HOWTO: Squish2SQL by example

This document expands on the notes in http://fireball.danbri.org/people/danbri/2002/02/java/rewriter/doc/inkling-query-rewriter.html with the goal of explaining the Squish2SQL convertor with a worked example. We take an example test1.squish and highlight the connection between the parts of both query (since URIs and literals as mangled into integers, this is especially useful).

Annotated output from Squish2SQL follows. The raw output was generated with:

Annotated Squish query

     SELECT  ?thumb, ?name, ?mbox 
      WHERE 
 
        ### predicate, subject, object ###########################
        ###                                                     ##
  	    (foaf::depiction ?x ?uri)				## a1
  *         (foaf::depiction ?z ?uri)				## a2
            (foaf::mbox ?x mailto:libby.miller@bristol.ac.uk)   ## a3
            (foaf::mbox ?z ?mbox)				## a4
  *         (foaf::name ?z ?name)				## a5
            (foaf::thumbnail ?uri ?thumb)			## a6
  *         (dc::description ?uri ?dd)				## a7

USING	 foaf for http://xmlns.com/foaf/0.1/  
          dc for http://purl.org/dc/elements/1.1/

resources 'b*': variables in order used are ?x, ?uri, ?z, ?mbox,
?name, ?thumb, ?dd. Are the '2', '5' and '7's in related to a2, a5,
a7? assuming not. 

 b1 
 b2: mbox
 b3
 b4
 b5: thumb
 b6: 
 b7: name

Notes: the numbers for a's correspond to the where-clause rows. b2, b5 and b7 are bound to specific resources/literals, and correspond as follows: b2=mbox, b5=thumb, b7=name.

Annotated SQL query

    SELECT DISTINCT 
    b2.value AS mbox, b5.value AS thumb, b7.value AS name  
   FROM   
    triples a1,  triples a2,  triples a3,  triples a4, triples a5, triples a6,  triples a7,  ## rows
    resources b2,  resources b5,  resources b7  ## num variables asked for

   WHERE a1.predicate = '116868652'		## foaf::depiction
   AND   a2.predicate = '116868652'		## foaf::depiction
   AND a3.predicate = '1547507681'              ## foaf::mbox
   AND    a3.object = '1145937192'              ## mailto:libby.miller@bristol.ac.uk
   AND a4.predicate = '1547507681'              ## foaf::mbox
   AND    a5.predicate = '1577895888'           ## foaf::name
   AND a6.predicate = '-1848367484'             ## foaf::thumbnail
   AND a7.predicate = '-221079518'		## dc::description  

   AND a1.subject=a3.subject		## ?x		     a1, a3
   AND a1.object=a2.object              ## ?uri		     a1, a2
   AND a2.object=a6.subject		## ?z		     a2, a6
   AND a6.subject=a7.subject		## ?uri		     a6, a7
   AND a2.subject=a4.subject		## ?z		     a2, a4
   AND a4.subject=a5.subject		## ?z		     a4, a5

   AND b2.key=a4.object			## ?mbox a4
   AND b5.key=a6.object			## ?thumb a6
   AND b7.key=a5.object			## ?name a5


Notes extracted from .java version:

    //this is just for the a* variables
    Hashtable realToSqlVariableNameMapA

     //we need a1.subject=b1.subject - that's what this hash does.
    Hashtable sqlVariableMatchAB;

    sqlVariableNamesB = new Vector();

...when looping through WHERE clauses, each time:

clauses.each { |clause|

  pred_bound = false
  subject_bound = false
  object_bound = false

   allQueryVariables.each { |varname|
     if s.equals(varname) 
	realToSqlVarname_A[varname]= "a"+id_a_clause
       subject_bound = 1
     end
     if p.equals(varname)
       realToSqlVarname_A[varname]= "a"+id+a_clause
     end
     if o.equals(varname)
       realToSqlVarname_A[varname]= "a"+id_a_clause
     end
   }
   sh1_sub=intToHashCode(s)
   sh1_pred=etc
   sh1_pobj= 
}

Conclusions

The conversion algorithm required should be obvious. Enjoy.