Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * jsonb_gin.c
4 : * GIN support functions for jsonb
5 : *
6 : * Copyright (c) 2014-2023, PostgreSQL Global Development Group
7 : *
8 : * We provide two opclasses for jsonb indexing: jsonb_ops and jsonb_path_ops.
9 : * For their description see json.sgml and comments in jsonb.h.
10 : *
11 : * The operators support, among the others, "jsonb @? jsonpath" and
12 : * "jsonb @@ jsonpath". Expressions containing these operators are easily
13 : * expressed through each other.
14 : *
15 : * jb @? 'path' <=> jb @@ 'EXISTS(path)'
16 : * jb @@ 'expr' <=> jb @? '$ ? (expr)'
17 : *
18 : * Thus, we're going to consider only @@ operator, while regarding @? operator
19 : * the same is true for jb @@ 'EXISTS(path)'.
20 : *
21 : * Result of jsonpath query extraction is a tree, which leaf nodes are index
22 : * entries and non-leaf nodes are AND/OR logical expressions. Basically we
23 : * extract following statements out of jsonpath:
24 : *
25 : * 1) "accessors_chain = const",
26 : * 2) "EXISTS(accessors_chain)".
27 : *
28 : * Accessors chain may consist of .key, [*] and [index] accessors. jsonb_ops
29 : * additionally supports .* and .**.
30 : *
31 : * For now, both jsonb_ops and jsonb_path_ops supports only statements of
32 : * the 1st find. jsonb_ops might also support statements of the 2nd kind,
33 : * but given we have no statistics keys extracted from accessors chain
34 : * are likely non-selective. Therefore, we choose to not confuse optimizer
35 : * and skip statements of the 2nd kind altogether. In future versions that
36 : * might be changed.
37 : *
38 : * In jsonb_ops statement of the 1st kind is split into expression of AND'ed
39 : * keys and const. Sometimes const might be interpreted as both value or key
40 : * in jsonb_ops. Then statement of 1st kind is decomposed into the expression
41 : * below.
42 : *
43 : * key1 AND key2 AND ... AND keyN AND (const_as_value OR const_as_key)
44 : *
45 : * jsonb_path_ops transforms each statement of the 1st kind into single hash
46 : * entry below.
47 : *
48 : * HASH(key1, key2, ... , keyN, const)
49 : *
50 : * Despite statements of the 2nd kind are not supported by both jsonb_ops and
51 : * jsonb_path_ops, EXISTS(path) expressions might be still supported,
52 : * when statements of 1st kind could be extracted out of their filters.
53 : *
54 : * IDENTIFICATION
55 : * src/backend/utils/adt/jsonb_gin.c
56 : *
57 : *-------------------------------------------------------------------------
58 : */
59 :
60 : #include "postgres.h"
61 :
62 : #include "access/gin.h"
63 : #include "access/stratnum.h"
64 : #include "catalog/pg_collation.h"
65 : #include "catalog/pg_type.h"
66 : #include "common/hashfn.h"
67 : #include "miscadmin.h"
68 : #include "utils/builtins.h"
69 : #include "utils/jsonb.h"
70 : #include "utils/jsonpath.h"
71 : #include "utils/varlena.h"
72 :
73 : typedef struct PathHashStack
74 : {
75 : uint32 hash;
76 : struct PathHashStack *parent;
77 : } PathHashStack;
78 :
79 : /* Buffer for GIN entries */
80 : typedef struct GinEntries
81 : {
82 : Datum *buf;
83 : int count;
84 : int allocated;
85 : } GinEntries;
86 :
87 : typedef enum JsonPathGinNodeType
88 : {
89 : JSP_GIN_OR,
90 : JSP_GIN_AND,
91 : JSP_GIN_ENTRY,
92 : } JsonPathGinNodeType;
93 :
94 : typedef struct JsonPathGinNode JsonPathGinNode;
95 :
96 : /* Node in jsonpath expression tree */
97 : struct JsonPathGinNode
98 : {
99 : JsonPathGinNodeType type;
100 : union
101 : {
102 : int nargs; /* valid for OR and AND nodes */
103 : int entryIndex; /* index in GinEntries array, valid for ENTRY
104 : * nodes after entries output */
105 : Datum entryDatum; /* path hash or key name/scalar, valid for
106 : * ENTRY nodes before entries output */
107 : } val;
108 : JsonPathGinNode *args[FLEXIBLE_ARRAY_MEMBER]; /* valid for OR and AND
109 : * nodes */
110 : };
111 :
112 : /*
113 : * jsonb_ops entry extracted from jsonpath item. Corresponding path item
114 : * may be: '.key', '.*', '.**', '[index]' or '[*]'.
115 : * Entry type is stored in 'type' field.
