| Michael Bouschen 2006-06-22, 7:11 pm |
| Hi,
I agree to what Craig mentioned in the JDO TCK minutes: method
addSubquery separates the outer from the subquery. So it looks like that
this approach cannot support correlated subqueries where the subquery
iterates a collection field defined in the outer query. My preference is
using variables declaration to support subqueries even in the Query API
case.
A query selecting employees earning more than the average salary could be:
SELECT FROM Employee WHERE this.salary > averageSalary
VARIABLES float averageSalary = SELECT AVG(this.salary) FROM Employee
But I still have problems with a correlated subquery. I'm looking for a
query selecting employees earning more than the average salary of their
department. The difference to the query above is that the candidates
collection of the subquery: the employees collection of the department
of the current employee and not the entire Employee extent. Here are
some ideas to stimulate the discussion, but I'm not quite happy with these.
(1) Use a collection variable in the FROM clause of the subquery:
SELECT FROM Employee WHERE this.salary > averageSalary
VARIABLES Collection emps = this.department.employees;
float averageSalary = SELECT AVG(this.salary) FROM emps
One issue is that 'this' is ambiguous: I would think that 'this' always
refers to an instance of the inner most scope. But this means the
subquery cannot directly access a field of the outer query. So the outer
query declares a variable emps that may be used in the inner query.
Maybe we need a special syntax to access the 'this' from the outer query.
(2) The second form does not define any FROM clause, instead it uses a
variable bound to a collection field of the outer query:
SELECT FROM Employee WHERE this.salary > averageSalary
VARIABLES Employee e;
float averageSalary = SELECT AVG(e.salary) WHERE
this.department.employees.contains(e)
Since the subquery does not have a FROM clause, 'this' refers to the
current Employee from the outer query.
BTW, this is the query as Java Persistence API Query (formerly EJB QL):
SELECT e FROM Employee e JOIN e.department d
WHERE e.salary > (SELECT AVG(e2.salary) FROM d.employees e2)
About the open issue: using the assignment operator ("=") versus the
JDOQL keyword "AS": I vote for the assignment operator. The "AS" keyword
is used in the result expression of the form 'expr AS name'. In a
variable declaration it would have the form 'type name AS expr'. It
might be confusing that the variable declaration swaps the order,
because here the expr is right of the keyword, where it is on the other
side in the result expression.
Regards Michael
> Hi everyone,
>
> Here are 2 proposals discussed in the Fri 9 Jun JDO conference call
> regarding support for subqueries in JDOQL, including single-string and Query
> API enhancements, inspired by JPOX's proposed enhancement, documented in
> JPOX JIRA issue CORE-2861
> (http://www.jpox.org/servlet/jira/browse/CORE-2861). The string and API
> enhancement proposals described here are designed to be used hand-in-hand,
> as the folks on the call wanted to continue to provide compatible
> string-based and API-based usages. In a nutshell, both proposals hinge on
> the use of the exising facility to declare JDOQL variables (not JDOQL
> parameters) to bind subqueries to superqueries.
>
> Please read thoroughly, consider, and comment.
>
> --matthew
>
> PS: Martin was on the hook to describe an alternative proposal based on a
> future object pattern. See separate proposal from him.
>
> <proposals>
>
> Query API support
> =================
> Proposal: Introduce new method Query.addSubquery(String variableName, Query
> subquery)
>
> This proposal entails utilizing the current Query API's declareVariables
> facility and adding a method Query.addSubquery(String,Query) to support
> subqueries. Essentially, a subquery is bound to a superquery via a variable
> declared for the superquery. The implementation handles coercing the
> subquery's result into the type of the variable(s) declared in the
> superquery. Queries can be nested to arbitrary levels.
>
> Example 1A: Find people with above average income
>
> Query superquery1a = pm.newQuery("SELECT FROM Person WHERE income >
> averageIncome");
> superquery1a.declareVariables("BigDecimal averageIncome;");
>
> Query subquery1a = pm.newQuery("SELECT avg(income) FROM Person");
> superquery1a.addSubquery("averageIncome", subquery1a); // binds subquery to
> superquery
>
>
> Example 2A: Find average income of fathers using subquery
>
> Query superquery2a = pm.newQuery("SELECT avg(income) FROM fathers");
> // in next line, Collection<Parent> derived from subquery
> superquery2a.declareVariables("Collection fathers;");
>
> Query subquery2a = pm.newQuery("SELECT FROM Person WHERE gender == 'M' &&
> children.size() > 0");
> superquery2a.addSubquery("fathers", subquery2a); // binds subquery to
> superquery
>
>
> Example 3A: Find average income of fathers using a single Query instance
> Note: this example's usage is required if the grammar specification of the
> variables clause remains the same (as it currently is) in the API and string
> forms (see JDO 2.0 spec sections 14.6.5 & 14.6.13).
