 attribute set with:
     <screen>
      Odr_oid *bib1 = odr_getoidbystr(odr, "1.2.840.10003.3.1");
     </screen>
    </para>
   </example>

   <para>
    The function
    <screen>
     char *oid_oid_to_dotstring(const Odr_oid *oid, char *oidbuf)
    </screen>
    does the reverse of <function>oid_oiddotstring_to_oid</function>. It
    converts an OID to the string-based representation using dots.
    The supplied char buffer <literal>oidbuf</literal> holds the resulting
    string and must be at least <literal>OID_STR_MAX</literal> in size.
   </para>

   <para>
    OIDs can be copied with <function>oid_oidcpy</function> which takes
    two OID lists as arguments. Alternativly, an OID copy can be allocated
    on a ODR stream with:
    <screen>
     Odr_oid *odr_oiddup(ODR odr, const Odr_oid *o);
    </screen>
   </para>
   
   <para>
    OIDs can be compared with <function>oid_oidcmp</function> which returns
    zero if the two OIDs provided are identical; non-zero otherwise.
   </para>
 
   <sect2 id="tools.oid.database"><title>OID database</title>
    <para>
     From YAZ version 3 and later, the oident system has been replaced
     by an OID database. OID database is a misnomer .. the old odient
     system was also a database.
    </para>
    <para>
     The OID database is really just a map between named Object Identifiers
     (string) and their OID raw equivalents. Most operations either
     convert from string to OID or other way around.
    </para>
    <para>
     Unfortunately, whenever we supply a string we must also specify the 
     <emphasis>OID class</emphasis>. The class is necessary because some
     strings correspond to multiple OIDs. An example of such a string is
     <literal>Bib-1</literal> which may either be an attribute-set 
     or a diagnostic-set.
    </para>
    <para>
     Applications using the YAZ database should include 
     <filename>yaz/oid_db.h</filename>.
    </para>
    <para>
     A YAZ database handle is of type <literal>yaz_oid_db_t</literal>.
     Actually that's a pointer. You need not think deal with that.
     YAZ has a built-in database which can be considered "constant" for
     most purposes. 
     We can get hold that by using function <function>yaz_oid_std</function>.
    </para>
    <para>
     All functions with prefix <function>yaz_string_to_oid</function>
     converts from class + string to OID. We have variants of this
     operation due to different memory allocation strategies.
    </para>
    <para>
     All functions with prefix
     <function>yaz_oid_to_string</function> converts from OID to string
     + class.
    </para>

    <example id="tools.oid.bib1.4"><title>Create OID with YAZ DB</title>
     <para>
      We can create an OID for the Bib-1 attribute set on the ODR stream
      odr with:
     <screen>
        Odr_oid *bib1 = 
         yaz_string_to_oid_odr(yaz_oid_std(), CLASS_ATTSET, "Bib-1", odr);
      </screen>
      This is more complex than using <function>odr_getoidbystr</function>.
      You would only use <function>yaz_string_to_oid_odr</function> when the
      string (here Bib-1) is supplied by a user or configuration.
     </para>
    </example>

   </sect2>
   <sect2 id="tools.oid.std"><title>Standard OIDs</title>
     
    <para>
     All the object identifers in the standard OID database as returned
     by <function>yaz_oid_std</function> can referenced directly in a
     program as a constant OID.
     Each constant OID is prefixed with <literal>yaz_oid_</literal> -
     followed by OID class (lowercase) - then by OID name (normalized and
     lowercase).
    </para>
    <para>
     See <xref linkend="list-oids"/> for list of all object identifiers
     built into YAZ.
     These are declared in <filename>yaz/oid_std.h</filename> but are
     included by <filename>yaz/oid_db.h</filename> as well.
    </para>

    <example id="tools.oid.bib1.5"><title>Use a built-in OID</title>
     <para>
      We can allocate our own OID filled with the constant OID for
      Bib-1 with:
      <screen>
        Odr_oid *bib1 = odr_oiddup(o, yaz_oid_attset_bib1);
      </screen>
     </para>
    </example>
   </sect2>
  </sect1>
  <sect1 id="tools.nmem"><title>Nibble Memory</title>