116 : */
117 : typedef struct JsonPathGinPathItem
118 : {
119 : struct JsonPathGinPathItem *parent;
120 : Datum keyName; /* key name (for '.key' path item) or NULL */
121 : JsonPathItemType type; /* type of jsonpath item */
122 : } JsonPathGinPathItem;
123 :
124 : /* GIN representation of the extracted json path */
125 : typedef union JsonPathGinPath
126 : {
127 : JsonPathGinPathItem *items; /* list of path items (jsonb_ops) */
128 : uint32 hash; /* hash of the path (jsonb_path_ops) */
129 : } JsonPathGinPath;
130 :
131 : typedef struct JsonPathGinContext JsonPathGinContext;
132 :
133 : /* Callback, which stores information about path item into JsonPathGinPath */
134 : typedef bool (*JsonPathGinAddPathItemFunc) (JsonPathGinPath *path,
135 : JsonPathItem *jsp);
136 :
137 : /*
138 : * Callback, which extracts set of nodes from statement of 1st kind
139 : * (scalar != NULL) or statement of 2nd kind (scalar == NULL).
140 : */
141 : typedef List *(*JsonPathGinExtractNodesFunc) (JsonPathGinContext *cxt,
142 : JsonPathGinPath path,
143 : JsonbValue *scalar,
144 : List *nodes);
145 :
146 : /* Context for jsonpath entries extraction */
147 : struct JsonPathGinContext
148 : {
149 : JsonPathGinAddPathItemFunc add_path_item;
150 : JsonPathGinExtractNodesFunc extract_nodes;
151 : bool lax;
152 : };
153 :
154 : static Datum make_text_key(char flag, const char *str, int len);
155 : static Datum make_scalar_key(const JsonbValue *scalarVal, bool is_key);
156 :
157 : static JsonPathGinNode *extract_jsp_bool_expr(JsonPathGinContext *cxt,
158 : JsonPathGinPath path, JsonPathItem *jsp, bool not);
159 :
160 :
161 : /* Initialize GinEntries struct */
162 : static void
163 11986 : init_gin_entries(GinEntries *entries, int preallocated)
164 : {
165 11986 : entries->allocated = preallocated;
166 11986 : entries->buf = preallocated ? palloc(sizeof(Datum) * preallocated) : NULL;
167 11986 : entries->count = 0;
168 11986 : }
169 :
170 : /* Add new entry to GinEntries */
171 : static int
172 93762 : add_gin_entry(GinEntries *entries, Datum entry)
173 : {
174 93762 : int id = entries->count;
175 :
176 93762 : if (entries->count >= entries->allocated)
177 : {
178 564 : if (entries->allocated)
179 : {
180 66 : entries->allocated *= 2;
181 66 : entries->buf = repalloc(entries->buf,
182 66 : sizeof(Datum) * entries->allocated);
183 : }
184 : else
185 : {
186 498 : entries->allocated = 8;
187 498 : entries->buf = palloc(sizeof(Datum) * entries->allocated);
188 : }
189 : }
190 :
191 93762 : entries->buf[entries->count++] = entry;
192 :
193 93762 : return id;
194 : }
195 :
196 : /*
197 : *
198 : * jsonb_ops GIN opclass support functions
199 : *
200 : */
201 :
202 : Datum
203 759558 : gin_compare_jsonb(PG_FUNCTION_ARGS)
204 : {
205 759558 : text *arg1 = PG_GETARG_TEXT_PP(0);
206 759558 : text *arg2 = PG_GETARG_TEXT_PP(1);
207 : int32 result;
208 : char *a1p,
209 : *a2p;
210 : int len1,
211 : len2;
212 :
213 759558 : a1p = VARDATA_ANY(arg1);
214 759558 : a2p = VARDATA_ANY(arg2);
215 :
216 759558 : len1 = VARSIZE_ANY_EXHDR(arg1);
217 759558 : len2 = VARSIZE_ANY_EXHDR(arg2);
218 :
219 : /* Compare text as bttextcmp does, but always using C collation */
220 759558 : result = varstr_cmp(a1p, len1, a2p, len2, C_COLLATION_OID);
221 :
222 759558 : PG_FREE_IF_COPY(arg1, 0);
223 759558 : PG_FREE_IF_COPY(arg2, 1);
224 :
225 759558 : PG_RETURN_INT32(result);
226 : }
227 :
228 : Datum
229 7210 : gin_extract_jsonb(PG_FUNCTION_ARGS)
230 : {
231 7210 : Jsonb *jb = (Jsonb *) PG_GETARG_JSONB_P(0);
232 7210 : int32 *nentries = (int32 *) PG_GETARG_POINTER(1);
233 7210 : int total = JB_ROOT_COUNT(jb);
234 : JsonbIterator *it;
235 : JsonbValue v;
236 : JsonbIteratorToken r;
237 : GinEntries entries;
238 :
239 : /* If the root level is empty, we certainly have no keys */
240 7210 : if (total == 0)
241 : {
242 720 : *nentries = 0;
243 720 : PG_RETURN_POINTER(NULL);
244 : }
245 :
246 : /* Otherwise, use 2 * root count as initial estimate of result size */
247 6490 : init_gin_entries(&entries, 2 * total);
248 :
249 6490 : it = JsonbIteratorInit(&jb->root);
250 :
251 83430 : while ((r = JsonbIteratorNext(&it, &v, false)) != WJB_DONE)
252 : {
253 76940 : switch (r)
254 : {
255 31860 : case WJB_KEY:
256 31860 : add_gin_entry(&entries, make_scalar_key(&v, true));
257 31860 : break;
