database/postgres
1
0
Fork 0
mirror of https://github.com/postgres/postgres.git synced 2025-07-28 23:42:10 +03:00
Code Activity

Phrase full text search.

Browse Source 
Patch introduces new text search operator (<-> or <DISTANCE>) into tsquery.
On-disk and binary in/out format of tsquery are backward compatible.
It has two side effect:
- change order for tsquery, so, users, who has a btree index over tsquery,
  should reindex it
- less number of parenthesis in tsquery output, and tsquery becomes more
  readable

Authors: Teodor Sigaev, Oleg Bartunov, Dmitry Ivanov
Reviewers: Alexander Korotkov, Artur Zakirov
This commit is contained in:
Teodor Sigaev
2016-04-07 18:44:18 +03:00
parent 015e88942a
commit bb140506df
30 changed files with 2542 additions and 450 deletions
Download Patch File Download Diff File Expand all files Collapse all files

57
src/backend/utils/adt/tsginidx.c
View File

@ -179,14 +179,16 @@ typedef struct
} GinChkVal;
static GinTernaryValue
checkcondition_gin(void *checkval, QueryOperand *val)
checkcondition_gin_internal(GinChkVal *gcv, QueryOperand *val, ExecPhraseData *data)
{
GinChkVal *gcv = (GinChkVal *) checkval;
int j;
/* if any val requiring a weight is used, set recheck flag */
if (val->weight != 0)
*(gcv->need_recheck) = true;
/*
* if any val requiring a weight is used or caller
* needs position information then set recheck flag
*/
if (val->weight != 0 || data != NULL)
*gcv->need_recheck = true;
/* convert item's number to corresponding entry's (operand's) number */
j = gcv->map_item_operand[((QueryItem *) val) - gcv->first_item];
@ -195,16 +197,22 @@ checkcondition_gin(void *checkval, QueryOperand *val)
return gcv->check[j];
}
/*
* Wrapper of check condition function for TS_execute.
*/
static bool
checkcondition_gin(void *checkval, QueryOperand *val, ExecPhraseData *data)
{
return checkcondition_gin_internal((GinChkVal *) checkval,
val,
data) != GIN_FALSE;
}
/*
* Evaluate tsquery boolean expression using ternary logic.
*
* chkcond is a callback function used to evaluate each VAL node in the query.
* checkval can be used to pass information to the callback. TS_execute doesn't
* do anything with it.
*/
static GinTernaryValue
TS_execute_ternary(QueryItem *curitem, void *checkval,
GinTernaryValue (*chkcond) (void *checkval, QueryOperand *val))
TS_execute_ternary(GinChkVal *gcv, QueryItem *curitem)
{
GinTernaryValue val1,
val2,
@ -214,22 +222,30 @@ TS_execute_ternary(QueryItem *curitem, void *checkval,
check_stack_depth();
if (curitem->type == QI_VAL)
return chkcond(checkval, (QueryOperand *) curitem);
return checkcondition_gin_internal(gcv,
(QueryOperand *) curitem,
NULL /* don't have any position info */);
switch (curitem->qoperator.oper)
{
case OP_NOT:
result = TS_execute_ternary(curitem + 1, checkval, chkcond);
result = TS_execute_ternary(gcv, curitem + 1);
if (result == GIN_MAYBE)
return result;
return !result;
case OP_PHRASE:
/*
* GIN doesn't contain any information about positions,
* treat OP_PHRASE as OP_AND with recheck requirement
*/
*gcv->need_recheck = true;
case OP_AND:
val1 = TS_execute_ternary(curitem + curitem->qoperator.left,
checkval, chkcond);
val1 = TS_execute_ternary(gcv, curitem + curitem->qoperator.left);
if (val1 == GIN_FALSE)
return GIN_FALSE;
val2 = TS_execute_ternary(curitem + 1, checkval, chkcond);
val2 = TS_execute_ternary(gcv, curitem + 1);
if (val2 == GIN_FALSE)
return GIN_FALSE;
if (val1 == GIN_TRUE && val2 == GIN_TRUE)
@ -238,11 +254,10 @@ TS_execute_ternary(QueryItem *curitem, void *checkval,
return GIN_MAYBE;
case OP_OR:
val1 = TS_execute_ternary(curitem + curitem->qoperator.left,
checkval, chkcond);
val1 = TS_execute_ternary(gcv, curitem + curitem->qoperator.left);
if (val1 == GIN_TRUE)
return GIN_TRUE;
val2 = TS_execute_ternary(curitem + 1, checkval, chkcond);
val2 = TS_execute_ternary(gcv, curitem + 1);
if (val2 == GIN_TRUE)
return GIN_TRUE;
if (val1 == GIN_FALSE && val2 == GIN_FALSE)
@ -327,9 +342,7 @@ gin_tsquery_triconsistent(PG_FUNCTION_ARGS)
gcv.map_item_operand = (int *) (extra_data[0]);
gcv.need_recheck = &recheck;
res = TS_execute_ternary(GETQUERY(query),
&gcv,
checkcondition_gin);
res = TS_execute_ternary(&gcv, GETQUERY(query));
if (res == GIN_TRUE && recheck)
res = GIN_MAYBE;

4
src/backend/utils/adt/tsgistidx.c
View File

@ -298,7 +298,7 @@ typedef struct
* is there value 'val' in array or not ?
*/
static bool
checkcondition_arr(void *checkval, QueryOperand *val)
checkcondition_arr(void *checkval, QueryOperand *val, ExecPhraseData *data)
{
int32 *StopLow = ((CHKVAL *) checkval)->arrb;
int32 *StopHigh = ((CHKVAL *) checkval)->arre;
@ -327,7 +327,7 @@ checkcondition_arr(void *checkval, QueryOperand *val)
}
static bool
checkcondition_bit(void *checkval, QueryOperand *val)
checkcondition_bit(void *checkval, QueryOperand *val, ExecPhraseData *data)
{
/*
* we are not able to find a prefix in signature tree

