DBIx::DBStag - Relational Database to Hierarchical (Stag/XML) Mapping


  use DBIx::DBStag;
  my $dbh = DBIx::DBStag->connect("dbi:Pg:dbname=moviedb");
  my $sql = q[
               studio NATURAL JOIN 
               movie NATURAL JOIN
               movie_to_star NATURAL JOIN
               movie.genre = 'sci-fi' AND star.lastname = 'Fisher'
              USE NESTING
  my $dataset = $dbh->selectall_stag($sql);
  my @studios = $dataset->get_studio;

  # returns nested data that looks like this -
  # (studio
  #  (name "20th C Fox")
  #  (movie
  #   (name "star wars") (genre "sci-fi")
  #   (star
  #    (firstname "Carrie")(lastname "Fisher")))))

  # iterate through result tree -
  foreach my $studio (@studios) {
        printf "STUDIO: %s\n", $studio->get_name;
        my @movies = $studio->get_movie;

        foreach my $movie (@movies) {
            printf "  MOVIE: %s (genre:%s)\n", 
              $movie->get_name, $movie->get_genre;
            my @stars = $movie->get_star;

            foreach my $star (@stars) {
                printf "    STARRING: %s:%s\n", 
                  $star->get_firstname, $star->get_lastname;
  # manipulate data then store it back in the database
  my @allstars = $dataset->get("movie/studio/star");
  $_->set_fullname($_->get_firstname.' '.$_->get_lastname)

  exit 0;

Or from the command line:

  unix> -d 'dbi:Pg:dbname=moviebase'     \
       'SELECT * FROM studio NATURAL JOIN movie NATURAL   \
          JOIN movie_to_star NATURAL JOIN star            \
          USE NESTING (set(studio(movie(star))))'

Or using a predefined template:

  unix> -d moviebase /mdb-movie genre=sci-fi


This module is for mapping between relational databases and Stag objects (Structured Tags - see Data::Stag). Stag objects can also be represented as XML. The module has two main uses:


This module can take the results of any SQL query and decompose the flattened results into a tree data structure which reflects the foreign keys in the underlying relational schema. It does this by looking at the SQL query and introspecting the database schema, rather than requiring metadata or an object model.

In this respect, the module works just like a regular DBI handle, with a few extra methods.

Queries can also make use of predefined templates

Storing Data

DBStag objects can store any tree-like datastructure (such as XML documents) into a database using normalized schema that reflects the structure of the tree being stored. This is done using little or no metadata.

XML can also be imported, and a relational schema automatically generated.

For a tutorial on using DBStag to build and query relational databases from XML sources, please see DBIx::DBStag::Cookbook


This is a general overview of the rules for turning SQL query results into a tree like data structure. You don't need to understand all these rules to be able to use this module - you can experiment by using the script which comes with this distribution.

Mapping Relations

Relations (i.e. tables and views) are elements (nodes) in the tree. The elements have the same name as the relation in the database.

These nodes are always non-terminal (ie they always have child nodes)

Mapping Columns

Table and view columns of a relation are sub-elements of the table or view to which they belong. These elements will be data elements (i.e. terminal nodes). Only the columns selected in the SQL query will be present.

For example, the following query

  SELECT name, job FROM person;

will return a data structure that looks like this:

    (name "fred")
    (job "forklift driver"))
    (name "joe")
    (job "steamroller mechanic")))

The data is shown as a lisp-style S-Expression - it can also be expressed as XML, or manipulated as an object within perl.

Handling table aliases

If an ALIAS is used in the FROM part of the SQL query, the relation element will be nested inside an element with the same name as the alias. For instance, the query

  SELECT name FROM person AS author WHERE job = 'author';

Will return a data structure like this:

     (name "Philip K Dick"))))

The underlying assumption is that aliasing is used for a purpose in the original query; for instance, to determine the context of the relation where it may be ambiguous.

  FROM person AS employee 
           INNER JOIN 
       person AS boss ON (employee.boss_id = boss.person_id)

Will generate a nested result structure similar to this -

     (person_id "...")
     (name "...")
     (salary  "...")
       (person_id "...")
       (name "...")
       (salary  "..."))))))

If we neglected the alias, we would have 'person' directly nested under 'person', and the meaning would not be obvious. Note how the contents of the SQL query dynamically modifies the schema/structure of the result tree.


Right now, DBStag is fussy about how you specify aliases; you must use AS - you must say

  SELECT name FROM person AS author;

instead of

  SELECT name FROM person author;

Nesting of relations

The main utility of querying using this module is in retrieving the nested relation elements from the flattened query results. Given a query over relations A, B, C, D,... there are a number of possible tree structures. Not all of the tree structures are meaningful or useful.