258 168 : case WJB_ELEM:
259 : /* Pretend string array elements are keys, see jsonb.h */
260 168 : add_gin_entry(&entries, make_scalar_key(&v, v.type == jbvString));
261 168 : break;
262 31788 : case WJB_VALUE:
263 31788 : add_gin_entry(&entries, make_scalar_key(&v, false));
264 31788 : break;
265 13124 : default:
266 : /* we can ignore structural items */
267 13124 : break;
268 : }
269 : }
270 :
271 6490 : *nentries = entries.count;
272 :
273 6490 : PG_RETURN_POINTER(entries.buf);
274 : }
275 :
276 : /* Append JsonPathGinPathItem to JsonPathGinPath (jsonb_ops) */
277 : static bool
278 852 : jsonb_ops__add_path_item(JsonPathGinPath *path, JsonPathItem *jsp)
279 : {
280 : JsonPathGinPathItem *pentry;
281 : Datum keyName;
282 :
283 852 : switch (jsp->type)
284 : {
285 384 : case jpiRoot:
286 384 : path->items = NULL; /* reset path */
287 384 : return true;
288 :
289 372 : case jpiKey:
290 : {
291 : int len;
292 372 : char *key = jspGetString(jsp, &len);
293 :
294 372 : keyName = make_text_key(JGINFLAG_KEY, key, len);
295 372 : break;
296 : }
297 :
298 96 : case jpiAny:
299 : case jpiAnyKey:
300 : case jpiAnyArray:
301 : case jpiIndexArray:
302 96 : keyName = PointerGetDatum(NULL);
303 96 : break;
304 :
305 0 : default:
306 : /* other path items like item methods are not supported */
307 0 : return false;
308 : }
309 :
310 468 : pentry = palloc(sizeof(*pentry));
311 :
312 468 : pentry->type = jsp->type;
313 468 : pentry->keyName = keyName;
314 468 : pentry->parent = path->items;
315 :
316 468 : path->items = pentry;
317 :
318 468 : return true;
319 : }
320 :
321 : /* Combine existing path hash with next key hash (jsonb_path_ops) */
322 : static bool
323 696 : jsonb_path_ops__add_path_item(JsonPathGinPath *path, JsonPathItem *jsp)
324 : {
325 696 : switch (jsp->type)
326 : {
327 306 : case jpiRoot:
328 306 : path->hash = 0; /* reset path hash */
329 306 : return true;
330 :
331 294 : case jpiKey:
332 : {
333 : JsonbValue jbv;
334 :
335 294 : jbv.type = jbvString;
336 294 : jbv.val.string.val = jspGetString(jsp, &jbv.val.string.len);
337 :
338 294 : JsonbHashScalarValue(&jbv, &path->hash);
339 294 : return true;
340 : }
341 :
342 96 : case jpiIndexArray:
343 : case jpiAnyArray:
344 96 : return true; /* path hash is unchanged */
345 :
346 0 : default:
347 : /* other items (wildcard paths, item methods) are not supported */
348 0 : return false;
349 : }
350 : }
351 :
352 : static JsonPathGinNode *
353 966 : make_jsp_entry_node(Datum entry)
354 : {
355 966 : JsonPathGinNode *node = palloc(offsetof(JsonPathGinNode, args));
356 :
357 966 : node->type = JSP_GIN_ENTRY;
358 966 : node->val.entryDatum = entry;
359 :
360 966 : return node;
361 : }
362 :
363 : static JsonPathGinNode *
364 420 : make_jsp_entry_node_scalar(JsonbValue *scalar, bool iskey)
365 : {
366 420 : return make_jsp_entry_node(make_scalar_key(scalar, iskey));
367 : }
368 :
369 : static JsonPathGinNode *
370 468 : make_jsp_expr_node(JsonPathGinNodeType type, int nargs)
371 : {
372 468 : JsonPathGinNode *node = palloc(offsetof(JsonPathGinNode, args) +
373 : sizeof(node->args[0]) * nargs);
374 :
375 468 : node->type = type;
376 468 : node->val.nargs = nargs;
377 :
378 468 : return node;
379 : }
380 :
381 : static JsonPathGinNode *
382 276 : make_jsp_expr_node_args(JsonPathGinNodeType type, List *args)
383 : {
384 276 : JsonPathGinNode *node = make_jsp_expr_node(type, list_length(args));
385 : ListCell *lc;
386 276 : int i = 0;
387 :
388 828 : foreach(lc, args)
389 552 : node->args[i++] = lfirst(lc);
390 :
391 276 : return node;
392 : }
393 :
394 : static JsonPathGinNode *
395 192 : make_jsp_expr_node_binary(JsonPathGinNodeType type,
396 : JsonPathGinNode *arg1, JsonPathGinNode *arg2)
397 : {
398 192 : JsonPathGinNode *node = make_jsp_expr_node(type, 2);
399 :
400 192 : node->args[0] = arg1;
401 192 : node->args[1] = arg2;
402 :
403 192 : return node;
404 : }
405 :
406 : /* Append a list of nodes from the jsonpath (jsonb_ops). */
407 : static List *
408 558 : jsonb_ops__extract_nodes(JsonPathGinContext *cxt, JsonPathGinPath path,
409 : JsonbValue *scalar, List *nodes)
410 : {
411 : JsonPathGinPathItem *pentry;
412 :
413 558 : if (scalar)
414 : {
415 : JsonPathGinNode *node;
416 :
417 : /*
418 : * Append path entry nodes only if scalar is provided. See header
419 : * comment for details.
