The London Perl and Raku Workshop takes place on 26th Oct 2024. If your company depends on Perl, please consider sponsoring and/or attending.


Net::Z3950::SimpleServer - Simple Perl API for building Z39.50 servers.


  use Net::Z3950::SimpleServer;

  sub my_search_handler {
        my $args = shift;

        my $set_id = $args->{SETNAME};
        my @database_list = @{ $args->{DATABASES} };
        my $query = $args->{QUERY};

        ## Perform the query on the specified set of databases
        ## and return the number of hits:

        $args->{HITS} = $hits;

  sub my_fetch_handler {        # Get a record for the user
        my $args = shift;

        my $set_id = $args->{SETNAME};

        my $record = fetch_a_record($args->{OFFSET});

        $args->{RECORD} = $record;
        if (number_of_hits() == $args->{OFFSET}) {      ## Last record in set?
                $args->{LAST} = 1;
        } else {
                $args->{LAST} = 0;

  ## Register custom event handlers:
  my $z = new Net::Z3950::SimpleServer(GHANDLE = $someObject,
                                       INIT   =>  \&my_init_handler,
                                       CLOSE  =>  \&my_close_handler,
                                       SEARCH =>  \&my_search_handler,
                                       FETCH  =>  \&my_fetch_handler);

  ## Launch server:
  $z->launch_server("", @ARGV);


The SimpleServer module is a tool for constructing Z39.50 "Information Retrieval" servers in Perl. The module is easy to use, but it does help to have an understanding of the Z39.50 query structure and the construction of structured retrieval records.

Z39.50 is a network protocol for searching remote databases and retrieving the results in the form of structured "records". It is widely used in libraries around the world, as well as in the US Federal Government. In addition, it is generally useful whenever you wish to integrate a number of different database systems around a shared, abstract data model.

The model of the module is simple: It implements a "generic" Z39.50 server, which invokes callback functions supplied by you to search for content in your database. You can use any tools available in Perl to supply the content, including modules like DBI and WWW::Search.

The server will take care of managing the network connections for you, and it will spawn a new process (or thread, in some environments) whenever a new connection is received.

The programmer can specify subroutines to take care of the following type of events:

  - Start service (called once).
  - Initialize request
  - Search request
  - Present request
  - Fetching of records
  - Scan request (browsing) 
  - Closing down connection

Note that only the Search and Fetch handler functions are required. The module can supply default responses to the other on its own.

After the launching of the server, all control is given away from the Perl script to the server. The server calls the registered subroutines to field incoming requests from Z39.50 clients.

A reference to an anonymous hash is passed to each handler. Some of the entries of these hashes are to be considered input and others output parameters.

The Perl programmer specifies the event handlers for the server by means of the SimpleServer object constructor

  my $z = new Net::Z3950::SimpleServer(
                        START   =>      \&my_start_handler,
                        INIT    =>      \&my_init_handler,
                        CLOSE   =>      \&my_close_handler,
                        SEARCH  =>      \&my_search_handler,
                        PRESENT =>      \&my_present_handler,
                        SCAN    =>      \&my_scan_handler,
                        FETCH   =>      \&my_fetch_handler,
                        EXPLAIN =>      \&my_explain_handler,
                        DELETE  =>      \&my_delete_handler,
                        SORT    =>      \&my_sort_handler);

In addition, the arguments to the constructor may include GHANDLE, a global handle which is made available to each invocation of every callback function. This is typically a reference to either a hash or an object.

If you want your SimpleServer to start a thread (threaded mode) to handle each incoming Z39.50 request instead of forking a process (forking mode), you need to register the handlers by symbol rather than by code reference. Thus, in threaded mode, you will need to register your handlers this way:

  my $z = new Net::Z3950::SimpleServer(
                        INIT    =>      "my_package::my_init_handler",
                        CLOSE   =>      "my_package::my_close_handler",
                        ....          );

where my_package is the Perl package in which your handler is located.

After the custom event handlers are declared, the server is launched by means of the method

  $z->launch_server("", @ARGV);

Notice, the first argument should be the name of your server script (for logging purposes), while the rest of the arguments are documented in the YAZ toolkit manual: The section on application invocation: <>

In particular, you need to use the -T switch to start your SimpleServer in threaded mode.