Usually it will make no sense to nest A under B if there is no foreign key relationship linking either A to B, or B to A. This is not always the case - it may be desirable to nest A under B if there is an intermediate linking table that is required at the relational level but not required in the tree structure.

DBStag will guess a structure/schema based on the ordering of the relations in your FROM clause. However, this guess can be over-ridden at either the SQL level (using DBStag specific SQL extensions) or at the API level.

The default algorithm is to nest each relation element under the relation element preceeding it in the FROM clause; for instance:


If there are appropriately named foreign keys, the following data will be returned (assuming one column 'x_foo' in each of a, b and c)

    (a_foo "...")
     (b_foo "...")
      (c_foo "...")))))

where 'x_foo' is a column in relation 'x'

This is not always desirable. If both b and c have foreign keys into table a, DBStag will not detect this - you have to guide it. There are two ways of doing this - you can guide by bracketing your FROM clause like this:


This will generate

    (a_foo "...")
     (b_foo "..."))
     (c_foo "..."))))

Now b and c are siblings in the tree. The algorithm is similar to before: nest each relation element under the relation element preceeding it; or, if the preceeding item in the FROM clause is a bracketed structure, nest it under the first relational element in the bracketed structure.

(Note that in MySQL you may not place brackets in the FROM clause in this way)

Another way to achieve the same thing is to specify the desired tree structure using a DBStag specific SQL extension. The DBStag specific component is removed from the SQL before being presented to the DBMS. The extension is the USE NESTING clause, which should come at the end of the SQL query (and is subsequently removed before processing by the DBMS).

  USE NESTING (set (a (b)(c)));

This will generate the same tree as above (i.e. 'b' and 'c' are siblings). Notice how the nesting in the clause is the same as the nesting in the resulting tree structure.

Note that 'set' is not a table in the underlying relational schema - the result data tree requires a named top level node to group all the 'a' relations under. You can call this top level element whatever you like.

If you are using the DBStag API directly, you can pass in the nesting structure as an argument to the select call; for instance:

  my $xmlstr =
    $dbh->selectall_xml(-sql=>q[SELECT * 
                                FROM a NATURAL JOIN b 
                                     NATURAL JOIN c],
                        -nesting=>'(set (a (b)(c)))');

or the equivalent -

  my $xmlstr =
    $dbh->selectall_xml(q[SELECT * 
                          FROM a NATURAL JOIN b 
                               NATURAL JOIN c],
                        '(set (a (b)(c)))');

If you like, you can also use XML here (only at the API level, not at the SQL level) -

  my $seq =
    $dbh->selectall_xml(-sql=>q[SELECT * 
                                FROM a NATURAL JOIN b 
                                     NATURAL JOIN c],

As you can see, this is a little more verbose than the S-Expression

Most command line scripts that use this module should allow pass-through via the '-nesting' switch.

Aliasing of functions and expressions

If you alias a function or an expression, DBStag needs to know where to put the resulting column; the column must be aliased.

This is inferred from the first named column in the function or expression; for example, the SQL below uses the minus function:

  SELECT blah.*, foo.*, foo.x-foo.y AS z

The z element will be nested under the foo element

You can force different nesting using a double underscore:

  SELECT blah.*, foo.*, foo.x - foo.y AS blah__z

This will nest the z element under the blah element

If you would like to override this behaviour and use the alias as the element name, pass in the -aliaspolicy=>'a' arg to the API call. If you wish to use the table names without nesting, use -aliaspolicy=>'t'.

Conformance to DTD/XML-Schema

DBStag returns Data::Stag structures that are equivalent to a simplified subset of XML (and also a simplified subset of lisp S-Expressions).

These structures are examples of semi-structured data - a good reference is this book -

  Data on the Web: From Relations to Semistructured Data and XML
  Serge Abiteboul, Dan Suciu, Peter Buneman
  Morgan Kaufmann; 1st edition (January 2000)

The schema for the resulting Stag structures can be seen to conform to a schema that is dynamically determined at query-time from the underlying relational schema and from the specification of the query itself.

