Refactor from Heikki Linnakangas <heikki@enterprisedb.com>:

* Defined new struct WordEntryPosVector that holds a uint16 length and a
variable size array of WordEntries. This replaces the previous
convention of a variable size uint16 array, with the first element
implying the length. WordEntryPosVector has the same layout in memory,
but is more readable in source code. The POSDATAPTR and POSDATALEN
macros are still used, though it would now be more readable to access
the fields in WordEntryPosVector directly.

* Removed needfree field from DocRepresentation. It was always set to false.

* Miscellaneous other commenting and refactoring

src/backend/utils/adt/tsginidx.c

@@ -7,7 +7,7 @@
  *
  *
  * IDENTIFICATION
- *	  $PostgreSQL: pgsql/src/backend/utils/adt/tsginidx.c,v 1.3 2007/09/07 16:03:40 teodor Exp $
+ *	  $PostgreSQL: pgsql/src/backend/utils/adt/tsginidx.c,v 1.4 2007/09/11 08:46:29 teodor Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -25,13 +25,12 @@ gin_extract_tsvector(PG_FUNCTION_ARGS)
 	int32	   *nentries = (int32 *) PG_GETARG_POINTER(1);
 	Datum	   *entries = NULL;
 
-	*nentries = 0;
+	*nentries = vector->size;
 	if (vector->size > 0)
 	{
 		int			i;
 		WordEntry  *we = ARRPTR(vector);
 
-		*nentries = (uint32) vector->size;
 		entries = (Datum *) palloc(sizeof(Datum) * vector->size);
 
 		for (i = 0; i < vector->size; i++)
@@ -134,11 +133,19 @@ gin_ts_consistent(PG_FUNCTION_ARGS)
 
 	if (query->size > 0)
 	{
-		int4		i,
+		int			i,
 					j = 0;
 		QueryItem  *item;
 		GinChkVal	gcv;
 
+		/*
+		 * check-parameter array has one entry for each value (operand) in the
+		 * query. We expand that array into mapped_check, so that there's one
+		 * entry in mapped_check for every node in the query, including 
+		 * operators, to allow quick lookups in checkcondition_gin. Only the 
+		 * entries corresponding operands are actually used.
+		 */
+
 		gcv.frst = item = GETQUERY(query);
 		gcv.mapped_check = (bool *) palloc(sizeof(bool) * query->size);
 

src/backend/utils/adt/tsgistidx.c

@@ -7,7 +7,7 @@
  *
  *
  * IDENTIFICATION
- *	  $PostgreSQL: pgsql/src/backend/utils/adt/tsgistidx.c,v 1.3 2007/09/07 15:09:56 teodor Exp $
+ *	  $PostgreSQL: pgsql/src/backend/utils/adt/tsgistidx.c,v 1.4 2007/09/11 08:46:29 teodor Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -133,20 +133,27 @@ gtsvectorout(PG_FUNCTION_ARGS)
 }
 
 static int
-compareint(const void *a, const void *b)
+compareint(const void *va, const void *vb)
 {
-	if (*((int4 *) a) == *((int4 *) b))
+	int4 a = *((int4 *) va);
+	int4 b = *((int4 *) vb);
+
+	if (a == b)
 		return 0;
-	return (*((int4 *) a) > *((int4 *) b)) ? 1 : -1;
+	return (a > b) ? 1 : -1;
 }
 
+/*
+ * Removes duplicates from an array of int4. 'l' is
+ * size of the input array. Returns the new size of the array.
+ */
 static int
 uniqueint(int4 *a, int4 l)
 {
 	int4	   *ptr,
 			   *res;
 
-	if (l == 1)
+	if (l <= 1)
 		return l;
 
 	ptr = res = a;
@@ -570,12 +577,15 @@ typedef struct
 } SPLITCOST;
 
 static int
-comparecost(const void *a, const void *b)
+comparecost(const void *va, const void *vb)
 {
-	if (((SPLITCOST *) a)->cost == ((SPLITCOST *) b)->cost)
+	SPLITCOST *a = (SPLITCOST *) va;
+	SPLITCOST *b = (SPLITCOST *) vb;
+
+	if (a->cost == b->cost)
 		return 0;
 	else
-		return (((SPLITCOST *) a)->cost > ((SPLITCOST *) b)->cost) ? 1 : -1;
+		return (a->cost > b->cost) ? 1 : -1;
 }
 