   <para>
    Sometimes when you need to allocate and construct a large,
    interconnected complex of structures, it can be a bit of a pain to
    release the associated memory again. For the structures describing the
    Z39.50 PDUs and related structures, it is convenient to use the
    memory-management system of the &odr; subsystem (see
    <xref linkend="odr.use"/>). However, in some circumstances
    where you might otherwise benefit from using a simple nibble memory
    management system, it may be impractical to use
    <function>odr_malloc()</function> and <function>odr_reset()</function>.
    For this purpose, the memory manager which also supports the &odr;
    streams is made available in the NMEM module. The external interface
    to this module is given in the <filename>nmem.h</filename> file.
   </para>

   <para>
    The following prototypes are given:
   </para>

   <screen>
    NMEM nmem_create(void);
    void nmem_destroy(NMEM n);
    void *nmem_malloc(NMEM n, size_t size);
    void nmem_reset(NMEM n);
    size_t nmem_total(NMEM n);
    void nmem_init(void);
    void nmem_exit(void);
   </screen>

   <para>
    The <function>nmem_create()</function> function returns a pointer to a
    memory control handle, which can be released again by
    <function>nmem_destroy()</function> when no longer needed.
    The function <function>nmem_malloc()</function> allocates a block of
    memory of the requested size. A call to <function>nmem_reset()</function>
    or <function>nmem_destroy()</function> will release all memory allocated
    on the handle since it was created (or since the last call to
    <function>nmem_reset()</function>. The function
    <function>nmem_total()</function> returns the number of bytes currently
    allocated on the handle.
   </para>

   <para>
    The nibble memory pool is shared amongst threads. POSIX
    mutex'es and WIN32 Critical sections are introduced to keep the
    module thread safe. Function <function>nmem_init()</function>
    initializes the nibble memory library and it is called automatically
    the first time the <literal>YAZ.DLL</literal> is loaded. &yaz; uses
    function <function>DllMain</function> to achieve this. You should
    <emphasis>not</emphasis> call <function>nmem_init</function> or
    <function>nmem_exit</function> unless you're absolute sure what
    you're doing. Note that in previous &yaz; versions you'd have to call
    <function>nmem_init</function> yourself. 
   </para>

  </sect1>

  <sect1 id="tools.log"><title>Log</title>
  <para>
   &yaz; has evolved a fairly complex log system which should be useful both 
   for debugging &yaz; itself, debugging applications that use &yaz;, and for
   production use of those applications.  
  </para>
  <para>
   The log functions are declared in header <filename>yaz/log.h</filename>
    and implemented in <filename>src/log.c</filename>.
    Due to name clash with syslog and some math utilities the logging
    interface has been modified as of YAZ 2.0.29. The obsolete interface
    is still available if in header file <filename>yaz/log.h</filename>.
    The key points of the interface are:
  </para>
  <screen>
   void yaz_log(int level, const char *fmt, ...)

   void yaz_log_init(int level, const char *prefix, const char *name);
   void yaz_log_init_file(const char *fname);
   void yaz_log_init_level(int level);
   void yaz_log_init_prefix(const char *prefix);
   void yaz_log_time_format(const char *fmt);
   void yaz_log_init_max_size(int mx);

   int yaz_log_mask_str(const char *str);
   int yaz_log_module_level(const char *name);
  </screen>

  <para>
   The reason for the whole log module is the <function>yaz_log</function>
   function. It takes a bitmask indicating the log levels, a
   <literal>printf</literal>-like format string, and a variable number of
   arguments to log.
  </para>

  <para>
   The <literal>log level</literal> is a bit mask, that says on which level(s)
   the log entry should be made, and optionally set some behaviour of the
   logging. In the most simple cases, it can be one of <literal>YLOG_FATAL,
   YLOG_DEBUG, YLOG_WARN, YLOG_LOG</literal>. Those can be combined with bits
   that modify the way the log entry is written:<literal>YLOG_ERRNO,
   YLOG_NOTIME, YLOG_FLUSH</literal>.
   Most of the rest of the bits are deprecated, and should not be used. Use
   the dynamic log levels instead.
  </para>