If you need to generate a DTD you can ause the script, which is part of the Data::Stag distribution


The following methods are for using the DBStag API to query a database


  Usage   - $dbh = DBIx::DBStag->connect($DSN);
  Returns - L<DBIx::DBStag>
  Args    - see the connect() method in L<DBI>

This will be the first method you call to initiate a DBStag object

The DSN may be a standard DBI DSN, or it can be a DBStag alias


 Usage   - $stag = $dbh->selectall_stag($sql);
           $stag = $dbh->selectall_stag($sql, $nesting_clause);
           $stag = $dbh->selectall_stag(-template=>$template,
 Returns - L<Data::Stag>
 Args    - sql string, 
           [nesting string], 
           [bind hashref],
           [template DBIx::DBStag::SQLTemplate]

Executes a query and returns a Data::Stag structure

An optional nesting expression can be passed in to control how the relation is decomposed into a tree. The nesting expression can be XML or an S-Expression; see above for details


 Usage   - $xml = $dbh->selectall_xml($sql);
 Returns - string
 Args    - See selectall_stag()

As selectall_stag(), but the results are transformed into an XML string


 Usage   - $sxpr = $dbh->selectall_sxpr($sql);
 Returns - string
 Args    - See selectall_stag()

As selectall_stag(), but the results are transformed into an S-Expression string; see Data::Stag for more details.


 Usage   - $dbh->selectall_sax(-sql=>$sql, -handler=>$sax_handler);
 Returns - string
 Args    - sql string, [nesting string], handler SAX

As selectall_stag(), but the results are transformed into SAX events

[currently this is just a wrapper to selectall_xml but a genuine event generation model will later be used]


 Usage   - $tbl = $dbh->selectall_rows($sql);
 Returns - arrayref of arrayref
 Args    - See selectall_stag()

As selectall_stag(), but the results of the SQL query are left undecomposed and unnested. The resulting structure is just a flat table; the first row is the column headings. This is similar to DBI->selectall_arrayref(). The main reason to use this over the direct DBI method is to take advantage of other stag functionality, such as templates

prepare_stag PRIVATE METHOD

 Usage   - $prepare_h = $dbh->prepare_stag(-template=>$template);
 Returns - hashref (see below)
 Args    - See selectall_stag()

Returns a hashref



The following methods are for using the DBStag API to store nested data in a database


  Usage   - $dbh->storenode($stag);
  Returns - 
  Args    - L<Data::Stag>

SEE ALSO: The script

Recursively stores a stag tree structure in the database.

The database schema is introspected for most of the mapping data, but you can supply your own (see later)

The Stag tree/XML must be a direct mapping of the relational schema. Column and table names must correspond to element names. Elements may be nested. Different styles of XML-Relational mapping may be used: XORT-style and the more compact Stag-style

XORT-style mapping

With a XORT-style mapping, elements corresponding to tables can be nested under elements corresponding to foreign keys.

For example, if the relational schema has a foreign key from table person to table address, the following XML is permissable:


The address node will be stored in the database and collapsed to whatever the value of the primary key is.

Stag-style mapping

Stag-style is more compact, but sometimes relies on the presence of a dbstag_metadata element to specify how foreign keys are mapped


Operations are specified as attributes inside elements, specifying whether the nod should be inserted, updated, looked up or stored/forced. Operations are optional (default is force/store).

  <person op="insert">
   <address_id op="lookup">

The above will always insert into the person table (which may be quite dangerous; if an entry with the same unique constraint exists, an error will be thrown). Assuming (streetaddr,city) is a unique constraint for the address table, this will lookup the specified address (and not modify the table) and use the returned pk value for the person.address_id foreign key

The operations are:

force (default)

looks up (by unique constraints) first; if exists, will do an update. if does not exist, will do an insert


insert only. DBMS will throw error if row with same UC exists


update only. DBMS will throw error if a row the with the specified UC cannot be found


finds the pk value using one of the unique constraints present in the XML node


deletes row that has matching UC

Operations can be used in either XORT or Stag mode


Macros can be used with either XORT or Stag style mappings. Macros allow you to refer to the same node later on in the XML

  <person op="lookup" id="joe">
  <person op="lookup" id="fred">

Assuming name is a unique constraint for person, and person_relationship has two foreign keys named person1_id and person2_id linking to the person table, DBStag will first lookup the two person rows by name (throwing an error if not present) and use the returned pk values to populate the person_relationship table

How it works

Before a node is stored, certain subnodes will be pre-stored; these are subnodes for which there is a foreign key mapping FROM the parent node TO the child node. This pre-storage is recursive.

After these nodes are stored, the current node is either INSERTed or UPDATEd. The database is introspected for UNIQUE constraints; these are used as keys. If there exists a row in the database with matching key, then the node is UPDATEd; otherwise it is INSERTed.

