Thursday, February 11, 2010

REF Cursor to JSON

REF Cursors are cool. They allow you to encapsulate SQL queries behind a PL/SQL package API. For example, you can create a function called GET_EMPLOYEES that returns a SYS_REFCURSOR containing the employees in a specific department:

function get_employees (p_deptno in number) return sys_refcursor
l_returnvalue sys_refcursor;

open l_returnvalue
select empno, ename, job, sal
from emp
where deptno = p_deptno;

return l_returnvalue;

end get_employees;

The client (an application written in Java, .NET, PHP, etc.) can call your API and process the returned REF Cursor just as if it was a normal result set from a SQL query. The benefits are legion. The client no longer needs to contain embedded SQL statements, or indeed know anything about the actual database structure and query text. Privileges on the underlying tables can be revoked. The API can be shared and reused among different clients, whether they are written in Java, .NET, or any number of other languages.

That is, unless your client is Oracle Application Express (Apex). Apex unfortunately lacks the ability to process REF Cursors, or, more accurately, you cannot create report regions in Apex based on REF Cursors. For standard reports, you have to either embed the SQL statement in the region definition, or return the SQL text string from a function (and hope that the string you built is valid SQL when it gets executed). For interactive reports, only embedded SQL statements are supported.

I dislike having to scatter literal SQL statements all around my Apex applications, and not be able to take advantage of a package-based, shared and reusable PL/SQL API to encapsulate queries. I submitted a feature request to the Apex team back in 2007, asking for the ability to base report regions on REF Cursors, but so far this has not been implemented.

The problem, as far as I know, is that Apex uses (and must use) DBMS_SQL to "describe" a SQL statement in order to get the metadata (column names, data types, etc.) for a report region. But not until Oracle 11g did DBMS_SQL include a function (TO_CURSOR_NUMBER) that allows you to convert a REF Cursor into a DBMS_SQL cursor handle. So, as long as the minimum supported database version for Apex is Oracle 10g, support for REF Cursors is unlikely to be implemented.

In the meantime, there are a couple of alternatives:

Option 1: Pipelined functions

It's possible to encapsulate your queries behind a PL/SQL API by using pipelined functions. For example, the above example could be rewritten as...

create type t_employee as object (
empno number(4),
ename varchar2(10),
job varchar2(9),
sal number

create type t_employee_tab as table of t_employee;

function get_employees (p_deptno in number) return t_employee_tab pipelined

for l_rec in (select empno, ename, job, sal from emp where deptno = p_deptno) loop
pipe row (t_employee (l_rec.empno, l_rec.ename, l_rec.job, l_rec.sal));
end loop;


end get_employees;

And used from Apex (in a report region) via the TABLE statement:
select *
from table(employee_pkg.get_employees (:p1_deptno))

Option 2: XML from REF Cursor

The DBMS_XMLGEN package can generate XML based on a REF Cursor. While this does not "describe" the REF Cursor per se, it does give us a way (from PL/SQL) to find the column names of an arbitrary REF Cursor query, and perhaps infer the data types from the data itself. A couple of blog posts from Tom Kyte explain how this can be used to generate HTML based on a REF Cursor.

So back to Apex, you could generate a "report" based on a PL/SQL region with code similar to this:

l_clob clob;
l_rc   sys_refcursor;
l_rc := get_employees (:p1_deptno);
l_clob := fncRefCursor2HTML (l_rc);
htp_print_clob (l_clob);

It would also be possible to pass your own XLST stylesheet into the conversion function (perhaps an Apex report region template fetched from the Apex data dictionary?) to control the appearance of the report.

I put "report" in quotes above, because until the Apex team implements report regions based on REF Cursors, you will miss all the nice built-in features of standard (and interactive) reports, such as sorting, paging, column formatting, linking, etc.

Option 3: JSON from REF Cursor

Bear with me, I am finally getting to the point of this blog post.