311
src/backend/utils/adt/tsquery.c
View File

@ -56,7 +56,7 @@ struct TSQueryParserStateData
/*
* subroutine to parse the modifiers (weight and prefix flag currently)
* part, like ':1AB' of a query.
* part, like ':AB*' of a query.
*/
static char *
get_modifiers(char *buf, int16 *weight, bool *prefix)
@ -100,6 +100,94 @@ get_modifiers(char *buf, int16 *weight, bool *prefix)
return buf;
}
/*
* Parse phrase operator. The operator
* may take the following forms:
*
* a <X> b (distance is no greater than X)
* a <-> b (default distance = 1)
*
* The buffer should begin with '<' char
*/
static char *
parse_phrase_operator(char *buf, int16 *distance)
{
enum
{
PHRASE_OPEN = 0,
PHRASE_DIST,
PHRASE_CLOSE,
PHRASE_ERR,
PHRASE_FINISH
} state = PHRASE_OPEN;
char *ptr = buf;
char *endptr;
long l = 1;
while (*ptr)
{
switch(state)
{
case PHRASE_OPEN:
Assert(t_iseq(ptr, '<'));
state = PHRASE_DIST;
ptr++;
break;
case PHRASE_DIST:
if (t_iseq(ptr, '-'))
{
state = PHRASE_CLOSE;
ptr++;
break;
}
else if (!t_isdigit(ptr))
{
state = PHRASE_ERR;
break;
}
l = strtol(ptr, &endptr, 10);
if (ptr == endptr)
state = PHRASE_ERR;
else if (errno == ERANGE || l > MAXENTRYPOS)
ereport(ERROR,
(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
errmsg("distance in phrase operator should not be greater than %d",
MAXENTRYPOS)));
else
{
state = PHRASE_CLOSE;
ptr = endptr;
}
break;
case PHRASE_CLOSE:
if (t_iseq(ptr, '>'))
{
state = PHRASE_FINISH;
ptr++;
}
else
state = PHRASE_ERR;
break;
case PHRASE_FINISH:
*distance = (int16) l;
return ptr;
case PHRASE_ERR:
default:
goto err;
}
}
err:
*distance = -1;
return buf;
}
/*
* token types for parsing
*/
@ -116,8 +204,10 @@ typedef enum
/*
* get token from query string
*
* *operator is filled in with OP_* when return values is PT_OPR
* *operator is filled in with OP_* when return values is PT_OPR,
* but *weight could contain a distance value in case of phrase operator.
* *strval, *lenval and *weight are filled in when return value is PT_VAL
*
*/
static ts_tokentype
gettoken_query(TSQueryParserState state,
@ -185,13 +275,23 @@ gettoken_query(TSQueryParserState state,
(state->buf)++;
return PT_OPR;
}
if (t_iseq(state->buf, '|'))
else if (t_iseq(state->buf, '|'))
{
state->state = WAITOPERAND;
*operator = OP_OR;
(state->buf)++;
return PT_OPR;
}
else if (t_iseq(state->buf, '<'))
{
state->state = WAITOPERAND;
*operator = OP_PHRASE;
/* weight var is used as storage for distance */
state->buf = parse_phrase_operator(state->buf, weight);
if (*weight < 0)
return PT_ERR;
return PT_OPR;
}
else if (t_iseq(state->buf, ')'))
{
(state->buf)++;
@ -223,15 +323,16 @@ gettoken_query(TSQueryParserState state,
* Push an operator to state->polstr
*/
void
pushOperator(TSQueryParserState state, int8 oper)
pushOperator(TSQueryParserState state, int8 oper, int16 distance)
{
QueryOperator *tmp;
Assert(oper == OP_NOT || oper == OP_AND || oper == OP_OR);
Assert(oper == OP_NOT || oper == OP_AND || oper == OP_OR || oper == OP_PHRASE);
tmp = (QueryOperator *) palloc0(sizeof(QueryOperator));
tmp->type = QI_OPR;
tmp->oper = oper;
tmp->distance = (oper == OP_PHRASE) ? distance : 0;
/* left is filled in later with findoprnd */
state->polstr = lcons(tmp, state->polstr);
@ -330,14 +431,18 @@ makepol(TSQueryParserState state,
PushFunction pushval,
Datum opaque)
{
int8 operator = 0;
ts_tokentype type;
int lenval = 0;
char *strval = NULL;
int8 opstack[STACKDEPTH];
int lenstack = 0;
int16 weight = 0;
bool prefix;
int8 operator = 0;
ts_tokentype type;
int lenval = 0;
char *strval = NULL;
struct
{
int8 op;
int16 distance;
} opstack[STACKDEPTH];
int lenstack = 0;
int16 weight = 0;
bool prefix;
/* since this function recurses, it could be driven to stack overflow */
check_stack_depth();
@ -348,39 +453,48 @@ makepol(TSQueryParserState state,
{
case PT_VAL:
pushval(opaque, state, strval, lenval, weight, prefix);
while (lenstack && (opstack[lenstack - 1] == OP_AND ||
opstack[lenstack - 1] == OP_NOT))
while (lenstack && (opstack[lenstack - 1].op == OP_AND ||
opstack[lenstack - 1].op == OP_PHRASE ||
opstack[lenstack - 1].op == OP_NOT))
{
lenstack--;
pushOperator(state, opstack[lenstack]);
pushOperator(state,
opstack[lenstack].op,
opstack[lenstack].distance);
}
break;
case PT_OPR:
if (lenstack && operator == OP_OR)
pushOperator(state, OP_OR);
pushOperator(state, OP_OR, 0);
else
{
if (lenstack == STACKDEPTH) /* internal error */
elog(ERROR, "tsquery stack too small");
opstack[lenstack] = operator;
opstack[lenstack].op = operator;
opstack[lenstack].distance = weight;
lenstack++;
}
break;
case PT_OPEN:
makepol(state, pushval, opaque);
while (lenstack && (opstack[lenstack - 1] == OP_AND ||
opstack[lenstack - 1] == OP_NOT))
while (lenstack && (opstack[lenstack - 1].op == OP_AND ||
opstack[lenstack - 1].op == OP_PHRASE ||
opstack[lenstack - 1].op == OP_NOT))
{
lenstack--;
pushOperator(state, opstack[lenstack]);
pushOperator(state,
opstack[lenstack].op,
opstack[lenstack].distance);
}
break;
case PT_CLOSE:
while (lenstack)
{
lenstack--;
pushOperator(state, opstack[lenstack]);
pushOperator(state,
opstack[lenstack].op,
opstack[lenstack].distance);
};
return;
case PT_ERR:
@ -394,12 +508,14 @@ makepol(TSQueryParserState state,
while (lenstack)
{
lenstack--;
pushOperator(state, opstack[lenstack]);
pushOperator(state,
opstack[lenstack].op,
opstack[lenstack].distance);
}
}
static void
findoprnd_recurse(QueryItem *ptr, uint32 *pos, int nnodes)
findoprnd_recurse(QueryItem *ptr, uint32 *pos, int nnodes, bool *needcleanup)
{
/* since this function recurses, it could be driven to stack overflow. */
check_stack_depth();
@ -407,33 +523,47 @@ findoprnd_recurse(QueryItem *ptr, uint32 *pos, int nnodes)
if (*pos >= nnodes)
elog(ERROR, "malformed tsquery: operand not found");
if (ptr[*pos].type == QI_VAL ||
ptr[*pos].type == QI_VALSTOP) /* need to handle VALSTOP here, they
* haven't been cleaned away yet. */
if (ptr[*pos].type == QI_VAL)
{
(*pos)++;
}
else if (ptr[*pos].type == QI_VALSTOP)
{
*needcleanup = true; /* we'll have to remove stop words */
(*pos)++;
}
else
{
Assert(ptr[*pos].type == QI_OPR);
if (ptr[*pos].qoperator.oper == OP_NOT)
{
ptr[*pos].qoperator.left = 1;
ptr[*pos].qoperator.left = 1; /* fixed offset */
(*pos)++;
findoprnd_recurse(ptr, pos, nnodes);
/* process the only argument */
findoprnd_recurse(ptr, pos, nnodes, needcleanup);
}
else
{
QueryOperator *curitem = &ptr[*pos].qoperator;
int tmp = *pos;
QueryOperator *curitem = &ptr[*pos].qoperator;
int tmp = *pos; /* save current position */
Assert(curitem->oper == OP_AND || curitem->oper == OP_OR);
Assert(curitem->oper == OP_AND ||
curitem->oper == OP_OR ||
curitem->oper == OP_PHRASE);
if (curitem->oper == OP_PHRASE)
*needcleanup = true; /* push OP_PHRASE down later */
(*pos)++;
findoprnd_recurse(ptr, pos, nnodes);
curitem->left = *pos - tmp;
findoprnd_recurse(ptr, pos, nnodes);
/* process RIGHT argument */