(primary keys from pre-stored nodes become foreign key values in the existing node)

Subsequently, all subnodes that were not pre-stored are now post-stored. The primary key for the existing node will become foreign keys for the post-stored subnodes.


  Usage   - $dbh->force_safe_node_names(1);
  Returns - bool
  Args    - bool [optional]

If this is set, then before storage, all node names are made DB-safe; they are lowercased, and the following transform is applied:



  Usage   - $dbh->mapping(["alias/table.col=fktable.fkcol"]);
  Returns - 
  Args    - array

Creates a stag-relational mapping (for storing data only)

Occasionally not enough information can be obtained from db introspection; you can provide extra mapping data this way.

Occasionally you stag objects/data/XML will contain aliases that do not correspond to actual SQL relations; the aliases are intermediate nodes that provide information on which foreign key column to use

For example, with data like this:

   (name "...")
    (film (....))
    (film (....)))))

There may only be two SQL tables: person and film; person would have two foreign key columns into film. The mapping may look like this


The mapping can also be supplied in the xml that is loaded; any node named "dbstag_metadata" will not be loaded; it is used to supply the mapping. For example:



  Usage   - $dbh->mapconf("mydb-stagmap.stm");
  Returns - 
  Args    - filename

sets the conf file containing the stag-relational mappings

This is not of any use for a XORT-style mapping, where foreign key columns are explicitly stated

See mapping() above

The file contains line like:



  Usage   - $dbh->noupdate_h({person=>1})
  Returns - 
  Args    - hashref

Keys of hash are names of nodes that do not get updated - if a unique key is queried for and does not exist, the node will be inserted and subnodes will be stored; if the unique key does exist in the db, then this will not be updated; subnodes will not be stored


  Usage   - $dbh->trust_primary_key_values(1)
  Returns - bool
  Args    - bool (optional)

The default behaviour of the storenode() method is to remap all surrogate PRIMARY KEY values it comes across.

A surrogate primary key is typically a primary key of type SERIAL (or AUTO_INCREMENT) in MySQL. They are identifiers assigned automatically be the database with no semantics.

It may be desirable to store the same data in two different databases. We would generally not expect the surrogate IDs to match between databases, even if the rest of the data does.

(If you do not use surrogate primary key columns in your load xml, then you can ignore this accessor)

You should NOT use this method in conjunction with Macros

If you use primary key columns in your XML, and the primary keys are not surrogate, then youshould set this. If this accessor is set to non-zero (true) then the primary key values in the XML will be used.

If your db has surrogate/auto-increment/serial PKs, and you wish to use these PK columns in your XML, yet you want to make XML that can be exported from one db and imported into another, then the default behaviour will be fine.

For example, if we extract a 'person' from a db with surrogate PK id and unique key ssno, we may get this:


If we then import this into an entirely fresh db, with no rows in table person, then the default behaviour of storenode() will create a row like this:


The PK val 23 has been mapped to 1 (all foreign keys that point to will now point to

If we were to first call $sdbh->trust_primary_key_values(1), then would remain to be 23. This would only be appropriate behaviour if we were storing back into the same db we retrieved from.


  Usage   - $dbh->tracenode('person/name')

Traces on STDERR inserts/updates on a particular element type (table), displaying the sub-element (column value).

is_caching_on ADVANCED OPTION

  Usage   - $dbh->is_caching_on('person', 1)
  Returns - number
  Args    - number
                   0: off (default)
                   1: memory-caching ON
                   2: memory-caching OFF, bulkload ON
                   3: memory-caching ON, bulkload ON


By default no in-memory caching is used. If this is set to 1, then an in-memory cache is used for any particular element. No cache management is used, so you should be sure not to cache elements that will cause memory overloads.

Setting this will not affect the final result, it is purely an efficiency measure for use with storenode().

The cache is indexed by all unique keys for that particular element/table, wherever those unique keys are set


If bulkload is used without memory-caching (set to 2), then only INSERTs will be performed for this element. Note that this could potentially cause a unique key violation, if the same element is present twice

If bulkload is used with memory-caching (set to 3) then only INSERTs will be performed; the unique serial/autoincrement identifiers for those inserts will be cached and used. This means you can have the same element twice. However, the load must take place in one session, otherwise the contents of memory will be lost


  Usage   - $dbh->clear_cache;
  Returns - 
  Args    - none

Clears the in-memory cache

Caches are not automatically managed - the API user is responsible for making suring the cache does not get too big