Start handler

The start handler is called when service is started. The argument hash passed to the start handler has the form

  $args = {
             CONFIG =>  "default-config" ## GFS config (as given by -c)

The purpose of the start handler is to read the configuration file for the Generic Frontend Server . This is specified by option -c. If -c is omitted, the configuration file is set to "default-config".

The start handler is optional. It is supported in Simpleserver 1.16 and later.

Init handler

The init handler is called whenever a Z39.50 client is attempting to logon to the server. The exchange of parameters between the server and the handler is carried out via an anonymous hash reached by a reference, i.e.

  $args = shift;

The argument hash passed to the init handler has the form

  $args = {
                                    ## Response parameters:

             PEER_NAME =>  "",      ## Name or IP address of connecting client
             IMP_ID    =>  "",      ## Z39.50 Implementation ID
             IMP_NAME  =>  "",      ## Z39.50 Implementation name
             IMP_VER   =>  "",      ## Z39.50 Implementation version
             ERR_CODE  =>  0,       ## Error code, cnf. Z39.50 manual
             ERR_STR   =>  "",      ## Error string (additional info.)
             USER      =>  "xxx"    ## If Z39.50 authentication is used,
                                    ## this member contains user name
             PASS      =>  "yyy"    ## Under same conditions, this member
                                    ## contains the password in clear text
             GHANDLE   =>  $obj     ## Global handle specified at creation
             HANDLE    =>  undef    ## Handler of Perl data structure

The HANDLE member can be used to store any scalar value which will then be provided as input to all subsequent calls (i.e. for searching, record retrieval, etc.). A common use of the handle is to store a reference to a hash which may then be used to store session-specific parameters. If you have any session-specific information (such as a list of result sets or a handle to a back-end search engine of some sort), it is always best to store them in a private session structure - rather than leaving them in global variables in your script.

The Implementation ID, name and version are only really used by Z39.50 client developers to see what kind of server they're dealing with. Filling these in is optional.

The ERR_CODE should be left at 0 (the default value) if you wish to accept the connection. Any other value is interpreted as a failure and the client will be shown the door, with the code and the associated additional information, ERR_STR returned.

Search handler

Similarly, the search handler is called with a reference to an anonymous hash. The structure is the following:

  $args = {
                                     ## Request parameters:

    GHANDLE             =>  $obj     # Global handle specified at creation
    HANDLE              =>  ref,     # Your session reference.
    SETNAME             =>  "id",    # ID of the result set
    REPL_SET            =>  0,       # Replace set if already existing?
    DATABASES           =>  ["xxx"], # Reference to a list of databases to search
    QUERY               =>  "query", # The query expression as a PQF string
    RPN                 =>  $obj,    # Reference to a Net::Z3950::APDU::Query
    CQL                 =>  $x,      # A CQL query, if this is provided instead of Type-1
    SRW_SORTKEYS        =>  $x,      # XXX to be described
    PID                 =>  $x,      # XXX to be described
    PRESENT_NUMBER      =>  $x,      # XXX to be described
    EXTRA_ARGS          =>  $x,      # XXX to be described
    INPUTFACETS         =>  $x,      # Specification of facets required: see below.

                                     ## Response parameters:

    ERR_CODE            =>  0,       # Error code (0=Successful search)
    ERR_STR             =>  "",      # Error string
    HITS                =>  0,       # Number of matches
    ESTIMATED_HIT_COUNT =>  $x,      # XXX to be described
    EXTRA_RESPONSE_DATA =>  $x,      # XXX to be described
    OUTPUTFACETS        =>  $x       # Facets returned: see below.

Note that a search which finds 0 hits is considered successful in Z39.50 terms - you should only set the ERR_CODE to a non-zero value if there was a problem processing the request. The Z39.50 standard provides a comprehensive list of standard diagnostic codes, and you should use these whenever possible.

Query structures

In Z39.50, the most comment kind of query is the so-called Type-1 _query, a tree-structure of terms combined by operators, the terms being qualified by lists of attributes.

The QUERY parameter presented this tree to the search function in the Prefix Query Format (PQF) which is used in many applications based on the YAZ toolkit. The full grammar is described in the YAZ manual.

The following are all examples of valid queries in the PQF.