420 : */
421 648 : for (pentry = path.items; pentry; pentry = pentry->parent)
422 : {
423 372 : if (pentry->type == jpiKey) /* only keys are indexed */
424 276 : nodes = lappend(nodes, make_jsp_entry_node(pentry->keyName));
425 : }
426 :
427 : /* Append scalar node for equality queries. */
428 276 : if (scalar->type == jbvString)
429 : {
430 144 : JsonPathGinPathItem *last = path.items;
431 : GinTernaryValue key_entry;
432 :
433 : /*
434 : * Assuming that jsonb_ops interprets string array elements as
435 : * keys, we may extract key or non-key entry or even both. In the
436 : * latter case we create OR-node. It is possible in lax mode
437 : * where arrays are automatically unwrapped, or in strict mode for
438 : * jpiAny items.
439 : */
440 :
441 144 : if (cxt->lax)
442 144 : key_entry = GIN_MAYBE;
443 0 : else if (!last) /* root ($) */
444 0 : key_entry = GIN_FALSE;
445 0 : else if (last->type == jpiAnyArray || last->type == jpiIndexArray)
446 0 : key_entry = GIN_TRUE;
447 0 : else if (last->type == jpiAny)
448 0 : key_entry = GIN_MAYBE;
449 : else
450 0 : key_entry = GIN_FALSE;
451 :
452 144 : if (key_entry == GIN_MAYBE)
453 : {
454 144 : JsonPathGinNode *n1 = make_jsp_entry_node_scalar(scalar, true);
455 144 : JsonPathGinNode *n2 = make_jsp_entry_node_scalar(scalar, false);
456 :
457 144 : node = make_jsp_expr_node_binary(JSP_GIN_OR, n1, n2);
458 : }
459 : else
460 : {
461 0 : node = make_jsp_entry_node_scalar(scalar,
462 : key_entry == GIN_TRUE);
463 : }
464 : }
465 : else
466 : {
467 132 : node = make_jsp_entry_node_scalar(scalar, false);
468 : }
469 :
470 276 : nodes = lappend(nodes, node);
471 : }
472 :
473 558 : return nodes;
474 : }
475 :
476 : /* Append a list of nodes from the jsonpath (jsonb_path_ops). */
477 : static List *
478 480 : jsonb_path_ops__extract_nodes(JsonPathGinContext *cxt, JsonPathGinPath path,
479 : JsonbValue *scalar, List *nodes)
480 : {
481 480 : if (scalar)
482 : {
483 : /* append path hash node for equality queries */
484 270 : uint32 hash = path.hash;
485 :
486 270 : JsonbHashScalarValue(scalar, &hash);
487 :
488 270 : return lappend(nodes,
489 270 : make_jsp_entry_node(UInt32GetDatum(hash)));
490 : }
491 : else
492 : {
493 : /* jsonb_path_ops doesn't support EXISTS queries => nothing to append */
494 210 : return nodes;
495 : }
496 : }
497 :
498 : /*
499 : * Extract a list of expression nodes that need to be AND-ed by the caller.
500 : * Extracted expression is 'path == scalar' if 'scalar' is non-NULL, and
501 : * 'EXISTS(path)' otherwise.
502 : */
503 : static List *
504 1038 : extract_jsp_path_expr_nodes(JsonPathGinContext *cxt, JsonPathGinPath path,
505 : JsonPathItem *jsp, JsonbValue *scalar)
506 : {
507 : JsonPathItem next;
508 1038 : List *nodes = NIL;
509 :
510 : for (;;)
511 : {
512 2220 : switch (jsp->type)
513 : {
514 348 : case jpiCurrent:
515 348 : break;
516 :
517 324 : case jpiFilter:
518 : {
519 : JsonPathItem arg;
520 : JsonPathGinNode *filter;
521 :
522 324 : jspGetArg(jsp, &arg);
523 :
524 324 : filter = extract_jsp_bool_expr(cxt, path, &arg, false);
525 :
526 324 : if (filter)
527 324 : nodes = lappend(nodes, filter);
528 :
529 324 : break;
530 : }
531 :
532 1548 : default:
533 1548 : if (!cxt->add_path_item(&path, jsp))
534 :
535 : /*
536 : * Path is not supported by the index opclass, return only
537 : * the extracted filter nodes.
538 : */
539 0 : return nodes;
540 1548 : break;
541 : }
542 :
543 2220 : if (!jspGetNext(jsp, &next))
544 1038 : break;
545 :
546 1182 : jsp = &next;
547 : }
548 :
549 : /*
550 : * Append nodes from the path expression itself to the already extracted
551 : * list of filter nodes.