  Usage   - print $dbh->cache_summary->xml
  Returns -  L<Data::Stag>
  Args    - 

Gives a summary of the size of the in-memory cache by keys. This can be used for automatic cache management:

  $person_cache = $dbh->cache_summary->get_person;
  my @index_nodes = $person_cache->tnodes;
  foreach (@index_nodes) {
    if ($_->data > MAX_PERSON_CACHE_SIZE) {


DBStag comes with its own SQL templating system. This allows you to reuse the same canned SQL or similar SQL qeuries in different contexts. See DBIx::DBStag::SQLTemplate


  Usage   - $template = $dbh->find_template("my-template-name");
  Returns - L<DBIx::DBStag::SQLTemplate>
  Args    - str

Returns an object representing a canned paramterized SQL query. See DBIx::DBStag::SQLTemplate for documentation on templates


  Usage   - $templates = $dbh->list_templates();
  Returns - Arrayref of L<DBIx::DBStag::SQLTemplate>
  Args    - 

Returns a list of ALL defined templates - See DBIx::DBStag::SQLTemplate


  Usage   - $templates = $dbh->find_templates_by_schema($schema_name);
  Returns - Arrayref of L<DBIx::DBStag::SQLTemplate>
  Args    - str

Returns a list of templates for a particular schema - See DBIx::DBStag::SQLTemplate


  Usage   - $templates = $dbh->find_templates_by_dbname("mydb");
  Returns - Arrayref of L<DBIx::DBStag::SQLTemplate>
  Args    - db name

Returns a list of templates for a particular db

Requires resources to be set up (see below)


Generally when connecting to a database, it is necessary to specify a DBI style DSN locator. DBStag also allows you specify a resource list file which maps logical names to full locators

The following methods allows you to use a resource list


  Usage   - $rlist = $dbh->resources_list
  Returns - arrayref to a hashref
  Args    - none

Returns a list of resources; each resource is a hash



The above methods rely on you having a file describing all the relational dbs available to you, and setting the env var DBSTAG_DBIMAP_FILE set (this is a : separated list of paths).

This is alpha code - not fully documented, API may change

Currently a resources file is a whitespace delimited text file - XML/Sxpr/IText definitions may be available later

Here is an example of a resources file:

  mytestdb         rdb        Pg:mytestdb                      schema=test
  worldfactbook    rdb  schema=wfb
  employees        rdb      schema=employees

The first column is the nickname or logical name of the resource/db. This nickname can be used instead of the full DBI locator path (eg you can just use employees instead of dbi:Pg:dbname=employees;

The second column is the resource type - rdb is for relational database. You can use the same file to track other system datasources available to you, but DBStag is only interested in relational dbs.

The 3rd column is a way of locating the resource - driver:name@host

The 4th column is a ; separated list of tag=value pairs; the most important tag is the schema tag. Multiple dbs may share the same schema, and hence share SQL Templates


DBStag is usable without writing any perl, you can use command line scripts and files that utilise tree structures (XML, S-Expressions) -d <DSN> [-n <nestexpr>] <SQL>

Queries database and writes decomposed relation as XML

Can also be used with templates: -d <DSN> /<templatename> <var1> <var2> ... <varN> -d <DSN> [-n <nestexpr>] <SQL>

Queries database and writes decomposed relation as HTML with nested tables indicating the nested structures. -d <DSN> <file>

Stores data from a file (Supported formats: XML, Sxpr, IText - see Data::Stag) in a normalized database. Gets it right most of the time.

TODO - metadata help [-l <linktable>]* <file>

Takes data from a file (Supported formats: XML, Sxpr, IText - see Data::Stag) and generates a relational schema in the form of SQL CREATE TABLE statements.



setting this environment will cause all SQL statements to be printed on STDERR, as well as a full trace of how nodes are stored


The SQL parsing can be quite particular - sometimes the SQL can be parsed by the DBMS but not by DBStag. The error messages are not always helpful.

There are probably a few cases the SQL SELECT parsing grammar cannot deal with.

If you want to select from views, you need to hack DBIx::DBSchema (as of v0.21)


Use SQL::Translator to make SQL DDL generation less Pg-specific; also for deducing foreign keys (right now foreign keys are guessed by the name of the column, eg table_id)

Can we cache the grammar so that startup is not so slow?

Improve algorithm so that events are fired rather than building up entire structure in-memory

Tie in all DBI attributes accessible by hash, i.e.: $dbh->{...}

Error handling



Chris Mungall <cjm AT fruitfly DOT org>


Copyright (c) 2004 Chris Mungall

This module is free software. You may distribute this module under the same terms as perl itself