 

src/backend/utils/adt/tsrank.c

@@ -7,7 +7,7 @@
  *
  *
  * IDENTIFICATION
- *	  $PostgreSQL: pgsql/src/backend/utils/adt/tsrank.c,v 1.4 2007/09/07 16:03:40 teodor Exp $
+ *	  $PostgreSQL: pgsql/src/backend/utils/adt/tsrank.c,v 1.5 2007/09/11 08:46:29 teodor Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -53,22 +53,24 @@ cnt_length(TSVector t)
 {
 	WordEntry  *ptr = ARRPTR(t),
 			   *end = (WordEntry *) STRPTR(t);
-	int			len = 0,
-				clen;
+	int			len = 0;
 
 	while (ptr < end)
 	{
-		if ((clen = POSDATALEN(t, ptr)) == 0)
+		int clen = POSDATALEN(t, ptr);
+
+		if (clen == 0)
 			len += 1;
 		else
 			len += clen;
+
 		ptr++;
 	}
 
 	return len;
 }
 
-static int4
+static int
 WordECompareQueryItem(char *eval, char *qval, WordEntry *ptr, QueryOperand *item)
 {
 	if (ptr->len == item->length)
@@ -80,6 +82,10 @@ WordECompareQueryItem(char *eval, char *qval, WordEntry *ptr, QueryOperand *item
 	return (ptr->len > item->length) ? 1 : -1;
 }
 
+/*
+ * Returns a pointer to a WordEntry corresponding 'item' from tsvector 't'. 'q'
+ * is the TSQuery containing 'item'. Returns NULL if not found.
+ */
 static WordEntry *
 find_wordentry(TSVector t, TSQuery q, QueryOperand *item)
 {
@@ -178,15 +184,15 @@ SortAndUniqItems(TSQuery q, int *size)
 }
 
 /* A dummy WordEntryPos array to use when haspos is false */
-static WordEntryPos POSNULL[] = {
+static WordEntryPosVector POSNULL = {
 	1, /* Number of elements that follow */
-	0
+	{ 0 }
 };
 
 static float
 calc_rank_and(float *w, TSVector t, TSQuery q)
 {
-	uint16	  **pos;
+	WordEntryPosVector	 **pos;
 	int			i,
 				k,
 				l,
@@ -207,9 +213,8 @@ calc_rank_and(float *w, TSVector t, TSQuery q)
 		pfree(item);
 		return calc_rank_or(w, t, q);
 	}
-	pos = (uint16 **) palloc(sizeof(uint16 *) * q->size);
-	memset(pos, 0, sizeof(uint16 *) * q->size);
-	WEP_SETPOS(POSNULL[1], MAXENTRYPOS - 1);
+	pos = (WordEntryPosVector **) palloc0(sizeof(WordEntryPosVector *) * q->size);
+	WEP_SETPOS(POSNULL.pos[0], MAXENTRYPOS - 1);
 
 	for (i = 0; i < size; i++)
 	{
@@ -218,25 +223,25 @@ calc_rank_and(float *w, TSVector t, TSQuery q)
 			continue;
 
 		if (entry->haspos)
-			pos[i] = (uint16 *) _POSDATAPTR(t, entry);
+			pos[i] = _POSVECPTR(t, entry);
 		else
-			pos[i] = (uint16 *) POSNULL;
+			pos[i] = &POSNULL;
 
 
-		dimt = *(uint16 *) (pos[i]);
-		post = (WordEntryPos *) (pos[i] + 1);
+		dimt = pos[i]->npos;
+		post = pos[i]->pos;
 		for (k = 0; k < i; k++)
 		{
 			if (!pos[k])
 				continue;
-			lenct = *(uint16 *) (pos[k]);
-			ct = (WordEntryPos *) (pos[k] + 1);
+			lenct = pos[k]->npos;
+			ct = pos[k]->pos;
 			for (l = 0; l < dimt; l++)
 			{
 				for (p = 0; p < lenct; p++)
 				{
 					dist = Abs((int) WEP_GETPOS(post[l]) - (int) WEP_GETPOS(ct[p]));
-					if (dist || (dist == 0 && (pos[i] == (uint16 *) POSNULL || pos[k] == (uint16 *) POSNULL)))
+					if (dist || (dist == 0 && (pos[i] == &POSNULL || pos[k] == &POSNULL)))
 					{
 						float		curw;
 
@@ -285,8 +290,8 @@ calc_rank_or(float *w, TSVector t, TSQuery q)
 		}
 		else
 		{
-			dimt = *(uint16 *) POSNULL;
-			post = POSNULL + 1;
+			dimt = POSNULL.npos;
+			post = POSNULL.pos;
 		}
 
 		resj = 0.0;
@@ -456,17 +461,19 @@ typedef struct
 {
 	QueryItem **item;
 	int16		nitem;
-	bool		needfree;
 	uint8		wclass;
 	int32		pos;
 } DocRepresentation;
 
 static int
-compareDocR(const void *a, const void *b)
+compareDocR(const void *va, const void *vb)
 {
-	if (((DocRepresentation *) a)->pos == ((DocRepresentation *) b)->pos)
+	DocRepresentation *a = (DocRepresentation *) va;
+	DocRepresentation *b = (DocRepresentation *) vb;
+
+	if (a->pos == b->pos)
 		return 0;
-	return (((DocRepresentation *) a)->pos > ((DocRepresentation *) b)->pos) ? 1 : -1;
+	return (a->pos > b->pos) ? 1 : -1;
 }
 
 static bool
@@ -547,11 +554,11 @@ Cover(DocRepresentation *doc, int len, TSQuery query, Extention *ext)
 
 	ptr = doc + lastpos;
 