552 : */
553 1038 : return cxt->extract_nodes(cxt, path, scalar, nodes);
554 : }
555 :
556 : /*
557 : * Extract an expression node from one of following jsonpath path expressions:
558 : * EXISTS(jsp) (when 'scalar' is NULL)
559 : * jsp == scalar (when 'scalar' is not NULL).
560 : *
561 : * The current path (@) is passed in 'path'.
562 : */
563 : static JsonPathGinNode *
564 1038 : extract_jsp_path_expr(JsonPathGinContext *cxt, JsonPathGinPath path,
565 : JsonPathItem *jsp, JsonbValue *scalar)
566 : {
567 : /* extract a list of nodes to be AND-ed */
568 1038 : List *nodes = extract_jsp_path_expr_nodes(cxt, path, jsp, scalar);
569 :
570 1038 : if (nodes == NIL)
571 : /* no nodes were extracted => full scan is needed for this path */
572 168 : return NULL;
573 :
574 870 : if (list_length(nodes) == 1)
575 594 : return linitial(nodes); /* avoid extra AND-node */
576 :
577 : /* construct AND-node for path with filters */
578 276 : return make_jsp_expr_node_args(JSP_GIN_AND, nodes);
579 : }
580 :
581 : /* Recursively extract nodes from the boolean jsonpath expression. */
582 : static JsonPathGinNode *
583 834 : extract_jsp_bool_expr(JsonPathGinContext *cxt, JsonPathGinPath path,
584 : JsonPathItem *jsp, bool not)
585 : {
586 834 : check_stack_depth();
587 :
588 834 : switch (jsp->type)
589 : {
590 72 : case jpiAnd: /* expr && expr */
591 : case jpiOr: /* expr || expr */
592 : {
593 : JsonPathItem arg;
594 : JsonPathGinNode *larg;
595 : JsonPathGinNode *rarg;
596 : JsonPathGinNodeType type;
597 :
598 72 : jspGetLeftArg(jsp, &arg);
599 72 : larg = extract_jsp_bool_expr(cxt, path, &arg, not);
600 :
601 72 : jspGetRightArg(jsp, &arg);
602 72 : rarg = extract_jsp_bool_expr(cxt, path, &arg, not);
603 :
604 72 : if (!larg || !rarg)
605 : {
606 24 : if (jsp->type == jpiOr)
607 12 : return NULL;
608 :
609 12 : return larg ? larg : rarg;
610 : }
611 :
612 48 : type = not ^ (jsp->type == jpiAnd) ? JSP_GIN_AND : JSP_GIN_OR;
613 :
614 48 : return make_jsp_expr_node_binary(type, larg, rarg);
615 : }
616 :
617 0 : case jpiNot: /* !expr */
618 : {
619 : JsonPathItem arg;
620 :
621 0 : jspGetArg(jsp, &arg);
622 :
623 : /* extract child expression inverting 'not' flag */
624 0 : return extract_jsp_bool_expr(cxt, path, &arg, !not);
625 : }
626 :
627 216 : case jpiExists: /* EXISTS(path) */
628 : {
629 : JsonPathItem arg;
630 :
631 216 : if (not)
632 0 : return NULL; /* NOT EXISTS is not supported */
633 :
634 216 : jspGetArg(jsp, &arg);
635 :
636 216 : return extract_jsp_path_expr(cxt, path, &arg, NULL);
637 : }
638 :
639 0 : case jpiNotEqual:
640 :
641 : /*
642 : * 'not' == true case is not supported here because '!(path !=
643 : * scalar)' is not equivalent to 'path == scalar' in the general
644 : * case because of sequence comparison semantics: 'path == scalar'
645 : * === 'EXISTS (path, @ == scalar)', '!(path != scalar)' ===
646 : * 'FOR_ALL(path, @ == scalar)'. So, we should translate '!(path
647 : * != scalar)' into GIN query 'path == scalar || EMPTY(path)', but
648 : * 'EMPTY(path)' queries are not supported by the both jsonb
649 : * opclasses. However in strict mode we could omit 'EMPTY(path)'
650 : * part if the path can return exactly one item (it does not
651 : * contain wildcard accessors or item methods like .keyvalue()
652 : * etc.).