JSON is cool, too, just like REF Cursors. It's the fat-free alternative to XML, and JSON data is really easy to work with in Javascript.

For triple coolness, I want to use an API based on REF Cursors in PL/SQL, client-side data manipulation based on JSON, and Apex to glue the two together.

What I need is the ability to generate JSON based on a REF Cursor.

Apex does include a few JSON-related procedures in the APEX_UTIL package, including JSON_FROM_SQL. Although this procedure does support bind variables, it cannot generate JSON from a REF Cursor. (Also, the fact that is is a procedure rather than a function makes it less flexible than it could be. Dear Apex Team, can we please have overloaded (function) versions of these JSON procedures?)

REF Cursor to JSON: The (10g) solution

So I came up with this solution: Use DBMS_XMLGEN to generate XML based on a REF Cursor, and then transform the XML into JSON by using an XSLT stylesheet.

Note: As mentioned above, in Oracle 11g you can use DBMS_SQL to describe a REF Cursor, so you could write your own function to generate JSON from a REF Cursor, without going through XML first. (And perhaps in Oracle 12g the powers that be at Redwood Shores will provide us with a built-in DBMS_JSON package that can both generate and parse JSON?)

In the meantime, for Oracle 10g, I created the JSON_UTIL_PKG package.

Here is the code for the REF_CURSOR_TO_JSON function:

function ref_cursor_to_json (p_ref_cursor in sys_refcursor,
p_max_rows in number := null,
p_skip_rows in number := null) return clob
l_ctx         dbms_xmlgen.ctxhandle;
l_num_rows    pls_integer;
l_xml         xmltype;
l_json        xmltype;
l_returnvalue clob;


Purpose:    generate JSON from REF Cursor


Who     Date        Description
------  ----------  -------------------------------------
MBR     30.01.2010  Created


l_ctx := dbms_xmlgen.newcontext (p_ref_cursor);

dbms_xmlgen.setnullhandling (l_ctx, dbms_xmlgen.empty_tag);

-- for pagination
if p_max_rows is not null then
dbms_xmlgen.setmaxrows (l_ctx, p_max_rows);
end if;

if p_skip_rows is not null then
dbms_xmlgen.setskiprows (l_ctx, p_skip_rows);
end if;

-- get the XML content
l_xml := dbms_xmlgen.getxmltype (l_ctx, dbms_xmlgen.none);

l_num_rows := dbms_xmlgen.getnumrowsprocessed (l_ctx);

dbms_xmlgen.closecontext (l_ctx);

close p_ref_cursor;

if l_num_rows > 0 then
-- perform the XSL transformation
l_json := l_xml.transform (xmltype(get_xml_to_json_stylesheet));
l_returnvalue := l_json.getclobval();
l_returnvalue := g_json_null_object;
end if;

l_returnvalue := dbms_xmlgen.convert (l_returnvalue, dbms_xmlgen.entity_decode);

return l_returnvalue;

end ref_cursor_to_json;

Examples of usage

Get a small dataset

l_clob clob;
l_cursor sys_refcursor;
l_cursor := employee_pkg.get_employees (10);
l_clob := json_util_pkg.ref_cursor_to_json (l_cursor);
dbms_output.put_line (substr(l_clob, 1, 200));


A large dataset, with paging

l_clob clob;
l_cursor sys_refcursor;
l_cursor := test_pkg.get_all_objects;
l_clob := json_util_pkg.ref_cursor_to_json (l_cursor, p_max_rows => 3, p_skip_rows => 5000);
dbms_output.put_line (substr(l_clob, 1, 1000));


It works with nested datasets, too.. !

select d.deptno, d.dname,
cursor (select e.*
from emp e
where e.deptno = d.deptno) as the_emps
from dept d

l_json clob;
l_json := json_util_pkg.sql_to_json ('select d.deptno, d.dname,
cursor (select e.*
from emp e
where e.deptno = d.deptno) as the_emps
from dept d');
dbms_output.put_line (substr(l_json, 1, 10000));


Passing a REF Cursor directly to the function call by using the CURSOR function:

select json_util_pkg.ref_cursor_to_json(cursor(select * from emp))
from dual


Download the package

You can download the complete package, including the XSLT stylsheet, here (spec) and here (body).