  <para>
   Applications that use &yaz;, should not use the LOG_LOG for ordinary
   messages, but should make use of the dynamic loglevel system. This consists
   of two parts, defining the loglevel and checking it.
  </para>

  <para>
   To define the log levels, the (main) program should pass a string to
   <function>yaz_log_mask_str</function> to define which log levels are to be
   logged. This string should be a comma-separated list of log level names,
   and can contain both hard-coded names and dynamic ones. The log level
   calculation starts with <literal>YLOG_DEFAULT_LEVEL</literal> and adds a bit
   for each word it meets, unless the word starts with a '-', in which case it 
   clears the bit. If the string <literal>'none'</literal> is found,
   all bits are cleared. Typically this string comes from the command-line,
   often identified by <literal>-v</literal>. The
   <function>yaz_log_mask_str</function> returns a log level that should be
   passed to <function>yaz_log_init_level</function> for it to take effect.
  </para>

  <para>
   Each module should check what log bits it should be used, by calling 
   <function>yaz_log_module_level</function> with a suitable name for the
   module. The name is cleared from a preceding path and an extension, if any,
   so it is quite possible to use <literal>__FILE__</literal> for it. If the
   name has been passed to <function>yaz_log_mask_str</function>, the routine
   returns a non-zero bitmask, which should then be used in consequent calls
   to yaz_log. (It can also be tested, so as to avoid unnecessary calls to
   yaz_log, in time-critical places, or when the log entry would take time 
   to construct.) 
  </para>

  <para>
   Yaz uses the following dynamic log levels:
   <literal>server, session, request, requestdetail</literal> for the server
   functionality.
   <literal>zoom</literal> for the zoom client api.
   <literal>ztest</literal> for the simple test server.
   <literal>malloc, nmem, odr, eventl</literal> for internal debugging of yaz itself.
   Of course, any program using yaz is welcome to define as many new ones, as
   it needs.
  </para>

  <para>
   By default the log is written to stderr, but this can be changed by a call
   to <function>yaz_log_init_file</function> or
   <function>yaz_log_init</function>. If the log is directed to a file, the
   file size is checked at every write, and if it exceeds the limit given in
   <function>yaz_log_init_max_size</function>, the log is rotated. The
   rotation keeps one old version (with a <literal>.1</literal> appended to
   the name). The size defaults to 1GB. Setting it to zero will disable the
   rotation feature.
  </para>

  <screen>
  A typical yaz-log looks like this
  13:23:14-23/11 yaz-ztest(1) [session] Starting session from tcp:127.0.0.1 (pid=30968)
  13:23:14-23/11 yaz-ztest(1) [request] Init from 'YAZ' (81) (ver 2.0.28) OK
  13:23:17-23/11 yaz-ztest(1) [request] Search Z: @attrset Bib-1 foo  OK:7 hits
  13:23:22-23/11 yaz-ztest(1) [request] Present: [1] 2+2  OK 2 records returned
  13:24:13-23/11 yaz-ztest(1) [request] Close OK
  </screen>

  <para>
   The log entries start with a time stamp. This can be omitted by setting the
   <literal>YLOG_NOTIME</literal> bit in the loglevel. This way automatic tests
   can be hoped to produce identical log files, that are easy to diff. The
   format of the time stamp can be set with
   <function>yaz_log_time_format</function>, which takes a format string just
   like <function>strftime</function>.
  </para>

  <para>
   Next in a log line comes the prefix, often the name of the program. For
   yaz-based servers, it can also contain the session number. Then
   comes one or more logbits in square brackets, depending on the logging
   level set by <function>yaz_log_init_level</function> and the loglevel
   passed to <function>yaz_log_init_level</function>. Finally comes the format
   string and additional values passed to <function>yaz_log</function>
  </para>

  <para>
   The log level <literal>YLOG_LOGLVL</literal>, enabled by the string
   <literal>loglevel</literal>, will log all the log-level affecting
   operations. This can come in handy if you need to know what other log
   levels would be useful. Grep the logfile for <literal>[loglevel]</literal>.
  </para>

  <para>
   The log system is almost independent of the rest of &yaz;, the only
   important dependence is of <filename>nmem</filename>, and that only for
   using the semaphore definition there. 
  </para>