-	/* find lower bound of cover from founded upper bound, move down */
+	/* find lower bound of cover from found upper bound, move down */
 	while (ptr >= doc + ext->pos)
 	{
 		for (i = 0; i < ptr->nitem; i++)
-			if(ptr->item[i]->type  == QI_VAL) /* XXX */
+			if(ptr->item[i]->type  == QI_VAL)
 				ptr->item[i]->operand.istrue = 1;
 		if (TS_execute(GETQUERY(query), NULL, true, checkcondition_QueryOperand))
 		{
@@ -620,8 +627,8 @@ get_docrep(TSVector txt, TSQuery query, int *doclen)
 		}
 		else
 		{
-			dimt = *(uint16 *) POSNULL;
-			post = POSNULL + 1;
+			dimt = POSNULL.npos;
+			post = POSNULL.pos;
 		}
 
 		while (cur + dimt >= len)
@@ -636,7 +643,6 @@ get_docrep(TSVector txt, TSQuery query, int *doclen)
 			{
 				int			k;
 
-				doc[cur].needfree = false;
 				doc[cur].nitem = 0;
 				doc[cur].item = (QueryItem **) palloc(sizeof(QueryItem *) * query->size);
 
@@ -658,7 +664,6 @@ get_docrep(TSVector txt, TSQuery query, int *doclen)
 			}
 			else
 			{
-				doc[cur].needfree = false;
 				doc[cur].nitem = doc[cur - 1].nitem;
 				doc[cur].item = doc[cur - 1].item;
 			}
@@ -764,9 +769,6 @@ calc_rank_cd(float4 *arrdata, TSVector txt, TSQuery query, int method)
 	if ((method & RANK_NORM_LOGUNIQ) && txt->size > 0)
 		Wdoc /= log((double) (txt->size + 1)) / log(2.0);
 
-	for (i = 0; i < doclen; i++)
-		if (doc[i].needfree)
-			pfree(doc[i].item);
 	pfree(doc);
 
 	return (float4) Wdoc;

src/backend/utils/adt/tsvector_op.c

@@ -7,7 +7,7 @@
  *
  *
  * IDENTIFICATION
- *	  $PostgreSQL: pgsql/src/backend/utils/adt/tsvector_op.c,v 1.4 2007/09/07 16:03:40 teodor Exp $
+ *	  $PostgreSQL: pgsql/src/backend/utils/adt/tsvector_op.c,v 1.5 2007/09/11 08:46:29 teodor Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -269,7 +269,7 @@ compareEntry(char *ptra, WordEntry * a, char *ptrb, WordEntry * b)
 static int4
 add_pos(TSVector src, WordEntry * srcptr, TSVector dest, WordEntry * destptr, int4 maxpos)
 {
-	uint16	   *clen = (uint16 *) _POSDATAPTR(dest, destptr);
+	uint16	   *clen = &_POSVECPTR(dest, destptr)->npos;
 	int			i;
 	uint16		slen = POSDATALEN(src, srcptr),
 				startlen;
@@ -354,7 +354,7 @@ tsvector_concat(PG_FUNCTION_ARGS)
 			if (ptr->haspos)
 			{
 				cur += SHORTALIGN(ptr1->len);
-				memcpy(cur, _POSDATAPTR(in1, ptr1), POSDATALEN(in1, ptr1) * sizeof(WordEntryPos) + sizeof(uint16));
+				memcpy(cur, _POSVECPTR(in1, ptr1), POSDATALEN(in1, ptr1) * sizeof(WordEntryPos) + sizeof(uint16));
 				cur += POSDATALEN(in1, ptr1) * sizeof(WordEntryPos) + sizeof(uint16);
 			}
 			else
@@ -399,7 +399,7 @@ tsvector_concat(PG_FUNCTION_ARGS)
 				cur += SHORTALIGN(ptr1->len);
 				if (ptr1->haspos)
 				{
-					memcpy(cur, _POSDATAPTR(in1, ptr1), POSDATALEN(in1, ptr1) * sizeof(WordEntryPos) + sizeof(uint16));
+					memcpy(cur, _POSVECPTR(in1, ptr1), POSDATALEN(in1, ptr1) * sizeof(WordEntryPos) + sizeof(uint16));
 					cur += POSDATALEN(in1, ptr1) * sizeof(WordEntryPos) + sizeof(uint16);
 					if (ptr2->haspos)
 						cur += add_pos(in2, ptr2, out, ptr, maxpos) * sizeof(WordEntryPos);
@@ -434,7 +434,7 @@ tsvector_concat(PG_FUNCTION_ARGS)
 		if (ptr->haspos)
 		{
 			cur += SHORTALIGN(ptr1->len);
-			memcpy(cur, _POSDATAPTR(in1, ptr1), POSDATALEN(in1, ptr1) * sizeof(WordEntryPos) + sizeof(uint16));
+			memcpy(cur, _POSVECPTR(in1, ptr1), POSDATALEN(in1, ptr1) * sizeof(WordEntryPos) + sizeof(uint16));
 			cur += POSDATALEN(in1, ptr1) * sizeof(WordEntryPos) + sizeof(uint16);
 		}
 		else
@@ -499,10 +499,17 @@ ValCompare(CHKVAL * chkval, WordEntry * ptr, QueryOperand * item)
  * check weight info
  */
 static bool
-checkclass_str(CHKVAL * chkval, WordEntry * val, QueryOperand * item)
+checkclass_str(CHKVAL *chkval, WordEntry *val, QueryOperand *item)
 {
-	WordEntryPos *ptr = (WordEntryPos *) (chkval->values + SHORTALIGN(val->pos + val->len) + sizeof(uint16));
-	uint16		len = *((uint16 *) (chkval->values + SHORTALIGN(val->pos + val->len)));
+	WordEntryPosVector *posvec;
+	WordEntryPos *ptr;
+	uint16		len;
+
+	posvec = (WordEntryPosVector *) 
+		(chkval->values + SHORTALIGN(val->pos + val->len));
+
+	len = posvec->npos;
+	ptr = posvec->pos;
 
