Modify partial-index-predicate applicability tester to test whether

clauses are equal(), before trying to match them up using btree opclass
inference rules.  This allows it to recognize many simple cases involving
non-btree operations, for example 'x IS NULL'.  Clean up code a little.

doc/src/sgml/ref/create_index.sgml

 <!--
-$Header: /cvsroot/pgsql/doc/src/sgml/ref/create_index.sgml,v 1.20 2001/07/16 05:06:57 tgl Exp $
+$Header: /cvsroot/pgsql/doc/src/sgml/ref/create_index.sgml,v 1.21 2001/08/06 18:09:45 tgl Exp $
 Postgres documentation
 -->
 
@@ -256,21 +256,26 @@ ERROR: Cannot create index: 'index_name' already exists.
     billed and unbilled orders where the unbilled orders take up a small
     fraction of the total table and yet that is an often used section, you
     can improve performance by creating an index on just that portion.
+    Another possible application is to use <command>WHERE</command> with
+    <command>UNIQUE</command> to enforce uniqueness over a subset of a
+    table.
   </para>
 
   <para>
     The expression used in the <command>WHERE</command> clause may refer
     only to columns of the underlying table (but it can use all columns,
-    not only the one(s) being indexed).  Currently, the
-    <productname>PostgreSQL</productname> planner can only devise query
-    plans that make use of a partial index when the predicate is built from
-    <command>AND</command> and <command>OR</command> combinations of
-    elements of the form
-    <firstterm>column</firstterm>
-    <firstterm>operator</firstterm>
-    <firstterm>constant</firstterm>.
-    However, more general predicates may still be useful in conjunction
-    with UNIQUE indexes, to enforce uniqueness over a subset of a table.
+    not only the one(s) being indexed).  Presently, sub-SELECTs and
+    aggregate expressions are also forbidden in <command>WHERE</command>.
+  </para>
+
+  <para>
+   All functions and operators used in an index definition must be
+   <firstterm>cachable</>, that is, their results must depend only on
+   their input arguments and never on any outside influence (such as
+   the contents of another table or the current time).  This restriction
+   ensures that the behavior of the index is well-defined.  To use a
+   user-defined function in an index, remember to mark the function cachable
+   when you create it.
   </para>
 
   <para>

src/backend/commands/indexcmds.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/commands/indexcmds.c,v 1.53 2001/07/17 21:53:01 tgl Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/commands/indexcmds.c,v 1.54 2001/08/06 18:09:45 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -146,10 +146,12 @@ DefineIndex(char *heapRelationName,
 	/*
 	 * Convert the partial-index predicate from parsetree form to
 	 * an implicit-AND qual expression, for easier evaluation at runtime.
+	 * While we are at it, we reduce it to a canonical (CNF or DNF) form
+	 * to simplify the task of proving implications.
 	 */
 	if (predicate != NULL && rangetable != NIL)
 	{
-		cnfPred = cnfify((Expr *) copyObject(predicate), true);
+		cnfPred = canonicalize_qual((Expr *) copyObject(predicate), true);
 		fix_opids((Node *) cnfPred);
 		CheckPredicate(cnfPred, rangetable, relationId);
 	}