  <para>
   The dynamic log levels and log rotation were introduced in &yaz; 2.0.28. At
   the same time, the log bit names were changed from
   <literal>LOG_something</literal> to <literal>YLOG_something</literal>, 
   to avoid collision with <filename>syslog.h</filename>.
  </para>

  </sect1>
  
  <sect1 id="marc"><title>MARC</title>
   
   <para>
    YAZ provides a fast utility for working with MARC records.
    Early versions of the MARC utility only allowed decoding of ISO2709.
    Today the utility may both encode - and decode to a varity of formats.
   </para>
   <synopsis><![CDATA[
    #include <yaz/marcdisp.h>

    /* create handler */
    yaz_marc_t yaz_marc_create(void);
    /* destroy */
    void yaz_marc_destroy(yaz_marc_t mt);

    /* set XML mode YAZ_MARC_LINE, YAZ_MARC_SIMPLEXML, ... */
    void yaz_marc_xml(yaz_marc_t mt, int xmlmode);
    #define YAZ_MARC_LINE      0
    #define YAZ_MARC_SIMPLEXML 1
    #define YAZ_MARC_OAIMARC   2
    #define YAZ_MARC_MARCXML   3
    #define YAZ_MARC_ISO2709   4
    #define YAZ_MARC_XCHANGE   5
    #define YAZ_MARC_CHECK     6
    #define YAZ_MARC_TURBOMARC 7

    /* supply iconv handle for character set conversion .. */
    void yaz_marc_iconv(yaz_marc_t mt, yaz_iconv_t cd);

    /* set debug level, 0=none, 1=more, 2=even more, .. */
    void yaz_marc_debug(yaz_marc_t mt, int level);

    /* decode MARC in buf of size bsize. Returns >0 on success; <=0 on failure.
    On success, result in *result with size *rsize. */
    int yaz_marc_decode_buf(yaz_marc_t mt, const char *buf, int bsize,
                            const char **result, size_t *rsize);

    /* decode MARC in buf of size bsize. Returns >0 on success; <=0 on failure.
       On success, result in WRBUF */
    int yaz_marc_decode_wrbuf(yaz_marc_t mt, const char *buf,
                              int bsize, WRBUF wrbuf);
]]>
   </synopsis>
   <note>
    <para>
     The synopsis is just a basic subset of all functionality. Refer
     to the actual header file <filename>marcdisp.h</filename> for
     details.
    </para>
   </note>
   <para>
    A MARC conversion handle must be created by using
    <function>yaz_marc_create</function> and destroyed
    by calling <function>yaz_marc_destroy</function>.
  </para>
   <para>
    All other function operate on a <literal>yaz_marc_t</literal> handle.
    The output is specified by a call to <function>yaz_marc_xml</function>.
    The <literal>xmlmode</literal> must be one of
    <variablelist>
     <varlistentry>
      <term>YAZ_MARC_LINE</term>
      <listitem>
       <para>
        A simple line-by-line format suitable for display but not
        recommend for further (machine) processing.
       </para>
      </listitem>
     </varlistentry>

     <varlistentry>
      <term>YAZ_MARC_MARCXML</term>
      <listitem>
       <para>
        <ulink url="&url.marcxml;">MARCXML</ulink>.
       </para>
      </listitem>
     </varlistentry>

     <varlistentry>
      <term>YAZ_MARC_ISO2709</term>
      <listitem>
       <para>
        ISO2709 (sometimes just referred to as "MARC").
       </para>
      </listitem>
     </varlistentry>

     <varlistentry>
      <term>YAZ_MARC_XCHANGE</term>
      <listitem>
       <para>
        <ulink url="&url.marcxchange;">MarcXchange</ulink>.
       </para>
      </listitem>
     </varlistentry>

     <varlistentry>
      <term>YAZ_MARC_CHECK</term>
      <listitem>
       <para>
        Pseudo format for validation only. Does not generate
        any real output except diagnostics.
       </para>
      </listitem>
     </varlistentry>

