15.3.3.XML Views - Representing SQL Data as Dynamic and Persistent XML

The XML view mechanism allows generating XML content from relational data and to query relational data as if it were XML without first converting it to XML.

CREATE XML VIEW statement

<xml view> ::=
     CREATE XML VIEW <name> as [ <namespaces_def> ] <element list> <opt_public>

<xml view> ::=
     CREATE XML VIEW <name> as [ <namespaces_def> ] <query spec> [ELEMENT] <opt_public>

<namespaces_def> ::= '[' <id_namespace> '='( name | <path string> ) [ ...]  ']'

<element list> ::= <element> [, ...]*

<element> ::=
         <table> <correlation name> AS [ <id_namespace> ':' ] <element> <columns>
         [ ON '('  <search condition> ')']
        [primary key '(' column_commalist ')' ]
        [ELEMENT]
         [ '{' <element list>'}' ]

<opt_public> ::=  PUBLIC <path string> OWNER <DAV owner name> [PERSISTENT] [INTERVAL <minutes>]

<columns> ::= '(' <column> [, ...] ')'

<column> ::=   <column name> | <column name> AS [ <id_namespace> ':' ] <attribute name>

<id_namespace> :: = identifier

The XML view declaration establishes a 'virtual document' a context within which XML hierarchy relationships can be translated into arbitrary joins. The virtual document can be then materialized into an actual set of persistent XML elements or used to generate SQL from XPATH.

Each table in the declaration generates an element into the result document. SQL views can be used as tables to accommodate for hidden joins, sub-queries, ordering and aggregates. If a view is used, which by nature has no primary key, the primary key clause should be used to define a uniquely identifying set of view columns.

The only restriction on the XML view declaration is that each branch has a fixed depth.

The structure of joins used to make the text can be specified in two ways: As a SQL query specification, that is a SELECT from a list of tables with a WHERE clause specifying the joins, or as a tree of join elements. The first form is called automatic and the second is called explicit. The automatic form allows generating a tree with as many levels as there are tables in the join, with elements derived from the rows at each level occupying each level of the hierarchy

With both forms the columns of the tables are mapped into either attributes or child elements of the element representing each row. In the explicit mode, the attribute / element choice can be made for each table, in the automatic mode for the entire view. The explicit mode also allows specifying a different element / attribute name whereas the automatic mode takes the name from the column name. Even if the columns are presented as child elements in the output text, they should be referenced as attributes in XPATH queries evaluated in the context of the view.

An XML namespaces can be used in XML view with two restrictions - all namespace names must be different and the first three letters of namespace name must not be 'xml', except for default namespace with name 'xmlns'.

Explicit XML Views

In the explicit form each level of the hierarchy is declared as a list of child elements. Each such element maps one table or view into an entity according to a join condition. The join conditions can reference columns from the associated table and columns from tables in parent elements. The join condition can also have scalar filtering conditions. A top element's join condition may only specify scalar conditions.

Each set of sibling child nodes is delimited by braces {}. The top level of the view typically consists of one element in the outermost braces. This element has itself a child list delimited by braces. Each such list can have more than one different element.

Each element specifies:

  • SQL table

  • Correlation name for use in subsequent joins for this table

  • XML element name to use for delimiting a row of this table

  • List of columns, with optional XML element or attribute names

  • join condition - will relate rows of this table to rows of the table in the enclosing element. If this element is at the top level, this can only consist of scalar conditions

  • Optional PRIMARY KEY clause, needed if the table in this element is a view, does not have a primary key or if a non-primary key unique identity is desired

  • Optional ELEMENT flag

  • Optional list of child elements, delimited by braces

[Note] Note:

A correlation name is mandatory for all the tables.

The column list can mention a single column or a single column renamed into an XML attribute of a different name. If a column of a table is referenced in a subsequent join condition it must appear in the output columns list. Expressions are not directly allowed but a view with expression columns can be used.

The opt_public clause, when present, offers a shorthand for calling xml_view_publish at the same time as making the definition. This makes a DAV resource reflecting the contents of the view. The contents may either be generated on demand or persisted as a DAV accessible XML document. In the latter case the document may be regenerated at a fixed interval. The interval is expressed in minutes.

The path is expressed as an absolute path from the root collection of the DAV server.