findoprnd_recurse(ptr, pos, nnodes, needcleanup);
curitem->left = *pos - tmp; /* set LEFT arg's offset */
/* process LEFT argument */
findoprnd_recurse(ptr, pos, nnodes, needcleanup);
}
}
}
@ -444,12 +574,13 @@ findoprnd_recurse(QueryItem *ptr, uint32 *pos, int nnodes)
* QueryItems must be in polish (prefix) notation.
*/
static void
findoprnd(QueryItem *ptr, int size)
findoprnd(QueryItem *ptr, int size, bool *needcleanup)
{
uint32 pos;
*needcleanup = false;
pos = 0;
findoprnd_recurse(ptr, &pos, size);
findoprnd_recurse(ptr, &pos, size, needcleanup);
if (pos != size)
elog(ERROR, "malformed tsquery: extra nodes");
@ -466,9 +597,6 @@ findoprnd(QueryItem *ptr, int size)
*
* opaque is passed on to pushval as is, pushval can use it to store its
* private state.
*
* The returned query might contain QI_STOPVAL nodes. The caller is responsible
* for cleaning them up (with clean_fakeval)
*/
TSQuery
parse_tsquery(char *buf,
@ -482,6 +610,7 @@ parse_tsquery(char *buf,
int commonlen;
QueryItem *ptr;
ListCell *cell;
bool needcleanup;
/* init state */
state.buffer = buf;
@ -531,7 +660,7 @@ parse_tsquery(char *buf,
i = 0;
foreach(cell, state.polstr)
{
QueryItem *item = (QueryItem *) lfirst(cell);
QueryItem *item = (QueryItem *) lfirst(cell);
switch (item->type)
{
@ -555,7 +684,14 @@ parse_tsquery(char *buf,
pfree(state.op);
/* Set left operand pointers for every operator. */
findoprnd(ptr, query->size);
findoprnd(ptr, query->size, &needcleanup);
/*
* QI_VALSTOP nodes should be cleaned and
* and OP_PHRASE should be pushed down
*/
if (needcleanup)
return cleanup_fakeval_and_phrase(query);
return query;
}
@ -600,12 +736,15 @@ while( ( (inf)->cur - (inf)->buf ) + (addsize) + 1 >= (inf)->buflen ) \
(inf)->cur = (inf)->buf + len; \
}
#define PRINT_PRIORITY(x) \
( (QO_PRIORITY(x) == OP_NOT) ? OP_NOT_PHRASE : QO_PRIORITY(x) )
/*
* recursive walk on tree and print it in
* infix (human-readable) view
* recursively traverse the tree and
* print it in infix (human-readable) form
*/
static void
infix(INFIX *in, bool first)
infix(INFIX *in, int parentPriority)
{
/* since this function recurses, it could be driven to stack overflow. */
check_stack_depth();
@ -674,24 +813,22 @@ infix(INFIX *in, bool first)
}
else if (in->curpol->qoperator.oper == OP_NOT)
{
bool isopr = false;
int priority = PRINT_PRIORITY(in->curpol);
if (priority < parentPriority)
{
RESIZEBUF(in, 2);
sprintf(in->cur, "( ");
in->cur = strchr(in->cur, '\0');
}
RESIZEBUF(in, 1);
*(in->cur) = '!';
in->cur++;
*(in->cur) = '\0';
in->curpol++;
if (in->curpol->type == QI_OPR)
{
isopr = true;
RESIZEBUF(in, 2);
sprintf(in->cur, "( ");
in->cur = strchr(in->cur, '\0');
}
infix(in, isopr);
if (isopr)
infix(in, priority);
if (priority < parentPriority)
{
RESIZEBUF(in, 2);
sprintf(in->cur, " )");
@ -701,11 +838,18 @@ infix(INFIX *in, bool first)
else
{
int8 op = in->curpol->qoperator.oper;
int priority = PRINT_PRIORITY(in->curpol);
int16 distance = in->curpol->qoperator.distance;
INFIX nrm;
bool needParenthesis = false;
in->curpol++;
if (op == OP_OR && !first)
if (priority < parentPriority ||
(op == OP_PHRASE &&
(priority == parentPriority || /* phrases are not commutative! */
parentPriority == OP_PRIORITY(OP_AND))))
{
needParenthesis = true;
RESIZEBUF(in, 2);
sprintf(in->cur, "( ");
in->cur = strchr(in->cur, '\0');
@ -717,14 +861,14 @@ infix(INFIX *in, bool first)
nrm.cur = nrm.buf = (char *) palloc(sizeof(char) * nrm.buflen);
/* get right operand */
infix(&nrm, false);
infix(&nrm, priority);
/* get & print left operand */
in->curpol = nrm.curpol;
infix(in, false);
infix(in, priority);
/* print operator & right operand */
RESIZEBUF(in, 3 + (nrm.cur - nrm.buf));
RESIZEBUF(in, 3 + (2 + 10 /* distance */) + (nrm.cur - nrm.buf));
switch (op)
{
case OP_OR:
@ -733,6 +877,12 @@ infix(INFIX *in, bool first)
case OP_AND:
sprintf(in->cur, " & %s", nrm.buf);
break;
case OP_PHRASE:
if (distance != 1)
sprintf(in->cur, " <%d> %s", distance, nrm.buf);
else
sprintf(in->cur, " <-> %s", nrm.buf);
break;
default:
/* OP_NOT is handled in above if-branch */
elog(ERROR, "unrecognized operator type: %d", op);
@ -740,7 +890,7 @@ infix(INFIX *in, bool first)
in->cur = strchr(in->cur, '\0');
pfree(nrm.buf);
if (op == OP_OR && !first)
if (needParenthesis)
{
RESIZEBUF(in, 2);
sprintf(in->cur, " )");
@ -749,7 +899,6 @@ infix(INFIX *in, bool first)
}
}
Datum
tsqueryout(PG_FUNCTION_ARGS)
{
@ -768,7 +917,7 @@ tsqueryout(PG_FUNCTION_ARGS)
nrm.cur = nrm.buf = (char *) palloc(sizeof(char) * nrm.buflen);
*(nrm.cur) = '\0';
nrm.op = GETOPERAND(query);
infix(&nrm, true);
infix(&nrm, -1 /* lowest priority */);
PG_FREE_IF_COPY(query, 0);
PG_RETURN_CSTRING(nrm.buf);
@ -789,7 +938,8 @@ tsqueryout(PG_FUNCTION_ARGS)
*
* For each operator:
* uint8 type, QI_OPR
* uint8 operator, one of OP_AND, OP_OR, OP_NOT.
* uint8 operator, one of OP_AND, OP_PHRASE OP_OR, OP_NOT.
* uint16 distance (only for OP_PHRASE)
*/
Datum
tsquerysend(PG_FUNCTION_ARGS)
@ -815,6 +965,9 @@ tsquerysend(PG_FUNCTION_ARGS)
break;
case QI_OPR:
pq_sendint(&buf, item->qoperator.oper, sizeof(item->qoperator.oper));
if (item->qoperator.oper == OP_PHRASE)
pq_sendint(&buf, item->qoperator.distance,
sizeof(item->qoperator.distance));
break;
default:
elog(ERROR, "unrecognized tsquery node type: %d", item->type);
@ -830,15 +983,16 @@ tsquerysend(PG_FUNCTION_ARGS)
Datum
tsqueryrecv(PG_FUNCTION_ARGS)
{
StringInfo buf = (StringInfo) PG_GETARG_POINTER(0);
TSQuery query;
int i,
len;
QueryItem *item;
int datalen;
char *ptr;
uint32 size;
const char **operands;
StringInfo buf = (StringInfo) PG_GETARG_POINTER(0);
TSQuery query;
int i,
len;
QueryItem *item;
int datalen;
char *ptr;
uint32 size;
const char **operands;
bool needcleanup;
size = pq_getmsgint(buf, sizeof(uint32));
if (size > (MaxAllocSize / sizeof(QueryItem)))
@ -907,13 +1061,15 @@ tsqueryrecv(PG_FUNCTION_ARGS)
int8 oper;
oper = (int8) pq_getmsgint(buf, sizeof(int8));
if (oper != OP_NOT && oper != OP_OR && oper != OP_AND)
if (oper != OP_NOT && oper != OP_OR && oper != OP_AND && oper != OP_PHRASE)
elog(ERROR, "invalid tsquery: unrecognized operator type %d",
(int) oper);
if (i == size - 1)
elog(ERROR, "invalid pointer to right operand");
item->qoperator.oper = oper;
if (oper == OP_PHRASE)
item->qoperator.distance = (int16) pq_getmsgint(buf, sizeof(int16));
}
else
elog(ERROR, "unrecognized tsquery node type: %d", item->type);
@ -930,7 +1086,7 @@ tsqueryrecv(PG_FUNCTION_ARGS)
* Fill in the left-pointers. Checks that the tree is well-formed as a
* side-effect.
*/
findoprnd(item, size);
findoprnd(item, size, &needcleanup);
/* Copy operands to output struct */
for (i = 0; i < size; i++)
@ -949,7 +1105,10 @@ tsqueryrecv(PG_FUNCTION_ARGS)
SET_VARSIZE(query, len + datalen);
PG_RETURN_TSVECTOR(query);
if (needcleanup)
PG_RETURN_TSQUERY(cleanup_fakeval_and_phrase(query));
PG_RETURN_TSQUERY(query);
}
/*