        "bob dylan"

        @or "dylan" "zimmerman"

        @set Result-1

        @or @and bob dylan @set Result-1

        @and @attr 1=1 "bob dylan" @attr 1=4 "slow train coming"

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

You will need to write a recursive function or something similar to parse incoming query expressions, and this is usually where a lot of the work in writing a database-backend happens. Fortunately, you don't need to support any more functionality than you want to. For instance, it is perfectly legal to not accept boolean operators, but you should try to return good error codes if you run into something you can't or won't support.

A more convenient alternative to the QUERY member is the RPN member, which is a reference to a Net::Z3950::APDU::Query object representing the RPN query tree. The structure of that object is supposed to be self-documenting, but here's a brief summary of what you get:

  • Net::Z3950::APDU::Query is a hash with two fields:


    Optional. If present, it is a reference to a Net::Z3950::APDU::OID. This is a string of dot-separated integers representing the OID of the query's top-level attribute set.


    Mandatory: a reference to the RPN tree itself.

  • Each node of the tree is an object of one of the following types:


    These three classes are all arrays of two elements, each of which is a node.


    A query term. See below for details.


    A reference to a result-set ID indicating a previous search. The ID of the result-set is in the id element.

  • Net::Z3950::RPN::Term is a hash with two fields:


    A string containing the search term itself.


    A reference to a Net::Z3950::RPN::Attributes object.

  • Net::Z3950::RPN::Attributes is an array of references to Net::Z3950::RPN::Attribute objects. (Note the plural/singular distinction.)

  • Net::Z3950::RPN::Attribute is a hash with three elements:


    Optional. If present, it is dot-separated OID string, as above.


    An integer indicating the type of the attribute - for example, under the BIB-1 attribute set, type 1 indicates a ``use'' attribute, type 2 a ``relation'' attribute, etc.


    An integer or string indicating the value of the attribute - for example, under BIB-1, if the attribute type is 1, then value 4 indicates a title search and 7 indicates an ISBN search; but if the attribute type is 2, then value 4 indicates a ``greater than or equal'' search, and 102 indicates a relevance match.

All of these classes except Attributes and Attribute are subclasses of the abstract class Net::Z3950::RPN::Node. That class has a single method, toPQF(), which may be used to turn an RPN tree, or part of one, back into a textual prefix query.

Note that, apart to toPQF(), none of these classes have any methods at all: the blessing into classes is largely just a documentation thing so that, for example, if you do

        { use Data::Dumper; print Dumper($args->{RPN}) }

you get something fairly human-readable. But of course, the type distinction between the three different kinds of boolean node is important.

By adding your own methods to these classes (building what I call ``augmented classes''), you can easily build code that walks the tree of the incoming RPN. Take a look at samples/ for a sample implementation of such an augmented classes technique.

Finally, when SimpleServer is invoked using SRU/SRW (and indeed using Z39.50 if the unusual type-104 query is used), the query that is _passed is expressed in CQL, the Contextual Query Language. In this case, the query string is made available in the CQL argument.


Servers may support the provision of facets -- counted lists of field values which may subsequently be be used as query terms to narrow the search.

In SRU, facets may be requested by the facetLimit parameter, as documented in the OASIS standard that formalises the SRU specification. Its value is a string consisting of a comma-separated list of facet specifications. Each facet specification consists of of a count, a colon and a fieldname. For example, facetLimit=10:title,5:author asks for ten title facets and five author facets.

Request format

The facet request is passed to the search-handler function in the INPUTFACETS parameter. Its value is rather complex, due to backwards compatibility with Z39.50:

  • The top-level value is a Net::Z3950::FacetList array.

  • This is an array of Net::Z3950::FacetField objects.

  • Each of these is an object with two members, attributes and terms.

  • attributes has type Net::Z3950::RPN::Attributes and is a list of objects of type Net::Z3950::RPN::Attribute.

  • Each attribute has two elements, attributeType and attributeValue. Each value is interpreted according to its type. The meanings of the types are as follows:

    1. The name of the field to provide values of the facets.

    2. The order in which to sort the values. (But it's not clear how this is to be interpreted: it may be implementation dependent.)

    3. The number of facets to include for the specified field.

    4. The first facet to include in the response: for example, if this is 11, then the first ten facts should be skipped.

So for example, the SRU facet specification facetLimit=10:title,5:author would be translated as a Net::Z3950::FacetList list of two Net::Z3950::FacetFields. The attributes of the first would be [1="title", 3=10], and those of the second would be [1="author", 3=5].