[Note] Note:

This root collection may be mapped into various places in the web server's URL space.

create xml view xx ... public '/xx.xml' owner 'dav' persistent interval 1;

is equivalent to:

create xml view xx ...;
xml_view_publish (xx, /xx.xml', 'dav', 1, 1);

A DAV resource created in this manner can be deleted as any DAV resource. The XML view itself is not affected but a possibly existing refresh job will be automatically deleted.

One XML view can be published several times with different names and owners. There may also exist persistent and non-persistent publications of the same view.

The CREATE XML VIEW statement defines stored procedures for generating an XML text fragment corresponding to each element declared in the view.

The names of the procedures are composed as follows:

create procedure http_view_<view name>
        (inout output_mode)
create procedure http_<view name>_<element name>_<correlation name>
        (in pk1 any, ..., in output_mode integer)

An http output procedure is created for each <element> in the create xml view declaration. It takes the primary key columns of the table in question in key order plus a mode flag. It then outputs the serialization of the specified element and any child elements. For an output mode of 0 the result goes directly to an HTTP client. For an output_mode of 1 the procedure returns the serialization as a string. Note that for this to work the tables in question must be real tables and the join conditions must only reference the next higher table in the create xml view tree. Further, the primary key columns of each table should be mentioned in the columns list for that table along with any foreign keys referenced in subsequent join conditions.

Example15.4.Examples

create xml view "cat" as
{
  "Demo"."demo"."Categories" "C" as "category"
        ("CategoryID", "Description" as "description")
    {
      "Demo"."demo"."Products" "P" as "product"  ("ProductName")
        on ("P"."CategoryID" = "C"."CategoryID")
    }
}

This declares a two level hierarchy with a category node for each category and a product child node for each product in the category.

create xml view "cats_e" as
  select "category"."CategoryID", "CategoryName",
    "ProductName", "ProductID"
    from "Demo".."Categories" "category", "Demo".."Products" as "product"
    where "product"."CategoryID" = "category"."CategoryID" element;

Here is a similar example, this time using the element option.


The procedures are

xmlg_cat
http_cat_category_C (in categoryid any, in _out integer);
http_cat_product_P (in productid any, in _out integer);

In the following example the function returns the selected items as an XML fragment. Consecutive elements are separated by new-lines for readability.

Example15.5.Example

SQL> call "http_cat_category_C" (1, 1);

1 sets? Done. -- 5 msec.
RESULT=
<category CategoryID="1" description="Soft drinks, coffees, teas, beers, and ales" >
<product ProductName="Chai" ></product>
<product ProductName="Chang" ></product>
<product ProductName="Guarana Fantastica" ></product>
<product ProductName="Sasquatch Ale" ></product>
<product ProductName="Steeleye Stout" ></product>
<product ProductName="Côte de Blaye" ></product>
<product ProductName="Chartreuse verte" ></product>
<product ProductName="Ipoh Coffee" ></product>
<product ProductName="Laughing Lumberjack Lager" ></product>
<product ProductName="Outback Lager" ></product>
<product ProductName="Rhönbrooou Klosterbier" ></product>
<product ProductName="Lakkalikööri" ></product>
</category>

The below example shows how to use a SQL view for hiding a join. The below view generates for each table a set of column children and a set of index children, which in turn have column children.

create view KEY_COLS as select KP_KEY_ID, KP_NTH, C.*
        from SYS_KEY_PARTS, SYS_COLS C where COL_ID = KP_COL;

create xml view "schema" as
{
  DB.DBA.SYS_KEYS k as "table" ("KEY_TABLE" as "name",
        KEY_ID as "key_id", KEY_TABLE as "table")
        on (k.KEY_IS_MAIN = 1 and k.KEY_MIGRATE_TO is null)
        { DB.DBA.KEY_COLS  c as "column" (COLUMN as name)
                on (k.KEY_ID = c.KP_KEY_ID)
                primary key (COL_ID),
        DB.DBA.SYS_KEYS i as "index" (KEY_NAME
          as "name", KEY_ID as "key_id", KEY_N_SIGNIFICANT as "n_parts")
          on (i.KEY_TABLE = k.KEY_TABLE and i.KEY_IS_MAIN = 0 and i.KEY_MIGRATE_TO is null)
          {
            DB.DBA.KEY_COLS ic as "column" (COLUMN as "name")
              on (ic.KP_NTH < i.KEY_N_SIGNIFICANT and ic.KP_KEY_ID = i.KEY_ID)
              primary key (COL_ID)
              }
        }
};