364
src/backend/utils/adt/tsquery_cleanup.c
View File

@ -25,6 +25,12 @@ typedef struct NODE
QueryItem *valnode;
} NODE;
/* Non-operator nodes have fake (but highest) priority */
#define NODE_PRIORITY(x) \
( ((x)->valnode->qoperator.type == QI_OPR) ? \
QO_PRIORITY((x)->valnode) : \
TOP_PRIORITY )
/*
* make query tree from plain view of query
*/
@ -160,7 +166,8 @@ clean_NOT_intree(NODE *node)
{
NODE *res = node;
Assert(node->valnode->qoperator.oper == OP_AND);
Assert(node->valnode->qoperator.oper == OP_AND ||
node->valnode->qoperator.oper == OP_PHRASE);
node->left = clean_NOT_intree(node->left);
node->right = clean_NOT_intree(node->right);
@ -212,18 +219,20 @@ clean_NOT(QueryItem *ptr, int *len)
#define V_STOP 3 /* the expression is a stop word */
/*
* Clean query tree from values which is always in
* text (stopword)
* Remove QI_VALSTOP (stopword nodes) from query tree.
*/
static NODE *
clean_fakeval_intree(NODE *node, char *result)
clean_fakeval_intree(NODE *node, char *result, int *adddistance)
{
char lresult = V_UNKNOWN,
rresult = V_UNKNOWN;
char lresult = V_UNKNOWN,
rresult = V_UNKNOWN;
/* since this function recurses, it could be driven to stack overflow. */
check_stack_depth();
if (adddistance)
*adddistance = 0;
if (node->valnode->type == QI_VAL)
return node;
else if (node->valnode->type == QI_VALSTOP)
@ -237,7 +246,7 @@ clean_fakeval_intree(NODE *node, char *result)
if (node->valnode->qoperator.oper == OP_NOT)
{
node->right = clean_fakeval_intree(node->right, &rresult);
node->right = clean_fakeval_intree(node->right, &rresult, NULL);
if (!node->right)
{
*result = V_STOP;
@ -247,13 +256,30 @@ clean_fakeval_intree(NODE *node, char *result)
}
else
{
NODE *res = node;
NODE *res = node;
int ndistance, ldistance = 0, rdistance = 0;
node->left = clean_fakeval_intree(node->left, &lresult);
node->right = clean_fakeval_intree(node->right, &rresult);
ndistance = (node->valnode->qoperator.oper == OP_PHRASE) ?
node->valnode->qoperator.distance :
0;
node->left = clean_fakeval_intree(node->left,
&lresult,
ndistance ? &ldistance : NULL);
node->right = clean_fakeval_intree(node->right,
&rresult,
ndistance ? &rdistance : NULL);
/*
* ndistance, ldistance and rdistance are greater than zero
* if their corresponding nodes are OP_PHRASE
*/
if (lresult == V_STOP && rresult == V_STOP)
{
if (adddistance && ndistance)
*adddistance = ldistance + ndistance + rdistance;
freetree(node);
*result = V_STOP;
return NULL;
@ -261,33 +287,333 @@ clean_fakeval_intree(NODE *node, char *result)
else if (lresult == V_STOP)
{
res = node->right;
/*
* propagate distance from current node to the
* right upper subtree.
*/
if (adddistance && ndistance)
*adddistance = rdistance;
pfree(node);
}
else if (rresult == V_STOP)
{
res = node->left;
/*
* propagate distance from current node to the upper tree.
*/
if (adddistance && ndistance)
*adddistance = ndistance + ldistance;
pfree(node);
}
else if (ndistance)
{
node->valnode->qoperator.distance += ldistance;
if (adddistance)
*adddistance = 0;
}
else if (adddistance)
{
*adddistance = 0;
}
return res;
}
return node;
}
QueryItem *
clean_fakeval(QueryItem *ptr, int *len)
static NODE *
copyNODE(NODE *node)
{
NODE *root = maketree(ptr);
char result = V_UNKNOWN;
NODE *resroot;
NODE *cnode = palloc(sizeof(NODE));
resroot = clean_fakeval_intree(root, &result);
/* since this function recurses, it could be driven to stack overflow. */
check_stack_depth();
cnode->valnode = palloc(sizeof(QueryItem));
*(cnode->valnode) = *(node->valnode);
if (node->valnode->type == QI_OPR)
{
cnode->right = copyNODE(node->right);
if (node->valnode->qoperator.oper != OP_NOT)
cnode->left = copyNODE(node->left);
}
return cnode;
}
static NODE *
makeNODE(int8 op, NODE *left, NODE *right)
{
NODE *node = palloc(sizeof(NODE));
node->valnode = palloc(sizeof(QueryItem));
node->valnode->qoperator.type = QI_OPR;
node->valnode->qoperator.oper = op;
node->left = left;
node->right = right;
return node;
}
/*
* Move operation with high priority to the leaves. This guarantees
* that the phrase operator will be near the bottom of the tree.
* An idea behind is do not store position of lexemes during execution
* of ordinary operations (AND, OR, NOT) because it could be expensive.
* Actual transformation will be performed only on subtrees under the
* <-> (<n>) operation since it's needed solely for the phrase operator.
*
* Rules:
* a <-> (b | c) => (a <-> b) | (a <-> c)
* (a | b) <-> c => (a <-> c) | (b <-> c)
* a <-> !b => a & !(a <-> b)
* !a <-> b => b & !(a <-> b)
*
* Warnings for readers:
* a <-> b != b <-> a
*
* a <n> (b <n> c) != (a <n> b) <n> c since the phrase lengths are:
* n 2n-1
*/
static NODE *
normalize_phrase_tree(NODE *node)
{
/* there should be no stop words at this point */
Assert(node->valnode->type != QI_VALSTOP);
if (node->valnode->type == QI_VAL)
return node;
/* since this function recurses, it could be driven to stack overflow. */
check_stack_depth();
Assert(node->valnode->type == QI_OPR);
if (node->valnode->qoperator.oper == OP_NOT)
{
/* eliminate NOT sequence */
while (node->valnode->type == QI_OPR &&
node->valnode->qoperator.oper == node->right->valnode->qoperator.oper)
{
node = node->right->right;
}
node->right = normalize_phrase_tree(node->right);
}
else if (node->valnode->qoperator.oper == OP_PHRASE)
{
int16 distance;
NODE *X;
node->left = normalize_phrase_tree(node->left);
node->right = normalize_phrase_tree(node->right);
if (NODE_PRIORITY(node) <= NODE_PRIORITY(node->right) &&
NODE_PRIORITY(node) <= NODE_PRIORITY(node->left))
return node;
/*
* We can't swap left-right and works only with left child
* because of a <-> b != b <-> a
*/
distance = node->valnode->qoperator.distance;
if (node->right->valnode->type == QI_OPR)
{
switch (node->right->valnode->qoperator.oper)
{
case OP_AND:
/* a <-> (b & c) => (a <-> b) & (a <-> c) */
node = makeNODE(OP_AND,
makeNODE(OP_PHRASE,
node->left,
node->right->left),
makeNODE(OP_PHRASE,
copyNODE(node->left),
node->right->right));
node->left->valnode->qoperator.distance =
node->right->valnode->qoperator.distance = distance;
break;
case OP_OR:
/* a <-> (b | c) => (a <-> b) | (a <-> c) */
node = makeNODE(OP_OR,
makeNODE(OP_PHRASE,
node->left,
node->right->left),
makeNODE(OP_PHRASE,
copyNODE(node->left),
node->right->right));
node->left->valnode->qoperator.distance =
node->right->valnode->qoperator.distance = distance;
break;
case OP_NOT:
/* a <-> !b => a & !(a <-> b) */
X = node->right;
node->right = node->right->right;
X->right = node;
node = makeNODE(OP_AND,
copyNODE(node->left),
X);
break;
case OP_PHRASE:
/* no-op */
break;
default:
elog(ERROR,"Wrong type of tsquery node: %d",
node->right->valnode->qoperator.oper);
}
}
if (node->left->valnode->type == QI_OPR &&
node->valnode->qoperator.oper == OP_PHRASE)
{
/*
* if the node is still OP_PHRASE, check the left subtree,
* otherwise the whole node will be transformed later.
*/
switch(node->left->valnode->qoperator.oper)
{
case OP_AND:
/* (a & b) <-> c => (a <-> c) & (b <-> c) */
node = makeNODE(OP_AND,
makeNODE(OP_PHRASE,
node->left->left,
node->right),
makeNODE(OP_PHRASE,
node->left->right,
copyNODE(node->right)));
node->left->valnode->qoperator.distance =
node->right->valnode->qoperator.distance = distance;
break;
case OP_OR:
/* (a | b) <-> c => (a <-> c) | (b <-> c) */
node = makeNODE(OP_OR,
makeNODE(OP_PHRASE,
node->left->left,
node->right),
makeNODE(OP_PHRASE,
node->left->right,
copyNODE(node->right)));
node->left->valnode->qoperator.distance =
node->right->valnode->qoperator.distance = distance;
break;
case OP_NOT:
/* !a <-> b => b & !(a <-> b) */
X = node->left;
node->left = node->left->right;
X->right = node;
node = makeNODE(OP_AND,
X,
copyNODE(node->right));
break;
case OP_PHRASE:
/* no-op */
break;
default:
elog(ERROR,"Wrong type of tsquery node: %d",
node->left->valnode->qoperator.oper);
}
}
/* continue transformation */
node = normalize_phrase_tree(node);
}
else /* AND or OR */
{
node->left = normalize_phrase_tree(node->left);
node->right = normalize_phrase_tree(node->right);
}
return node;
}
/*
* Number of elements in query tree
*/
static int32
calcstrlen(NODE *node)
{
int32 size = 0;
if (node->valnode->type == QI_VAL)
{
size = node->valnode->qoperand.length + 1;
}
else
{
Assert(node->valnode->type == QI_OPR);
size = calcstrlen(node->right);
if (node->valnode->qoperator.oper != OP_NOT)
size += calcstrlen(node->left);
}
return size;
}
TSQuery
cleanup_fakeval_and_phrase(TSQuery in)
{
int32 len,
lenstr,
commonlen,
i;
NODE *root;
char result = V_UNKNOWN;
TSQuery out;
QueryItem *items;
char *operands;
if (in->size == 0)
return in;
/* eliminate stop words */
root = clean_fakeval_intree(maketree(GETQUERY(in)), &result, NULL);
if (result != V_UNKNOWN)
{
ereport(NOTICE,
(errmsg("text-search query contains only stop words or doesn't contain lexemes, ignored")));
*len = 0;
return NULL;
out = palloc(HDRSIZETQ);
out->size = 0;
SET_VARSIZE(out, HDRSIZETQ);
return out;
}
return plaintree(resroot, len);
/* push OP_PHRASE nodes down */
root = normalize_phrase_tree(root);
/*
* Build TSQuery from plain view
*/
lenstr = calcstrlen(root);
items = plaintree(root, &len);
commonlen = COMPUTESIZE(len, lenstr);
out = palloc(commonlen);
SET_VARSIZE(out, commonlen);
out->size = len;
memcpy(GETQUERY(out), items, len * sizeof(QueryItem));
items = GETQUERY(out);
operands = GETOPERAND(out);
for (i = 0; i < out->size; i++)
{
QueryOperand *op = (QueryOperand *) &items[i];
if (op->type != QI_VAL)
continue;
memcpy(operands, GETOPERAND(in) + op->distance, op->length);
operands[op->length] = '\0';
op->distance = operands - GETOPERAND(out);
operands += op->length + 1;
}
return out;
}