653 : */
654 0 : return NULL;
655 :
656 546 : case jpiEqual: /* path == scalar */
657 : {
658 : JsonPathItem left_item;
659 : JsonPathItem right_item;
660 : JsonPathItem *path_item;
661 : JsonPathItem *scalar_item;
662 : JsonbValue scalar;
663 :
664 546 : if (not)
665 0 : return NULL;
666 :
667 546 : jspGetLeftArg(jsp, &left_item);
668 546 : jspGetRightArg(jsp, &right_item);
669 :
670 546 : if (jspIsScalar(left_item.type))
671 : {
672 96 : scalar_item = &left_item;
673 96 : path_item = &right_item;
674 : }
675 450 : else if (jspIsScalar(right_item.type))
676 : {
677 450 : scalar_item = &right_item;
678 450 : path_item = &left_item;
679 : }
680 : else
681 0 : return NULL; /* at least one operand should be a scalar */
682 :
683 546 : switch (scalar_item->type)
684 : {
685 114 : case jpiNull:
686 114 : scalar.type = jbvNull;
687 114 : break;
688 48 : case jpiBool:
689 48 : scalar.type = jbvBool;
690 48 : scalar.val.boolean = !!*scalar_item->content.value.data;
691 48 : break;
692 96 : case jpiNumeric:
693 96 : scalar.type = jbvNumeric;
694 96 : scalar.val.numeric =
695 96 : (Numeric) scalar_item->content.value.data;
696 96 : break;
697 288 : case jpiString:
698 288 : scalar.type = jbvString;
699 288 : scalar.val.string.val = scalar_item->content.value.data;
700 288 : scalar.val.string.len =
701 288 : scalar_item->content.value.datalen;
702 288 : break;
703 0 : default:
704 0 : elog(ERROR, "invalid scalar jsonpath item type: %d",
705 : scalar_item->type);
706 : return NULL;
707 : }
708 :
709 546 : return extract_jsp_path_expr(cxt, path, path_item, &scalar);
710 : }
711 :
712 0 : default:
713 0 : return NULL; /* not a boolean expression */
714 : }
715 : }
716 :
717 : /* Recursively emit all GIN entries found in the node tree */
718 : static void
719 1434 : emit_jsp_gin_entries(JsonPathGinNode *node, GinEntries *entries)
720 : {
721 1434 : check_stack_depth();
722 :
723 1434 : switch (node->type)
724 : {
725 966 : case JSP_GIN_ENTRY:
726 : /* replace datum with its index in the array */
727 966 : node->val.entryIndex = add_gin_entry(entries, node->val.entryDatum);
728 966 : break;
729 :
730 468 : case JSP_GIN_OR:
731 : case JSP_GIN_AND:
732 : {
733 : int i;
734 :
735 1404 : for (i = 0; i < node->val.nargs; i++)
736 936 : emit_jsp_gin_entries(node->args[i], entries);
737 :
738 468 : break;
739 : }
740 : }
741 1434 : }
742 :
743 : /*
744 : * Recursively extract GIN entries from jsonpath query.
745 : * Root expression node is put into (*extra_data)[0].
746 : */
747 : static Datum *
748 642 : extract_jsp_query(JsonPath *jp, StrategyNumber strat, bool pathOps,
749 : int32 *nentries, Pointer **extra_data)
750 : {
751 : JsonPathGinContext cxt;
752 : JsonPathItem root;
753 : JsonPathGinNode *node;
754 642 : JsonPathGinPath path = {0};
755 642 : GinEntries entries = {0};
756 :
757 642 : cxt.lax = (jp->header & JSONPATH_LAX) != 0;
758 :
759 642 : if (pathOps)
760 : {
761 294 : cxt.add_path_item = jsonb_path_ops__add_path_item;
762 294 : cxt.extract_nodes = jsonb_path_ops__extract_nodes;
763 : }
764 : else
765 : {
766 348 : cxt.add_path_item = jsonb_ops__add_path_item;
767 348 : cxt.extract_nodes = jsonb_ops__extract_nodes;
768 : }
769 :
770 642 : jspInit(&root, jp);
771 :
772 642 : node = strat == JsonbJsonpathExistsStrategyNumber
773 276 : ? extract_jsp_path_expr(&cxt, path, &root, NULL)
774 642 : : extract_jsp_bool_expr(&cxt, path, &root, false);
775 :
776 642 : if (!node)
777 : {
778 144 : *nentries = 0;
779 144 : return NULL;
780 : }
781 :
782 498 : emit_jsp_gin_entries(node, &entries);
783 :
784 498 : *nentries = entries.count;
785 498 : if (!*nentries)
786 0 : return NULL;
787 :
788 498 : *extra_data = palloc0(sizeof(**extra_data) * entries.count);
789 498 : **extra_data = (Pointer) node;
790 :
791 498 : return entries.buf;
792 : }
793 :
794 : /*
795 : * Recursively execute jsonpath expression.
796 : * 'check' is a bool[] or a GinTernaryValue[] depending on 'ternary' flag.