It is not clear what the purpose of terms is, but for the record, this is how it is represented:

  • terms is a Net::Z3950::FacetTerms array.

  • This is an array of Net::Z3950::FacetTerm objects.

  • Each of these is an object with two members, term and count. The first of these is an integer, the second a string.

Response format

Having generated facets corresponding to the request, the search handler should return them in the OUTPUTFACETS argument. The structure of this response is similar to that of the request:

  • The top-level value is a Net::Z3950::FacetList array.

  • This is an array of Net::Z3950::FacetField objects.

  • Each of these is an object with two members, attributes and terms.

  • attributes has type Net::Z3950::RPN::Attributes and is a list of objects of type Net::Z3950::RPN::Attribute.

  • Each attribute has two elements, attributeType and attributeValue. Each value is interpreted according to its type. The meanings of the types are as follows:

    1. The name of the field for which terms are provided.

    (That is the only type used.)

  • terms is a Net::Z3950::FacetTerms array.

  • This is an array of Net::Z3950::FacetTerm objects.

  • Each of these is an object with two members, term and count. The first of these is a string containing one of the facet terms, and the second is an integer indicating how many times it occurs in the records that were found by the search.

The example SimpleServer applicaation server includes code that shows how to examine the INPUTFACETS data structure and create the OUTPUTFACETS structure.

Present handler

The presence of a present handler in a SimpleServer front-end is optional. Each time a client wishes to retrieve records, the present service is called. The present service allows the origin to request a certain number of records retrieved from a given result set. When the present handler is called, the front-end server should prepare a result set for fetching. In practice, this means to get access to the data from the backend database and store the data in a temporary fashion for fast and efficient fetching. The present handler does *not* fetch anything. This task is taken care of by the fetch handler, which will be called the correct number of times by the YAZ library. More about this below. If no present handler is implemented in the front-end, the YAZ toolkit will take care of a minimum of preparations itself. This default present handler is sufficient in many situations, where only a small amount of records are expected to be retrieved. If on the other hand, large result sets are likely to occur, the implementation of a reasonable present handler can gain performance significantly.

The information exchanged between client and present handle is:

  $args = {
                                    ## Client/server request:

             GHANDLE   =>  $obj     ## Global handle specified at creation
             HANDLE    =>  ref,     ## Reference to datastructure
             SETNAME   =>  "id",    ## Result set ID
             START     =>  xxx,     ## Start position
             COMP      =>  "",      ## Desired record composition
             NUMBER    =>  yyy,     ## Number of requested records

                                    ## Response parameters:

             HITS      =>  zzz,     ## Number of returned records
             ERR_CODE  =>  0,       ## Error code
             ERR_STR   =>  ""       ## Error message

Fetch handler

The fetch handler is asked to retrieve a SINGLE record from a given result set (the front-end server will automatically call the fetch handler as many times as required).

The parameters exchanged between the server and the fetch handler are:

  $args = {
                                    ## Client/server request:

             GHANDLE   =>  $obj     ## Global handle specified at creation
             HANDLE    =>  ref      ## Reference to data structure
             SETNAME   =>  "id"     ## ID of the requested result set
             OFFSET    =>  nnn      ## Record offset number
             REQ_FORM  =>  "n.m.k.l"## Client requested format OID
             COMP      =>  "xyz"    ## Formatting instructions
             SCHEMA    =>  "abc"    ## Requested schema, if any

                                    ## Handler response:

             RECORD    =>  ""       ## Record string
             BASENAME  =>  ""       ## Origin of returned record
             LAST      =>  0        ## Last record in set?
             ERR_CODE  =>  0        ## Error code
             ERR_STR   =>  ""       ## Error string
             SUR_FLAG  =>  0        ## Surrogate diagnostic flag
             REP_FORM  =>  "n.m.k.l"## Provided format OID
             SCHEMA    =>  "abc"    ## Provided schema, if any

The REP_FORM value has by default the REQ_FORM value, but can be set to something different if the handler desires. The BASENAME value should contain the name of the database from where the returned record originates. The ERR_CODE and ERR_STR works the same way they do in the search handler. If there is an error condition, the SUR_FLAG is used to indicate whether the error condition pertains to the record currently being retrieved, or whether it pertains to the operation as a whole (e.g. the client has specified a result set which does not exist.)