54
src/backend/utils/adt/tsquery_op.c
View File

@ -27,7 +27,7 @@ tsquery_numnode(PG_FUNCTION_ARGS)
}
static QTNode *
join_tsqueries(TSQuery a, TSQuery b, int8 operator)
join_tsqueries(TSQuery a, TSQuery b, int8 operator, uint16 distance)
{
QTNode *res = (QTNode *) palloc0(sizeof(QTNode));
@ -36,6 +36,8 @@ join_tsqueries(TSQuery a, TSQuery b, int8 operator)
res->valnode = (QueryItem *) palloc0(sizeof(QueryItem));
res->valnode->type = QI_OPR;
res->valnode->qoperator.oper = operator;
if (operator == OP_PHRASE)
res->valnode->qoperator.distance = distance;
res->child = (QTNode **) palloc0(sizeof(QTNode *) * 2);
res->child[0] = QT2QTN(GETQUERY(b), GETOPERAND(b));
@ -64,7 +66,7 @@ tsquery_and(PG_FUNCTION_ARGS)
PG_RETURN_POINTER(a);
}
res = join_tsqueries(a, b, OP_AND);
res = join_tsqueries(a, b, OP_AND, 0);
query = QTN2QT(res);
@ -94,7 +96,7 @@ tsquery_or(PG_FUNCTION_ARGS)
PG_RETURN_POINTER(a);
}
res = join_tsqueries(a, b, OP_OR);
res = join_tsqueries(a, b, OP_OR, 0);
query = QTN2QT(res);
@ -105,6 +107,52 @@ tsquery_or(PG_FUNCTION_ARGS)
PG_RETURN_POINTER(query);
}
Datum
tsquery_phrase_distance(PG_FUNCTION_ARGS)
{
TSQuery a = PG_GETARG_TSQUERY_COPY(0);
TSQuery b = PG_GETARG_TSQUERY_COPY(1);
QTNode *res;
TSQuery query;
int32 distance = PG_GETARG_INT32(2);
if (distance < 0 || distance > MAXENTRYPOS)
ereport(ERROR,
(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
errmsg("distance in phrase operator should be non-negative and less than %d",
MAXENTRYPOS)));
if (a->size == 0)
{
PG_FREE_IF_COPY(a, 1);
PG_RETURN_POINTER(b);
}
else if (b->size == 0)
{
PG_FREE_IF_COPY(b, 1);
PG_RETURN_POINTER(a);
}
res = join_tsqueries(a, b, OP_PHRASE, (uint16) distance);
query = QTN2QT(res);
QTNFree(res);
PG_FREE_IF_COPY(a, 0);
PG_FREE_IF_COPY(b, 1);
PG_RETURN_POINTER(cleanup_fakeval_and_phrase(query));
}
Datum
tsquery_phrase(PG_FUNCTION_ARGS)
{
PG_RETURN_POINTER(DirectFunctionCall3(
tsquery_phrase_distance,
PG_GETARG_DATUM(0),
PG_GETARG_DATUM(1),
Int32GetDatum(1)));
}
Datum
tsquery_not(PG_FUNCTION_ARGS)
{

11
src/backend/utils/adt/tsquery_util.c
View File

@ -110,6 +110,10 @@ QTNodeCompare(QTNode *an, QTNode *bn)
if ((res = QTNodeCompare(an->child[i], bn->child[i])) != 0)
return res;
}
if (ao->oper == OP_PHRASE && ao->distance != bo->distance)
return (ao->distance > bo->distance) ? -1 : 1;
return 0;
}
else if (an->valnode->type == QI_VAL)
@ -150,7 +154,7 @@ QTNSort(QTNode *in)
for (i = 0; i < in->nchild; i++)
QTNSort(in->child[i]);
if (in->nchild > 1)
if (in->nchild > 1 && in->valnode->qoperator.oper != OP_PHRASE)
qsort((void *) in->child, in->nchild, sizeof(QTNode *), cmpQTN);
}
@ -190,7 +194,10 @@ QTNTernary(QTNode *in)
{
QTNode *cc = in->child[i];
if (cc->valnode->type == QI_OPR && in->valnode->qoperator.oper == cc->valnode->qoperator.oper)
/* OP_Phrase isn't associative */
if (cc->valnode->type == QI_OPR &&
in->valnode->qoperator.oper == cc->valnode->qoperator.oper &&
in->valnode->qoperator.oper != OP_PHRASE)
{
int oldnchild = in->nchild;