 	while (len--)
 	{
@@ -674,7 +681,13 @@ ts_match_tq(PG_FUNCTION_ARGS)
 }
 
 /*
- * Statistics of tsvector
+ * ts_stat statistic function support
+ */
+
+
+/*
+ * Returns the number of positions in value 'wptr' within tsvector 'txt',
+ * that have a weight equal to one of the weights in 'weight' bitmask.
  */
 static int
 check_weight(TSVector txt, WordEntry * wptr, int8 weight)
@@ -824,6 +837,18 @@ formstat(tsstat * stat, TSVector txt, WordEntry ** entry, uint32 len)
 	return newstat;
 }
 
+/*
+ * This is written like a custom aggregate function, because the
+ * original plan was to do just that. Unfortunately, an aggregate function
+ * can't return a set, so that plan was abandoned. If that limitation is
+ * lifted in the future, ts_stat could be a real aggregate function so that 
+ * you could use it like this:
+ *
+ *   SELECT ts_stat(vector_column) FROM vector_table;
+ *
+ *  where vector_column is a tsvector-type column in vector_table.
+ */
+
 static tsstat *
 ts_accum(tsstat * stat, Datum data)
 {

src/include/tsearch/ts_type.h

@@ -5,7 +5,7 @@
  *
  * Copyright (c) 1998-2007, PostgreSQL Global Development Group
  *
- * $PostgreSQL: pgsql/src/include/tsearch/ts_type.h,v 1.4 2007/09/07 16:03:40 teodor Exp $
+ * $PostgreSQL: pgsql/src/include/tsearch/ts_type.h,v 1.5 2007/09/11 08:46:29 teodor Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -43,6 +43,13 @@ typedef struct
 
 typedef uint16 WordEntryPos;
 
+typedef struct
+{
+	uint16 npos;
+	WordEntryPos pos[1]; /* var length */
+} WordEntryPosVector;
+
+
 #define WEP_GETWEIGHT(x)	( (x) >> 14 )
 #define WEP_GETPOS(x)		( (x) & 0x3fff )
 
@@ -88,9 +95,9 @@ typedef TSVectorData *TSVector;
 /* returns a pointer to the beginning of lexemes */
 #define STRPTR(x)	( (char *) &(x)->entries[x->size] )
 
-#define _POSDATAPTR(x,e)	(STRPTR(x) + SHORTALIGN((e)->pos + (e)->len))
-#define POSDATALEN(x,e) ( ( ((WordEntry*)(e))->haspos ) ? (*(uint16*)_POSDATAPTR(x,e)) : 0 )
-#define POSDATAPTR(x,e) ( (WordEntryPos*)( _POSDATAPTR(x,e)+sizeof(uint16) ) )
+#define _POSVECPTR(x, e) 	((WordEntryPosVector *)(STRPTR(x) + SHORTALIGN((e)->pos + (e)->len)))
+#define POSDATALEN(x,e) ( ( (e)->haspos ) ? (_POSVECPTR(x,e)->npos) : 0 )
+#define POSDATAPTR(x,e) (_POSVECPTR(x,e)->pos)
 
 /*
  * fmgr interface macros