     <varlistentry>
      <term>YAZ_MARC_TURBOMARC</term>
      <listitem>
       <para>
        XML format with same semantics as MARCXML but more compact
        and geared towards fast processing with XSLT. Refer to
        <xref linkend="tools.turbomarc"/> for more information.
       </para>
      </listitem>
     </varlistentry>

    </variablelist>
   </para>
   <para>
    The actual conversion functions are 
    <function>yaz_marc_decode_buf</function> and
    <function>yaz_marc_decode_wrbuf</function> which decodes and encodes
    a MARC record. The former function operates on simple buffers, the
    stores the resulting record in a WRBUF handle (WRBUF is a simple string
    type).
   </para>
   <example id="example.marc.display">
    <title>Display of MARC record</title>
    <para>
     The following program snippet illustrates how the MARC API may
     be used to convert a MARC record to the line-by-line format:
     <programlisting><![CDATA[
      void print_marc(const char *marc_buf, int marc_buf_size)
      {
         char *result;      /* for result buf */
         size_t result_len;    /* for size of result */
         yaz_marc_t mt = yaz_marc_create();
         yaz_marc_xml(mt, YAZ_MARC_LINE);
         yaz_marc_decode_buf(mt, marc_buf, marc_buf_size,
                             &result, &result_len);
         fwrite(result, result_len, 1, stdout);
         yaz_marc_destroy(mt);  /* note that result is now freed... */
      }
]]>
      </programlisting>
    </para>
   </example>
   <sect2 id="tools.turbomarc">
    <title>TurboMARC</title>
    <para>
     TurboMARC is yet another XML encoding of a MARC record. The format
     was designed for fast processing with XSLT.
    </para>
    <para>
     Applications like
     Pazpar2 uses XSLT to convert an XML encoded MARC record to an internal
     representation. This conversion mostly check the tag of a MARC field
     to determine the basic rules in the conversion. This check is
     costly when that is tag is encoded as an attribute in MARCXML.
     By having the tag value as the element instead, makes processing
     many times faster (at least for Libxslt).
    </para>
    <para>
     TurboMARC is encoded as follows:
     <itemizedlist>
      <listitem><para>
        Record elements is part of namespace
        "<literal>http://www.indexdata.com/turbomarc</literal>".
       </para></listitem>
      <listitem><para>
        A record is enclosed in element <literal>r</literal>.
       </para></listitem>
      <listitem><para>
        A collection of records is enclosed in element
        <literal>collection</literal>.
       </para></listitem>
      <listitem><para>
        The leader is encoded as element <literal>l</literal> with the 
        leader content as its (text) value.
       </para></listitem>
      <listitem><para>
        A control field is encoded as element <literal>c</literal> concatenated
        with the tag value of the control field if the tag value
        matches the regular expression <literal>[a-zA-Z0-9]*</literal>.
        If the tag value do not match the regular expression
        <literal>[a-zA-Z0-9]*</literal> the control field is encoded
        as element <literal>c</literal> and attribute <literal>code</literal>
        will hold the tag value.
        This rule ensure that in the rare cases where a tag value might
        result in a non-wellformed XML YAZ encode it as a coded attribute
        (as in MARCXML).
       </para>
       <para>
        The control field content is the the text value of this element.
        Indicators are encoded as attribute names
        <literal>i1</literal>, <literal>i2</literal>, etc.. and
        corresponding values for each indicator.
       </para></listitem>
      <listitem><para>
        A data field is encoded as element <literal>d</literal> concatenated
        with the tag value of the data field or using the attribute
        <literal>code</literal> as described in the rules for control fields.
        The children of the data field element is subfield elements.
        Each subfield element is encoded as <literal>s</literal>
        concatenated with the sub field code.
        The text of the subfield element is the contents of the subfield.
        Indicators are encoded as attributes for the data field element similar
        to the encoding for control fields.
       </para></listitem>
     </itemizedlist>
    </para>
   </sect2>
  </sect1>

  <sect1 id="tools.retrieval">
   <title>Retrieval Facility</title>
   <para>
    YAZ version 2.1.20 or later includes a Retrieval facility tool
    which allows a SRU/Z39.50 to describe itself and perform record
    conversions. The idea is the following:
    