263
src/backend/utils/adt/tsrank.c
View File

@ -364,8 +364,10 @@ calc_rank(const float *w, TSVector t, TSQuery q, int32 method)
return 0.0;
/* XXX: What about NOT? */
res = (item->type == QI_OPR && item->qoperator.oper == OP_AND) ?
calc_rank_and(w, t, q) : calc_rank_or(w, t, q);
res = (item->type == QI_OPR && (item->qoperator.oper == OP_AND ||
item->qoperator.oper == OP_PHRASE)) ?
calc_rank_and(w, t, q) :
calc_rank_or(w, t, q);
if (res < 0)
res = 1e-20f;
@ -496,10 +498,17 @@ ts_rank_tt(PG_FUNCTION_ARGS)
typedef struct
{
QueryItem **item;
int16 nitem;
uint8 wclass;
int32 pos;
union {
struct { /* compiled doc representation */
QueryItem **items;
int16 nitem;
} query;
struct { /* struct is used for preparing doc representation */
QueryItem *item;
WordEntry *entry;
} map;
} data;
WordEntryPos pos;
} DocRepresentation;
static int
@ -508,26 +517,59 @@ compareDocR(const void *va, const void *vb)
const DocRepresentation *a = (const DocRepresentation *) va;
const DocRepresentation *b = (const DocRepresentation *) vb;
if (a->pos == b->pos)
return 0;
return (a->pos > b->pos) ? 1 : -1;
if (WEP_GETPOS(a->pos) == WEP_GETPOS(b->pos))
{
if (WEP_GETWEIGHT(a->pos) == WEP_GETWEIGHT(b->pos))
{
if (a->data.map.entry == b->data.map.entry)
return 0;
return (a->data.map.entry > b->data.map.entry) ? 1 : -1;
}
return (WEP_GETWEIGHT(a->pos) > WEP_GETWEIGHT(b->pos)) ? 1 : -1;
}
return (WEP_GETPOS(a->pos) > WEP_GETPOS(b->pos)) ? 1 : -1;
}
#define MAXQROPOS MAXENTRYPOS
typedef struct
{
bool operandexists;
bool reverseinsert; /* indicates insert order,
true means descending order */
uint32 npos;
WordEntryPos pos[MAXQROPOS];
} QueryRepresentationOperand;
typedef struct
{
TSQuery query;
bool *operandexist;
TSQuery query;
QueryRepresentationOperand *operandData;
} QueryRepresentation;
#define QR_GET_OPERAND_EXISTS(q, v) ( (q)->operandexist[ ((QueryItem*)(v)) - GETQUERY((q)->query) ] )
#define QR_SET_OPERAND_EXISTS(q, v) QR_GET_OPERAND_EXISTS(q,v) = true
#define QR_GET_OPERAND_DATA(q, v) \
( (q)->operandData + (((QueryItem*)(v)) - GETQUERY((q)->query)) )
static bool
checkcondition_QueryOperand(void *checkval, QueryOperand *val)
checkcondition_QueryOperand(void *checkval, QueryOperand *val, ExecPhraseData *data)
{
QueryRepresentation *qr = (QueryRepresentation *) checkval;
QueryRepresentation *qr = (QueryRepresentation *) checkval;
QueryRepresentationOperand *opData = QR_GET_OPERAND_DATA(qr, val);
return QR_GET_OPERAND_EXISTS(qr, val);
if (!opData->operandexists)
return false;
if (data)
{
data->npos = opData->npos;
data->pos = opData->pos;
if (opData->reverseinsert)
data->pos += MAXQROPOS - opData->npos;
}
return true;
}
typedef struct
@ -539,14 +581,65 @@ typedef struct
DocRepresentation *end;
} CoverExt;
static void
resetQueryRepresentation(QueryRepresentation *qr, bool reverseinsert)
{
int i;
for(i = 0; i < qr->query->size; i++)
{
qr->operandData[i].operandexists = false;
qr->operandData[i].reverseinsert = reverseinsert;
qr->operandData[i].npos = 0;
}
}
static void
fillQueryRepresentationData(QueryRepresentation *qr, DocRepresentation *entry)
{
int i;
int lastPos;
QueryRepresentationOperand *opData;
for (i = 0; i < entry->data.query.nitem; i++)
{
if (entry->data.query.items[i]->type != QI_VAL)
continue;
opData = QR_GET_OPERAND_DATA(qr, entry->data.query.items[i]);
opData->operandexists = true;
if (opData->npos == 0)
{
lastPos = (opData->reverseinsert) ? (MAXQROPOS - 1) : 0;
opData->pos[lastPos] = entry->pos;
opData->npos++;
continue;
}
lastPos = opData->reverseinsert ?
(MAXQROPOS - opData->npos) :
(opData->npos - 1);
if (WEP_GETPOS(opData->pos[lastPos]) != WEP_GETPOS(entry->pos))
{
lastPos = opData->reverseinsert ?
(MAXQROPOS - 1 - opData->npos) :
(opData->npos);
opData->pos[lastPos] = entry->pos;
opData->npos++;
}
}
}
static bool
Cover(DocRepresentation *doc, int len, QueryRepresentation *qr, CoverExt *ext)
{
DocRepresentation *ptr;
int lastpos = ext->pos;
int i;
bool found = false;
DocRepresentation *ptr;
int lastpos = ext->pos;
bool found = false;
/*
* since this function recurses, it could be driven to stack overflow.
@ -554,7 +647,7 @@ Cover(DocRepresentation *doc, int len, QueryRepresentation *qr, CoverExt *ext)
*/
check_stack_depth();
memset(qr->operandexist, 0, sizeof(bool) * qr->query->size);
resetQueryRepresentation(qr, false);
ext->p = INT_MAX;
ext->q = 0;
@ -563,16 +656,13 @@ Cover(DocRepresentation *doc, int len, QueryRepresentation *qr, CoverExt *ext)
/* find upper bound of cover from current position, move up */
while (ptr - doc < len)
{
for (i = 0; i < ptr->nitem; i++)
{
if (ptr->item[i]->type == QI_VAL)
QR_SET_OPERAND_EXISTS(qr, ptr->item[i]);
}
fillQueryRepresentationData(qr, ptr);