797 : */
798 : static GinTernaryValue
799 12612 : execute_jsp_gin_node(JsonPathGinNode *node, void *check, bool ternary)
800 : {
801 : GinTernaryValue res;
802 : GinTernaryValue v;
803 : int i;
804 :
805 12612 : switch (node->type)
806 : {
807 5088 : case JSP_GIN_AND:
808 5088 : res = GIN_TRUE;
809 8064 : for (i = 0; i < node->val.nargs; i++)
810 : {
811 6888 : v = execute_jsp_gin_node(node->args[i], check, ternary);
812 6888 : if (v == GIN_FALSE)
813 3912 : return GIN_FALSE;
814 2976 : else if (v == GIN_MAYBE)
815 240 : res = GIN_MAYBE;
816 : }
817 1176 : return res;
818 :
819 1056 : case JSP_GIN_OR:
820 1056 : res = GIN_FALSE;
821 2160 : for (i = 0; i < node->val.nargs; i++)
822 : {
823 1968 : v = execute_jsp_gin_node(node->args[i], check, ternary);
824 1968 : if (v == GIN_TRUE)
825 864 : return GIN_TRUE;
826 1104 : else if (v == GIN_MAYBE)
827 72 : res = GIN_MAYBE;
828 : }
829 192 : return res;
830 :
831 6468 : case JSP_GIN_ENTRY:
832 : {
833 6468 : int index = node->val.entryIndex;
834 :
835 6468 : if (ternary)
836 6468 : return ((GinTernaryValue *) check)[index];
837 : else
838 0 : return ((bool *) check)[index] ? GIN_TRUE : GIN_FALSE;
839 : }
840 :
841 0 : default:
842 0 : elog(ERROR, "invalid jsonpath gin node type: %d", node->type);
843 : return GIN_FALSE; /* keep compiler quiet */
844 : }
845 : }
846 :
847 : Datum
848 528 : gin_extract_jsonb_query(PG_FUNCTION_ARGS)
849 : {
850 528 : int32 *nentries = (int32 *) PG_GETARG_POINTER(1);
851 528 : StrategyNumber strategy = PG_GETARG_UINT16(2);
852 528 : int32 *searchMode = (int32 *) PG_GETARG_POINTER(6);
853 : Datum *entries;
854 :
855 528 : if (strategy == JsonbContainsStrategyNumber)
856 : {
857 : /* Query is a jsonb, so just apply gin_extract_jsonb... */
858 : entries = (Datum *)
859 108 : DatumGetPointer(DirectFunctionCall2(gin_extract_jsonb,
860 : PG_GETARG_DATUM(0),
861 : PointerGetDatum(nentries)));
862 : /* ...although "contains {}" requires a full index scan */
863 108 : if (*nentries == 0)
864 12 : *searchMode = GIN_SEARCH_MODE_ALL;
865 : }
866 420 : else if (strategy == JsonbExistsStrategyNumber)
867 : {
868 : /* Query is a text string, which we treat as a key */
869 48 : text *query = PG_GETARG_TEXT_PP(0);
870 :
871 48 : *nentries = 1;
872 48 : entries = (Datum *) palloc(sizeof(Datum));
873 96 : entries[0] = make_text_key(JGINFLAG_KEY,
874 48 : VARDATA_ANY(query),
875 48 : VARSIZE_ANY_EXHDR(query));
876 : }
877 372 : else if (strategy == JsonbExistsAnyStrategyNumber ||
878 : strategy == JsonbExistsAllStrategyNumber)
879 24 : {
880 : /* Query is a text array; each element is treated as a key */
881 24 : ArrayType *query = PG_GETARG_ARRAYTYPE_P(0);
882 : Datum *key_datums;
883 : bool *key_nulls;
884 : int key_count;
885 : int i,
886 : j;
887 :
888 24 : deconstruct_array_builtin(query, TEXTOID, &key_datums, &key_nulls, &key_count);
889 :
890 24 : entries = (Datum *) palloc(sizeof(Datum) * key_count);
891 :
892 72 : for (i = 0, j = 0; i < key_count; i++)
893 : {
894 : /* Nulls in the array are ignored */
895 48 : if (key_nulls[i])
896 0 : continue;
897 : /* We rely on the array elements not being toasted */
898 96 : entries[j++] = make_text_key(JGINFLAG_KEY,
899 48 : VARDATA_ANY(key_datums[i]),
900 48 : VARSIZE_ANY_EXHDR(key_datums[i]));
901 : }
902 :
903 24 : *nentries = j;
904 : /* ExistsAll with no keys should match everything */
905 24 : if (j == 0 && strategy == JsonbExistsAllStrategyNumber)
906 0 : *searchMode = GIN_SEARCH_MODE_ALL;
907 : }
908 348 : else if (strategy == JsonbJsonpathPredicateStrategyNumber ||
909 : strategy == JsonbJsonpathExistsStrategyNumber)
910 348 : {
911 348 : JsonPath *jp = PG_GETARG_JSONPATH_P(0);
912 348 : Pointer **extra_data = (Pointer **) PG_GETARG_POINTER(4);
913 :
914 348 : entries = extract_jsp_query(jp, strategy, false, nentries, extra_data);
915 :
916 348 : if (!entries)
917 96 : *searchMode = GIN_SEARCH_MODE_ALL;
918 : }
919 : else
920 : {
921 0 : elog(ERROR, "unrecognized strategy number: %d", strategy);
922 : entries = NULL; /* keep compiler quiet */
923 : }
924 :
925 528 : PG_RETURN_POINTER(entries);
926 : }
927 :
928 : Datum
929 0 : gin_consistent_jsonb(PG_FUNCTION_ARGS)
930 : {
931 0 : bool *check = (bool *) PG_GETARG_POINTER(0);
932 0 : StrategyNumber strategy = PG_GETARG_UINT16(1);
933 :
934 : /* Jsonb *query = PG_GETARG_JSONB_P(2); */
935 0 : int32 nkeys = PG_GETARG_INT32(3);
936 :
937 0 : Pointer *extra_data = (Pointer *) PG_GETARG_POINTER(4);
938 0 : bool *recheck = (bool *) PG_GETARG_POINTER(5);
939 0 : bool res = true;
940 : int32 i;
941 :
942 0 : if (strategy == JsonbContainsStrategyNumber)
943 : {
944 : /*
945 : * We must always recheck, since we can't tell from the index whether
946 : * the positions of the matched items match the structure of the query
947 : * object. (Even if we could, we'd also have to worry about hashed
948 : * keys and the index's failure to distinguish keys from string array
949 : * elements.) However, the tuple certainly doesn't match unless it
950 : * contains all the query keys.