Update 12.02.2011: This package can now be downloaded as part of the Alexandria library for PL/SQL.

Note that to compile the packages you need the following SQL type defined in your schema:

create type t_str_array as table of varchar2(4000);

Thursday, February 4, 2010

My first Apex 4 plugin: Flight Info from Web Service

One of the exciting new features in Apex 4 is the support for plugin regions and items. This feature has huge potential, and will make development with Apex even more efficient, productive, and fun. There are already several plugins out there, and I think we will see a lot of interesting work in this area after Apex 4 is released.

Here is my own first attempt at a (useful) plugin: A region plugin that displays up-to-date flight information for airports in Norway, based on public flight data provided by Avinor, the company that operates the Norwegian airport network.

Avinor has a simple web service that provides flight information in XML format.

I am sure there are similar (web) services for flight information in other countries (feel free to leave a comment below if you know of any).

Here is the PL/SQL code behind the plugin:

procedure render_my_plugin (
p_region              in apex_plugin.t_region,
p_plugin              in apex_plugin.t_plugin,
p_is_printer_friendly in boolean )
l_clob clob;
l_airport_code varchar2(20) := p_region.attribute_01;
l_direction    varchar2(20) := p_region.attribute_02;

l_clob := apex_web_service.make_rest_request(
p_url => '',
p_http_method => 'GET',
p_parm_name => apex_util.string_to_table('airport:direction'),
p_parm_value => apex_util.string_to_table(l_airport_code || ':' || l_direction )

if l_direction = 'D' then
htp.p('<p><b>Departures from ' || l_airport_code || '</b></p>');
htp.p('<p><b>Arrivals to ' || l_airport_code || '</b></p>');
end if;

htp.p('<table width="100%">');

for l_rec in (
FROM XMLTABLE ('//airport/flights/flight'
COLUMNS unique_id       varchar2(100) path '@uniqueID',
airline         varchar2(10) path 'airline',
flight_id       varchar2(20) path 'flight_id',
airport         varchar2(20) path 'airport',
schedule_time   varchar2(100) path 'schedule_time',
gate            varchar2(100) path 'gate')
ORDER BY airline, flight_id) loop

htp.p('<tr><td>' || l_rec.airline || '</td><td>' || l_rec.flight_id || '</td><td>' || l_rec.airport || '</td><td>' || l_rec.schedule_time || '</td><td>' || l_rec.gate || '</td></tr>');

end loop;


htp.p('<a href="">Flight data from Avinor.</a> Last updated: ' || to_char(sysdate, ' hh24:mi:ss'));

end render_my_plugin;

The code illustrates several concepts:
  • How to render a region plugin using the PL/SQL Web Toolkit (HTP.P) calls
  • How to retrieve values from the attributes defined for the plugin
  • Using the new APEX_WEB_SERVICE.MAKE_REST_REQUEST function to retrieve a web page as a CLOB
  • Using the XMLTABLE function to transform XML into a recordset that can be used in a SELECT

An export of my plugin can be downloaded here, and installed into your own Apex 4 application.

After the plugin has been installed, using the plugin is as simple as adding a Region (of type Plugin) to the page, and configuring the values for Airport and Direction (the plugin attributes) in the region definition.

You can see a live demo of the plugin here (public page, does not require authentication):

Note that for this page, I've also taken advantage of the built-in region caching feature of Apex. The region cache duration is set to 10 minutes, which prevents us from hitting the remote web service for every page view. I really like that you can switch on region caching in Apex without writing a single line of code.

Conclusion: Apex 4 plugins rock!