if (TS_execute(GETQUERY(qr->query), (void *) qr, false, checkcondition_QueryOperand))
{
if (ptr->pos > ext->q)
if (WEP_GETPOS(ptr->pos) > ext->q)
{
ext->q = ptr->pos;
ext->q = WEP_GETPOS(ptr->pos);
ext->end = ptr;
lastpos = ptr - doc;
found = true;
@ -585,22 +675,24 @@ Cover(DocRepresentation *doc, int len, QueryRepresentation *qr, CoverExt *ext)
if (!found)
return false;
memset(qr->operandexist, 0, sizeof(bool) * qr->query->size);
resetQueryRepresentation(qr, true);
ptr = doc + lastpos;
/* 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)
QR_SET_OPERAND_EXISTS(qr, ptr->item[i]);
/*
* we scan doc from right to left, so pos info in reverse order!
*/
fillQueryRepresentationData(qr, ptr);
if (TS_execute(GETQUERY(qr->query), (void *) qr, true, checkcondition_QueryOperand))
{
if (ptr->pos < ext->p)
if (WEP_GETPOS(ptr->pos) < ext->p)
{
ext->begin = ptr;
ext->p = ptr->pos;
ext->p = WEP_GETPOS(ptr->pos);
}
break;
}
@ -628,18 +720,20 @@ get_docrep(TSVector txt, QueryRepresentation *qr, int *doclen)
WordEntry *entry,
*firstentry;
WordEntryPos *post;
int32 dimt,
int32 dimt, /* number of 'post' items */
j,
i,
nitem;
int len = qr->query->size * 4,
cur = 0;
DocRepresentation *doc;
char *operand;
doc = (DocRepresentation *) palloc(sizeof(DocRepresentation) * len);
operand = GETOPERAND(qr->query);
/*
* Iterate through query to make DocRepresentaion for words and it's entries
* satisfied by query
*/
for (i = 0; i < qr->query->size; i++)
{
QueryOperand *curoperand;
@ -649,13 +743,11 @@ get_docrep(TSVector txt, QueryRepresentation *qr, int *doclen)
curoperand = &item[i].qoperand;
if (QR_GET_OPERAND_EXISTS(qr, &item[i]))
continue;
firstentry = entry = find_wordentry(txt, qr->query, curoperand, &nitem);
if (!entry)
continue;
/* iterations over entries in tsvector */
while (entry - firstentry < nitem)
{
if (entry->haspos)
@ -676,53 +768,67 @@ get_docrep(TSVector txt, QueryRepresentation *qr, int *doclen)
doc = (DocRepresentation *) repalloc(doc, sizeof(DocRepresentation) * len);
}
/* iterations over entry's positions */
for (j = 0; j < dimt; j++)
{
if (j == 0)
if (curoperand->weight == 0 ||
curoperand->weight & (1 << WEP_GETWEIGHT(post[j])))
{
int k;
doc[cur].nitem = 0;
doc[cur].item = (QueryItem **) palloc(sizeof(QueryItem *) * qr->query->size);
for (k = 0; k < qr->query->size; k++)
{
QueryOperand *kptr = &item[k].qoperand;
QueryOperand *iptr = &item[i].qoperand;
if (k == i ||
(item[k].type == QI_VAL &&
compareQueryOperand(&kptr, &iptr, operand) == 0))
{
/*
* if k == i, we've already checked above that
* it's type == Q_VAL
*/
doc[cur].item[doc[cur].nitem] = item + k;
doc[cur].nitem++;
QR_SET_OPERAND_EXISTS(qr, item + k);
}
}
doc[cur].pos = post[j];
doc[cur].data.map.entry = entry;
doc[cur].data.map.item = (QueryItem *) curoperand;
cur++;
}
else
{
doc[cur].nitem = doc[cur - 1].nitem;
doc[cur].item = doc[cur - 1].item;
}
doc[cur].pos = WEP_GETPOS(post[j]);
doc[cur].wclass = WEP_GETWEIGHT(post[j]);
cur++;
}
entry++;
}
}
*doclen = cur;
if (cur > 0)
{
DocRepresentation *rptr = doc + 1,
*wptr = doc,
storage;
/*
* Sort representation in ascending order by pos and entry
*/
qsort((void *) doc, cur, sizeof(DocRepresentation), compareDocR);
/*
* Join QueryItem per WordEntry and it's position
*/
storage.pos = doc->pos;
storage.data.query.items = palloc(sizeof(QueryItem *) * qr->query->size);
storage.data.query.items[0] = doc->data.map.item;
storage.data.query.nitem = 1;
while (rptr - doc < cur)
{
if (rptr->pos == (rptr-1)->pos &&
rptr->data.map.entry == (rptr-1)->data.map.entry)
{
storage.data.query.items[storage.data.query.nitem] = rptr->data.map.item;
storage.data.query.nitem++;
}
else
{
*wptr = storage;
wptr++;
storage.pos = rptr->pos;
storage.data.query.items = palloc(sizeof(QueryItem *) * qr->query->size);
storage.data.query.items[0] = rptr->data.map.item;
storage.data.query.nitem = 1;
}
rptr++;
}
*wptr = storage;
wptr++;
*doclen = wptr - doc;
return doc;
}
@ -758,12 +864,13 @@ calc_rank_cd(const float4 *arrdata, TSVector txt, TSQuery query, int method)
}
qr.query = query;
qr.operandexist = (bool *) palloc0(sizeof(bool) * query->size);
qr.operandData = (QueryRepresentationOperand *)
palloc0(sizeof(QueryRepresentationOperand) * query->size);
doc = get_docrep(txt, &qr, &doclen);
if (!doc)
{
pfree(qr.operandexist);
pfree(qr.operandData);
return 0.0;
}
@ -777,7 +884,7 @@ calc_rank_cd(const float4 *arrdata, TSVector txt, TSQuery query, int method)
while (ptr <= ext.end)
{
InvSum += invws[ptr->wclass];
InvSum += invws[WEP_GETWEIGHT(ptr->pos)];
ptr++;
}
@ -827,7 +934,7 @@ calc_rank_cd(const float4 *arrdata, TSVector txt, TSQuery query, int method)
pfree(doc);
pfree(qr.operandexist);
pfree(qr.operandData);
return (float4) Wdoc;
}