    <itemizedlist>
     <listitem>
      <para>
       An SRU/Z39.50 client sends a retrieval request which includes
       a combination of the following parameters: syntax (format),
       schema (or element set name).
      </para>
     </listitem>

     <listitem>
      <para>
       The retrieval facility is invoked with parameters in a
       server/proxy. The retrieval facility matches the parameters a set of
       "supported" retrieval types.
       If there is no match, the retrieval signals an error
       (syntax and / or schema not supported).
      </para>
     </listitem>

     <listitem>
      <para>
       For a successful match, the backend is invoked with the same
       or altered retrieval parameters (syntax, schema). If
       a record is received from the backend, it is converted to the
       frontend name / syntax.
      </para>
     </listitem>

     <listitem>
      <para>
       The resulting record is sent back the client and tagged with
       the frontend syntax / schema.
      </para>
     </listitem>

    </itemizedlist>
   </para>
   <para>
    The Retrieval facility is driven by an XML configuration. The
    configuration is neither Z39.50 ZeeRex or SRU ZeeRex. But it
    should be easy to generate both of them from the XML configuration.
    (unfortunately the two versions
    of ZeeRex differ substantially in this regard).
   </para>
   <sect2 id="tools.retrieval.format">
    <title>Retrieval XML format</title>
    <para>
     All elements should be covered by namespace 
     <literal>http://indexdata.com/yaz</literal> .
     The root element node must be <literal>retrievalinfo</literal>.
    </para>
    <para>
     The <literal>retrievalinfo</literal> must include one or
     more <literal>retrieval</literal> elements. Each 
    <literal>retrieval</literal> defines specific combination of
     syntax, name and identifier supported by this retrieval service.
    </para>
    <para>
     The <literal>retrieval</literal> element may include any of the
     following attributes:
     <variablelist>
      <varlistentry><term><literal>syntax</literal> (REQUIRED)</term>
       <listitem>
        <para>
         Defines the record syntax. Possible values is any
         of the names defined in YAZ' OID database or a raw
         OID in (n.n ... n).
        </para>
       </listitem>
      </varlistentry>
      <varlistentry><term><literal>name</literal> (OPTIONAL)</term>
       <listitem>
        <para>
         Defines the name of the retrieval format. This can be
         any string. For SRU, the value, is equivalent to schema (short-hand);
         for Z39.50 it's equivalent to simple element set name.
         For YAZ 3.0.24 and later this name may be specified as a glob
         expression with operators
         <literal>*</literal> and <literal>?</literal>.
        </para>
       </listitem>
      </varlistentry>
      <varlistentry><term><literal>identifier</literal> (OPTIONAL)</term>
       <listitem>
        <para>
         Defines the URI schema name of the retrieval format. This can be
         any string. For SRU, the value, is equivalent to URI schema.
         For Z39.50, there is no equivalent.
        </para>
       </listitem>
      </varlistentry>
     </variablelist>
    </para>
    <para>
     The <literal>retrieval</literal> may include one 
     <literal>backend</literal> element. If a <literal>backend</literal>
     element is given, it specifies how the records are retrieved by
     some backend and how the records are converted from the backend to
     the "frontend".
    </para>
    <para>
     The attributes, <literal>name</literal> and <literal>syntax</literal>
     may be specified for the <literal>backend</literal> element. These
     semantics of these attributes is equivalent to those for the
     <literal>retrieval</literal>. However, these values are passed to
     the "backend".
    </para>
    <para>
     The <literal>backend</literal> element may includes one or more
     conversion instructions (as children elements). The supported
     conversions are:
     <variablelist>
      <varlistentry><term><literal>marc</literal></term>
       <listitem>
        <para>
         The <literal>marc</literal> element specifies a conversion 
         to - and from ISO2709 encoded MARC and 
         <ulink url="&url.marcxml;">&acro.marcxml;</ulink>/MarcXchange.
         The following attributes may be specified:

         <variablelist>
          <varlistentry><term><literal>inputformat</literal> (REQUIRED)</term>
           <listitem>
            <para>
             Format of input. Supported values are 
            <literal>marc</literal> (for ISO2709); and <literal>xml</literal>
             for MARCXML/MarcXchange.
            </para>
           </listitem>
          </varlistentry>