If you need to return USMARC records, you might want to have a look at the MARC module on CPAN, if you don't already have a way of generating these.

NOTE: The record offset is 1-indexed, so 1 is the offset of the first record in the set.

Scan handler

A full featured Z39.50 server should support scan (or in some literature browse). The client specifies a starting term of the scan, and the server should return an ordered list of specified length consisting of terms actually occurring in the data base. Each of these terms should be close to or equal to the term originally specified. The quality of scan compared to simple search is a guarantee of hits. It is simply like browsing through an index of a book, you always find something! The parameters exchanged are:

  $args = {
                                                ## Client request

                GHANDLE         => $obj,        ## Global handle specified at creation
                HANDLE          => $ref,        ## Reference to data structure
                DATABASES       => ["xxx"],     ## Reference to a list of data-
                                                ## bases to search
                TERM            => 'start',     ## The start term
                RPN             =>  $obj,       ## Reference to a Net::Z3950::RPN::Term

                NUMBER          => xx,          ## Number of requested terms
                POS             => yy,          ## Position of starting point
                                                ## within returned list
                STEP            => 0,           ## Step size

                                                ## Server response

                ERR_CODE        => 0,           ## Error code
                ERR_STR         => '',          ## Diagnostic message
                NUMBER          => zz,          ## Number of returned terms
                STATUS          => $status,     ## ScanSuccess/ScanFailure
                ENTRIES         => $entries     ## Referenced list of terms

where the term list is returned by reference in the scalar $entries, which should point at a data structure of this kind,

  my $entries = [
                        {       TERM            => 'energy',
                                OCCURRENCE      => 5            },

                        {       TERM            => 'energy density',
                                OCCURRENCE      => 6,           },

                        {       TERM            => 'energy flow',
                                OCCURRENCE      => 3            },



The $status flag is only meaningful after a successful scan, and should be assigned one of two values:

  Net::Z3950::SimpleServer::ScanSuccess  Full success (default)
  Net::Z3950::SimpleServer::ScanPartial  Fewer terms returned than requested

The STEP member contains the requested number of entries in the term-list between two adjacent entries in the response.

A better alternative to the TERM member is the the RPN member, which is a reference to a Net::Z3950::RPN::Term object representing the scan clause. The structure of that object is the same as for Term objects included as part of the RPN tree passed to search handlers. This is more useful than the simple TERM because it includes attributes (e.g. access points associated with the term), which are discarded by the TERM element.

Close handler

The argument hash received by the close handler has two elements only:

  $args = {
                                    ## Server provides:

             GHANDLE   =>  $obj     ## Global handle specified at creation
             HANDLE    =>  ref      ## Reference to data structure

What ever data structure the HANDLE value points at goes out of scope after this call. If you need to close down a connection to your server or something similar, this is the place to do it.

Explain handler

The argument hash received by the explain handler has the following elements:

  $args = {
                            ## Request parameters:
     GHANDLE   =>  $obj,    # Global handle specified at creation
     HANDLE    =>  ref,     # Reference to data structure
     DATABASE  =>  $dbname, # Name of database to be explained

                            ## Response parameters:
     EXPLAIN   =>  $zeerex  # ZeeRex record for specified database

The handler should return a string containing the ZeeRex XML that describes that nominated database.

Delete handler

The argument hash received by the delete handler has the following elements:

  $args = {
                                    ## Client request:
             GHANDLE   =>  $obj,    ## Global handle specified at creation
             HANDLE    =>  ref,     ## Reference to data structure
             SETNAME   =>  "id",    ## Result set ID

                                    ## Server response:
             STATUS    => 0         ## Deletion status

The SETNAME element of the argument hash may or may not be defined. If it is, then SETNAME is the name of a result set to be deleted; if not, then all result-sets associated with the current session should be deleted. In either case, the callback function should report on success or failure by setting the STATUS element either to zero, on success, or to an integer from 1 to 10, to indicate one of the ten possible failure codes described in section of the Z39.50 standard -- see

Sort handler

The argument hash received by the sort handler has the following elements:

        $args = {
                                        ## Client request:
                GHANDLE => $obj,        ## Global handle specified at creation
                HANDLE => ref,          ## Reference to data structure
                INPUT => [ a, b ... ],  ## Names of result-sets to sort
                OUTPUT => "name",       ## Name of result-set to sort into
                SEQUENCE                ## Sort specification: see below

                                        ## Server response:
                STATUS => 0,            ## Success, Partial or Failure
                ERR_CODE => 0,          ## Error code
                ERR_STR => '',          ## Diagnostic message


The SEQUENCE element is a reference to an array, each element of which is a hash representing a sort key. Each hash contains the following elements:


0 for an ascending sort, 1 for descending, 3 for ascending by frequency, or 4 for descending by frequency.


0 for a case-sensitive sort, 1 for case-insensitive


How to respond if one or more records in the set to be sorted are missing the fields indicated in the sort specification. 1 to abort the sort, 2 to use a "null value", 3 if a value is provided to use in place of the missing data (although in the latter case, the actual value to use is currently not made available, so this is useless).

And one or other of the following:


A string indicating the field to be sorted, which the server may interpret as it sees fit (presumably by an out-of-band agreement with the client).


I have no idea what this is.


ATTRSET is the attribute set from which the attributes are taken, and SORT_ATTR is a reference to an array containing the attributes themselves. Each attribute is represented by (are you following this carefully?) yet another hash, this one containing the elements ATTR_TYPE and ATTR_VALUE: for example, type=1 and value=4 in the BIB-1 attribute set would indicate access-point 4 which is title, so that a sort of title is requested.

Precisely why all of the above is so is not clear, but goes some way to explain why, in the Z39.50 world, the developers of the standard are not so much worshiped as blamed.

The backend function should set STATUS to 0 on success, 1 for "partial success" (don't ask) or 2 on failure, in which case ERR_CODE and ERR_STR should be set.

Support for SRU and SRW

Since release 1.0, SimpleServer includes support for serving the SRU and SRW protocols as well as Z39.50. These ``web-friendly'' protocols enable similar functionality to that of Z39.50, but by means of rich URLs in the case of SRU, and a SOAP-based web-service in the case of SRW. These protocols are described at

In order to serve these protocols from a SimpleServer-based application, it is necessary to launch the application with a YAZ Generic Frontend Server (GFS) configuration file, which can be specified using the command-line argument -f filename. A minimal configuration file looks like this:


This file specifies only that should be used to translate the CQL queries of SRU and SRW into corresponding Z39.50 Type-1 queries. For more information about YAZ GFS configuration, including how to specify an Explain record, see the Virtual Hosts section of the YAZ manual at

The mapping of CQL queries into Z39.50 Type-1 queries is specified by a file that indicates which BIB-1 attributes should be generated for each CQL index, relation, modifiers, etc. A typical section of this file looks like this:

  index.dc.title                        = 1=4
  index.dc.subject                      = 1=21
  index.dc.creator                      = 1=1003
  relation.<                            = 2=1
  relation.le                           = 2=2

This file specifies the BIB-1 access points (type=1) for the Dublin Core indexes title, subject and creator, and the BIB-1 relations (type=2) corresponding to the CQL relations < and <=. For more information about the format of this file, see the CQL section of the YAZ manual at

The YAZ distribution includes a sample CQL-to-PQF mapping configuration file called; this is sufficient for many applications, and a good base to work from for most others.

If a SimpleServer-based application is run without this SRU-specific configuration, it can still serve SRU; however, CQL queries will not be translated, but passed straight through to the search-handler function, as the CQL member of the parameters hash. It is then the responsibility of the back-end application to parse and handle the CQL query, which is most easily done using Ed Summers' fine CQL::Parser module, available from CPAN at


Anders Sønderberg (, Sebastian Hammer (, Mike Taylor (


Copyright (C) 2000-2016 by Index Data.

This library is free software; you can redistribute it and/or modify it under the same terms as Perl itself, either Perl version 5.8.4 or, at your option, any later version of Perl 5 you may have available.


Any Perl module which is useful for accessing the data source of your choice.

1 POD Error

The following errors were encountered while parsing the POD:

Around line 1104:

Non-ASCII character seen before =encoding in 'Sønderberg'. Assuming UTF-8