2
src/backend/utils/adt/tsvector.c
View File

@ -28,7 +28,7 @@ typedef struct
/* Compare two WordEntryPos values for qsort */
static int
int
comparePos(const void *a, const void *b)
{
int apos = WEP_GETPOS(*(const WordEntryPos *) a);

330
src/backend/utils/adt/tsvector_op.c
View File

@ -1121,35 +1121,124 @@ tsCompareString(char *a, int lena, char *b, int lenb, bool prefix)
}
/*
* check weight info
* Check weight info or/and fill 'data' with the required positions
*/
static bool
checkclass_str(CHKVAL *chkval, WordEntry *val, QueryOperand *item)
checkclass_str(CHKVAL *chkval, WordEntry *entry, QueryOperand *val,
ExecPhraseData *data)
{
WordEntryPosVector *posvec;
WordEntryPos *ptr;
uint16 len;
bool result = false;
posvec = (WordEntryPosVector *)
(chkval->values + SHORTALIGN(val->pos + val->len));
len = posvec->npos;
ptr = posvec->pos;
while (len--)
if (entry->haspos && (val->weight || data))
{
if (item->weight & (1 << WEP_GETWEIGHT(*ptr)))
return true;
ptr++;
WordEntryPosVector *posvec;
/*
* We can't use the _POSVECPTR macro here because the pointer to the
* tsvector's lexeme storage is already contained in chkval->values.
*/
posvec = (WordEntryPosVector *)
(chkval->values + SHORTALIGN(entry->pos + entry->len));
if (val->weight && data)
{
WordEntryPos *posvec_iter = posvec->pos;
WordEntryPos *dptr;
/*
* Filter position information by weights
*/
dptr = data->pos = palloc(sizeof(WordEntryPos) * posvec->npos);
data->allocated = true;
/* Is there a position with a matching weight? */
while (posvec_iter < posvec->pos + posvec->npos)
{
/* If true, append this position to the data->pos */
if (val->weight & (1 << WEP_GETWEIGHT(*posvec_iter)))
{
*dptr = WEP_GETPOS(*posvec_iter);
dptr++;
}
posvec_iter++;
}
data->npos = dptr - data->pos;
if (data->npos > 0)
result = true;
}
else if (val->weight)
{
WordEntryPos *posvec_iter = posvec->pos;
/* Is there a position with a matching weight? */
while (posvec_iter < posvec->pos + posvec->npos)
{
if (val->weight & (1 << WEP_GETWEIGHT(*posvec_iter)))
{
result = true;
break; /* no need to go further */
}
posvec_iter++;
}
}
else /* data != NULL */
{
data->npos = posvec->npos;
data->pos = posvec->pos;
data->allocated = false;
result = true;
}
}
return false;
else
{
result = true;
}
return result;
}
/*
* Removes duplicate pos entries. We can't use uniquePos() from
* tsvector.c because array might be longer than MAXENTRYPOS
*
* Returns new length.
*/
static int
uniqueLongPos(WordEntryPos *pos, int npos)
{
WordEntryPos *pos_iter,
*result;
if (npos <= 1)
return npos;
qsort((void *) pos, npos, sizeof(WordEntryPos), comparePos);
result = pos;
pos_iter = pos + 1;
while (pos_iter < pos + npos)
{
if (WEP_GETPOS(*pos_iter) != WEP_GETPOS(*result))
{
result++;
*result = WEP_GETPOS(*pos_iter);
}
pos_iter++;
}
return result + 1 - pos;
}
/*
* is there value 'val' in array or not ?
*/
static bool
checkcondition_str(void *checkval, QueryOperand *val)
checkcondition_str(void *checkval, QueryOperand *val, ExecPhraseData *data)
{
CHKVAL *chkval = (CHKVAL *) checkval;
WordEntry *StopLow = chkval->arrb;
@ -1162,14 +1251,16 @@ checkcondition_str(void *checkval, QueryOperand *val)
while (StopLow < StopHigh)
{
StopMiddle = StopLow + (StopHigh - StopLow) / 2;
difference = tsCompareString(chkval->operand + val->distance, val->length,
chkval->values + StopMiddle->pos, StopMiddle->len,
difference = tsCompareString(chkval->operand + val->distance,
val->length,
chkval->values + StopMiddle->pos,
StopMiddle->len,
false);
if (difference == 0)
{
res = (val->weight && StopMiddle->haspos) ?
checkclass_str(chkval, StopMiddle, val) : true;
/* Check weight info & fill 'data' with positions */
res = checkclass_str(chkval, StopMiddle, val, data);
break;
}
else if (difference > 0)
@ -1178,30 +1269,199 @@ checkcondition_str(void *checkval, QueryOperand *val)
StopHigh = StopMiddle;
}
if (!res && val->prefix)
if ((!res || data) && val->prefix)
{
WordEntryPos *allpos = NULL;
int npos = 0,
totalpos = 0;
/*
* there was a failed exact search, so we should scan further to find
* a prefix match.
* a prefix match. We also need to do so if caller needs position info
*/
if (StopLow >= StopHigh)
StopMiddle = StopHigh;
while (res == false && StopMiddle < chkval->arre &&
tsCompareString(chkval->operand + val->distance, val->length,
chkval->values + StopMiddle->pos, StopMiddle->len,
while ((!res || data) && StopMiddle < chkval->arre &&
tsCompareString(chkval->operand + val->distance,
val->length,
chkval->values + StopMiddle->pos,
StopMiddle->len,
true) == 0)
{
res = (val->weight && StopMiddle->haspos) ?
checkclass_str(chkval, StopMiddle, val) : true;
if (data)
{
/*
* We need to join position information
*/
res = checkclass_str(chkval, StopMiddle, val, data);
if (res)
{
while (npos + data->npos >= totalpos)
{
if (totalpos == 0)
{
totalpos = 256;
allpos = palloc(sizeof(WordEntryPos) * totalpos);
}
else
{
totalpos *= 2;
allpos = repalloc(allpos, sizeof(WordEntryPos) * totalpos);
}
}
memcpy(allpos + npos, data->pos, sizeof(WordEntryPos) * data->npos);
npos += data->npos;
}
}
else
{
res = checkclass_str(chkval, StopMiddle, val, NULL);
}
StopMiddle++;
}
if (res && data)
{
/* Sort and make unique array of found positions */
data->pos = allpos;
data->npos = uniqueLongPos(allpos, npos);
data->allocated = true;
}
}
return res;
}
/*
* Check for phrase condition. Fallback to the AND operation
* if there is no positional information.
*/
static bool
TS_phrase_execute(QueryItem *curitem,
void *checkval, bool calcnot, ExecPhraseData *data,
bool (*chkcond) (void *, QueryOperand *, ExecPhraseData *))
{
/* since this function recurses, it could be driven to stack overflow */
check_stack_depth();
if (curitem->type == QI_VAL)
{
return chkcond(checkval, (QueryOperand *) curitem, data);
}
else
{
ExecPhraseData Ldata = {0, false, NULL},
Rdata = {0, false, NULL};
WordEntryPos *Lpos,
*Rpos,
*pos_iter = NULL;
Assert(curitem->qoperator.oper == OP_PHRASE);
if (!TS_phrase_execute(curitem + curitem->qoperator.left,
checkval, calcnot, &Ldata, chkcond))
return false;
if (!TS_phrase_execute(curitem + 1, checkval, calcnot, &Rdata, chkcond))
return false;
/*
* if at least one of the operands has no position
* information, fallback to AND operation.
*/
if (Ldata.npos == 0 || Rdata.npos == 0)
return true;
/*
* Result of the operation is a list of the
* corresponding positions of RIGHT operand.
*/
if (data)
{
if (!Rdata.allocated)
/*
* OP_PHRASE is based on the OP_AND, so the number of resulting
* positions could not be greater than the total amount of operands.
*/
data->pos = palloc(sizeof(WordEntryPos) * Min(Ldata.npos, Rdata.npos));
else
data->pos = Rdata.pos;
data->allocated = true;
data->npos = 0;
pos_iter = data->pos;
}
Lpos = Ldata.pos;
Rpos = Rdata.pos;
/*
* Find matches by distance, WEP_GETPOS() is needed because
* ExecPhraseData->data can point to the tsvector's WordEntryPosVector
*/
while (Rpos < Rdata.pos + Rdata.npos)
{
while (Lpos < Ldata.pos + Ldata.npos)
{
if (WEP_GETPOS(*Lpos) <= WEP_GETPOS(*Rpos))
{
/*
* Lpos is behind the Rpos, so we have to check the
* distance condition
*/
if (WEP_GETPOS(*Rpos) - WEP_GETPOS(*Lpos) <= curitem->qoperator.distance)
{
/* MATCH! */
if (data)
{
*pos_iter = WEP_GETPOS(*Rpos);
pos_iter++;
break; /* We need to build a unique result
* array, so go to the next Rpos */
}
else
{
/*
* We are in the root of the phrase tree and hence
* we don't have to store the resulting positions
*/
return true;
}
}
}
else
{
/*
* Go to the next Rpos, because Lpos
* is ahead of the current Rpos
*/
break;
}
Lpos++;
}
Rpos++;
}
if (data)
{
data->npos = pos_iter - data->pos;
if (data->npos > 0)
return true;
}
}
return false;
}
/*
* Evaluate tsquery boolean expression.
*
@ -1210,16 +1470,19 @@ checkcondition_str(void *checkval, QueryOperand *val)
* do anything with it.
* if calcnot is false, NOT expressions are always evaluated to be true. This
* is used in ranking.
* It believes that ordinary operators are always closier to root than phrase
* operator, so, TS_execute() may not take care of lexeme's position at all.
*/
bool
TS_execute(QueryItem *curitem, void *checkval, bool calcnot,
bool (*chkcond) (void *checkval, QueryOperand *val))
bool (*chkcond) (void *checkval, QueryOperand *val, ExecPhraseData *data))
{
/* since this function recurses, it could be driven to stack overflow */
check_stack_depth();
if (curitem->type == QI_VAL)
return chkcond(checkval, (QueryOperand *) curitem);
return chkcond(checkval, (QueryOperand *) curitem,
NULL /* we don't need position info */);
switch (curitem->qoperator.oper)
{
@ -1241,6 +1504,9 @@ TS_execute(QueryItem *curitem, void *checkval, bool calcnot,
else
return TS_execute(curitem + 1, checkval, calcnot, chkcond);
case OP_PHRASE:
return TS_phrase_execute(curitem, checkval, calcnot, NULL, chkcond);
default:
elog(ERROR, "unrecognized operator: %d", curitem->qoperator.oper);
}
@ -1277,6 +1543,10 @@ tsquery_requires_match(QueryItem *curitem)
*/
return false;
case OP_PHRASE:
/*
* Treat OP_PHRASE as OP_AND here
*/
case OP_AND:
/* If either side requires a match, we're good */
if (tsquery_requires_match(curitem + curitem->qoperator.left))

10
src/backend/utils/adt/tsvector_parser.c
View File

@ -89,7 +89,15 @@ do { \
} \
} while (0)
#define ISOPERATOR(x) ( pg_mblen(x)==1 && ( *(x)=='!' || *(x)=='&' || *(x)=='|' || *(x)=='(' || *(x)==')' ) )
/* phrase operator begins with '<' */
#define ISOPERATOR(x) \
( pg_mblen(x) == 1 && ( *(x) == '!' || \
*(x) == '&' || \
*(x) == '|' || \
*(x) == '(' || \
*(x) == ')' || \
*(x) == '<' \
) )
/* Fills gettoken_tsvector's output parameters, and returns true */
#define RETURN_TOKEN \