The following query will return the subtree describing the Customers table in the demo database:

XPATH [__view 'schema']
        /table[@name = 'Demo.demo.Customers'];

<table name="0" key_id="1011" table="Demo.demo.Customers" >
<column name="CustomerID" ></column>
<column name="CompanyName" ></column>
<column name="ContactName" ></column>
<column name="ContactTitle" ></column>
<column name="Address" ></column>
<column name="City" ></column>
<column name="Region" ></column>
<column name="PostalCode" ></column>
<column name="Country" ></column>
<column name="Phone" ></column>
<column name="Fax" ></column>
<index name="City" key_id="1012" n_parts="2" >
<column name="City" ></column>
<column name="CustomerID" ></column>
</index>
<index name="CompanyName2" key_id="1013" n_parts="2" >
<column name="CompanyName" ></column>
<column name="CustomerID" ></column>
</index>
<index name="PostalCode2" key_id="1014" n_parts="2" >
<column name="PostalCode" ></column>
<column name="CustomerID" ></column>
</index>
<index name="Region" key_id="1015" n_parts="2" >
<column name="Region" ></column>
<column name="CustomerID" ></column>
</index>
</table>

Automatic XML Views - Creating XML Views from SELECT Statements

The automatic form of CREATE XML VIEW will take a select statement and infer a hierarchy from it, based on the order of tables in the from clause. The parent table should be to the left of its children. This is practical if the tables form a hierarchy in application terms, like orders and order lines or departments and employees. This notation allows arbitrary depth but all siblings at the same level will be of the same type. Elements of child rows will be child elements of the element of their parent row, where the join condition identifies the child rows for one parent row.

The columns in the selection will appear as attributes or child elements of the rows selected. The names of the attributes will be the names of the columns. The names of the siblings will be the names of the tables in the from clause, without qualifiers or owners. Expressions should not appear in the selection. If the use of expressions is required then you may create a SQL view first to facilitate this.

The ELEMENT keyword may be present at the end of the select, before the publishing keywords. This will cause all columns to be represented as child elements of the element corresponding to the row. Note that even if the element switch is present, the values will appear like attributes in an XPATH query inside the view.

[Note] Note

SQL views or derived tables may not appear directly in the select. The reason for this is that a procedure is generated for each level of the generated XML tree and that this must take unique identifying column values for the element in question. If one desires to use a view, the explicit form should be used, with the primary key option specified where appropriate.

Add to text for explicit XML views

Each set of sibling child nodes is delimited by braces {}. The top level of the view typically consists of one element in the outermost braces. This element has itself a child list delimited by braces. Each such list can have more than one different element.

Each element specifies:

  • SQL table

  • Correlation name for use in subsequent joins for this table

  • XML element name to use for delimiting a row of this table

  • List of columns, with optional XML element or attribute names

  • join condition - will relate rows of this table to rows of the table in the enclosing element. If this element is at the top level, this can only consist of scalar conditions

  • Optional PRIMARY KEY clause, needed if the table in this element is a view, does not have a primary key or if a non-primary key unique identity is desired

  • Optional ELEMENT flag

  • Optional list of child elements, delimited by braces

Example15.6.Add to examples:

create xml view "cats_e" as
  select "category"."CategoryID", "CategoryName",
    "ProductName", "ProductID"
    from "Demo".."Categories" "category", "Demo".."Products" as "product"
    where "product"."CategoryID" = "category"."CategoryID" element;

Add to text: after 'free text and xml'

xml_view_publish

DB.DBA.xml_view_publish (in view_name varchar, in dav_path varchar,
        in dav_owner varchar, in is_persistent integer, in refresh_interval integer)

This presents an XML view as a DAV resource. The view name is the name in the create xml view statement, note that this is case sensitive and is never converted since it is a string, not an identifier. The path must be absolute and is interpreted as relative from the DAV root collection. The DAV user is the owner of the resource. If is_persistent is non-zero the resource will be materialized from the view's description. The refresh interval is only applicable if the resource is materialized. If so, this is an interval in minutes. A value of 0 means no automatic refresh.

The reverse operation of xml_view_publish is deleting the DAV resource. xml_view_publish may be called several times to alter the owner or refresh interval.