          <varlistentry><term><literal>outputformat</literal> (REQUIRED)</term>
           <listitem>
            <para>
             Format of output. Supported values are 
            <literal>line</literal> (MARC line format); 
            <literal>marcxml</literal> (for MARCXML),
            <literal>marc</literal> (ISO2709),
            <literal>marcxhcange</literal> (for MarcXchange).
            </para>
           </listitem>
          </varlistentry>

          <varlistentry><term><literal>inputcharset</literal> (OPTIONAL)</term>
           <listitem>
            <para>
             Encoding of input. For XML input formats, this need not
             be given, but for ISO2709 based inputformats, this should
             be set to the encoding used. For MARC21 records, a common
             inputcharset value  would be <literal>marc-8</literal>.
            </para>
           </listitem>
          </varlistentry>

          <varlistentry><term><literal>outputcharset</literal> (OPTIONAL)</term>
           <listitem>
            <para>
             Encoding of output. If outputformat is XML based, it is
             strongly recommened to use <literal>utf-8</literal>.
            </para>
           </listitem>
          </varlistentry>

         </variablelist>
        </para>
       </listitem>
      </varlistentry>
      <varlistentry><term><literal>xslt</literal></term>
       <listitem>
        <para>
         The <literal>xslt</literal> element specifies a conversion
         via &acro.xslt;. The following attributes may be specified:

         <variablelist>
          <varlistentry><term><literal>stylesheet</literal> (REQUIRED)</term>
           <listitem>
            <para>
             Stylesheet file.
            </para>
           </listitem>
          </varlistentry>
         </variablelist>

        </para>
       </listitem>
      </varlistentry>
     </variablelist>
    </para>
   </sect2>
   <sect2 id="tools.retrieval.examples">
    <title>Retrieval Facility Examples</title>
    <example id="tools.retrieval.marc21">
     <title>MARC21 backend</title>
     <para>
      A typical way to use the retrieval facility is to enable XML
      for servers that only supports ISO2709 encoded MARC21 records.
     </para>
     <programlisting><![CDATA[
     <retrievalinfo>
       <retrieval syntax="usmarc" name="F"/>
       <retrieval syntax="usmarc" name="B"/>
       <retrieval syntax="xml" name="marcxml"
                  identifier="info:srw/schema/1/marcxml-v1.1">
         <backend syntax="usmarc" name="F">
           <marc inputformat="marc" outputformat="marcxml"
                 inputcharset="marc-8"/>
         </backend>
       </retrieval>
       <retrieval syntax="xml" name="dc">
         <backend syntax="usmarc" name="F">
           <marc inputformat="marc" outputformat="marcxml"
                 inputcharset="marc-8"/>
           <xslt stylesheet="MARC21slim2DC.xsl"/>
         </backend>
       </retrieval>
     </retrievalinfo>
]]>
     </programlisting>
     <para>
      This means that our frontend supports:
      <itemizedlist>
       <listitem>
        <para>
         MARC21 F(ull) records.
        </para>
       </listitem>
       <listitem>
        <para>
         MARC21 B(rief) records.
        </para>
       </listitem>

       <listitem>
        <para>
         MARCXML records.
        </para>
       </listitem>

       <listitem>
        <para>
         Dublin core records.
        </para>
       </listitem>
      </itemizedlist>
     </para>
    </example>
   </sect2>
   <sect2 id="tools.retrieval.api">
    <title>API</title>
    <para>
     It should be easy to use the retrieval systems from applications. Refer
     to the headers
     <filename>yaz/retrieval.h</filename> and 
     <filename>yaz/record_conv.h</filename>.
    </para>
   </sect2>
  </sect1>
 </chapter>
 
 <!-- Keep this comment at the end of the file
 Local variables:
 mode: sgml
 sgml-omittag:t
 sgml-shorttag:t
 sgml-minimize-attributes:nil
 sgml-always-quote-attributes:t
 sgml-indent-step:1
 sgml-indent-data:t
 sgml-parent-document: "yaz.xml"
 sgml-local-catalogs: nil
 sgml-namecase-general:t
 End:
 -->