>
> Query superquery3a = pm.newQuery("SELECT avg(income) FROM fathers");
> // in next line, Collection<Parent> derived from subquery
> superquery3a.declareVariables(
> "Collection fathers = SELECT FROM Person WHERE gender == 'M' &&
> children.size() > 0;");
>
> Pros:
> * Maintains backward compatibility.
> * Enhances performance by allowing for the deferral of query execution
> until entire query with subqueries is defined. Current Query API support
> requires the execution of the subquery, then execution of the superquery;
> current JDOQL string spec doesn't allow for subqueries at all.
> * Grammar of the variables clause undergoes the same enhancements in both
> the single-string and the API.
> * Compatible with single-string enhancement proposal below
>
> Cons:
> * Type coercion becomes more complicated than just autoboxing.
> * Requires that variables may always have to be explicitly defined.
> * Possibility that type of candidate collection of superquery must be
> derived (see example 2A above), or may not be known.
> * Possibility of using variables in place of both parameters and candidate
> collections.
>
>
>
> Single-string proposal
> ======================
> Proposal: Use the existing VARIABLES JDOQL keyword in order to be
> compatible with the Query API proposal above.
>
> This proposal is very similar to JPOX JIRA CORE-2861
> (http://www.jpox.org/servlet/jira/browse/CORE-2861), which proposes the
> introduction of a new JDOQL keyword "WITH" to introduce typed and named
> subquery results. To make this more compatible with the Query API proposal
> above and to avoid the need to introduce a new keyword to JDOQL, the
> existing JDOQL keyword "VARIABLES" would be used to introduce typed and
> named subqueries, except that the variable(s) would be intialized via the
> assignment operator, "=", or the "AS" keyword (TBD) at declaration time with
> a valid JDOQL expression. Variables would continue to be
> semicolon-delimited. Additionally, Query.toString(), for queries that
> employ subqueries, returns JDOQL strings that use this syntax.
>
> Example 1S: Find people with above average income using subquery (similar
> to example 1A)
>
> SELECT FROM Person WHERE income > averageIncome
> VARIABLES float averageIncome =
> SELECT avg(income) FROM Person;
>
> Example 2S: Find average income of fathers using a subquery (similar to
> example 2A)
>
> SELECT avg(income) FROM parents
> VARIABLES Collection parents =
> SELECT FROM Parent WHERE gender == 'M' && children.size() > 0;
>
> Pros:
> * Continues to use existing JDOQL keywords.
> * Grammar of the variables clause undergoes the same enhancements in both
> the single-string and the API.
> * Compatible with Query API proposal above.
>
> Cons:
> * More verbose than introducing "WITH" keyword. Implicit, unnamed
> variables would not be supported, unless the following syntax were supported
> (from example 2S), where the tokens "VARIABLES Collection parents" is
> implied.
> SELECT avg(income) FROM parents =
> SELECT FROM Parent WHERE gender == 'M' && children.size() > 0;
> This syntax is admittedly less verbose and more like SQL subqueries, but
> leaves open the typing and naming of the implicit, unnamed variables.
> * Possibility of using variables in place of both parameters and candidate
> collections.
>
>
> Open issues
> ===========
>
> * Can type derivation & coercion of JDOQL variables be performed in all
> cases?
> * These proposals use the assignment operator ("="). Should we use
> assignment via the JDOQL keyword "AS" instead or in addition to the
> assignment operator?
> * This proposal requires that JDOQL variables be allowed to substitute for
> both JDOQL parameters and candidate collections. Should this be allowed?
>
>
> </proposals>
>
>
>
>
--
Michael Bouschen Tech@Spree Engineering GmbH
mailto:mbo.tech@spree.de http://www.tech.spree.de/
Tel.:++49/30/235 520-33 Buelowstr. 66
Fax.:++49/30/2175 2012 D-10783 Berlin
|