951 : */
952 0 : *recheck = true;
953 0 : for (i = 0; i < nkeys; i++)
954 : {
955 0 : if (!check[i])
956 : {
957 0 : res = false;
958 0 : break;
959 : }
960 : }
961 : }
962 0 : else if (strategy == JsonbExistsStrategyNumber)
963 : {
964 : /*
965 : * Although the key is certainly present in the index, we must recheck
966 : * because (1) the key might be hashed, and (2) the index match might
967 : * be for a key that's not at top level of the JSON object. For (1),
968 : * we could look at the query key to see if it's hashed and not
969 : * recheck if not, but the index lacks enough info to tell about (2).
970 : */
971 0 : *recheck = true;
972 0 : res = true;
973 : }
974 0 : else if (strategy == JsonbExistsAnyStrategyNumber)
975 : {
976 : /* As for plain exists, we must recheck */
977 0 : *recheck = true;
978 0 : res = true;
979 : }
980 0 : else if (strategy == JsonbExistsAllStrategyNumber)
981 : {
982 : /* As for plain exists, we must recheck */
983 0 : *recheck = true;
984 : /* ... but unless all the keys are present, we can say "false" */
985 0 : for (i = 0; i < nkeys; i++)
986 : {
987 0 : if (!check[i])
988 : {
989 0 : res = false;
990 0 : break;
991 : }
992 : }
993 : }
994 0 : else if (strategy == JsonbJsonpathPredicateStrategyNumber ||
995 : strategy == JsonbJsonpathExistsStrategyNumber)
996 : {
997 0 : *recheck = true;
998 :
999 0 : if (nkeys > 0)
1000 : {
1001 : Assert(extra_data && extra_data[0]);
1002 0 : res = execute_jsp_gin_node((JsonPathGinNode *) extra_data[0], check,
1003 : false) != GIN_FALSE;
1004 : }
1005 : }
1006 : else
1007 0 : elog(ERROR, "unrecognized strategy number: %d", strategy);
1008 :
1009 0 : PG_RETURN_BOOL(res);
1010 : }
1011 :
1012 : Datum
1013 63642 : gin_triconsistent_jsonb(PG_FUNCTION_ARGS)
1014 : {
1015 63642 : GinTernaryValue *check = (GinTernaryValue *) PG_GETARG_POINTER(0);
1016 63642 : StrategyNumber strategy = PG_GETARG_UINT16(1);
1017 :
1018 : /* Jsonb *query = PG_GETARG_JSONB_P(2); */
1019 63642 : int32 nkeys = PG_GETARG_INT32(3);
1020 63642 : Pointer *extra_data = (Pointer *) PG_GETARG_POINTER(4);
1021 63642 : GinTernaryValue res = GIN_MAYBE;
1022 : int32 i;
1023 :
1024 : /*
1025 : * Note that we never return GIN_TRUE, only GIN_MAYBE or GIN_FALSE; this
1026 : * corresponds to always forcing recheck in the regular consistent
1027 : * function, for the reasons listed there.
1028 : */
1029 63642 : if (strategy == JsonbContainsStrategyNumber ||
1030 : strategy == JsonbExistsAllStrategyNumber)
1031 : {
1032 : /* All extracted keys must be present */
1033 10140 : for (i = 0; i < nkeys; i++)
1034 : {
1035 3534 : if (check[i] == GIN_FALSE)
1036 : {
1037 2052 : res = GIN_FALSE;
1038 2052 : break;
1039 : }
1040 : }
1041 : }
1042 54984 : else if (strategy == JsonbExistsStrategyNumber ||
1043 : strategy == JsonbExistsAnyStrategyNumber)
1044 : {
1045 : /* At least one extracted key must be present */
1046 3240 : res = GIN_FALSE;
1047 4098 : for (i = 0; i < nkeys; i++)
1048 : {
1049 4098 : if (check[i] == GIN_TRUE ||
1050 864 : check[i] == GIN_MAYBE)
1051 : {
1052 3240 : res = GIN_MAYBE;
1053 3240 : break;
1054 : }
1055 : }
1056 : }
1057 51744 : else if (strategy == JsonbJsonpathPredicateStrategyNumber ||
1058 : strategy == JsonbJsonpathExistsStrategyNumber)
1059 : {
1060 51744 : if (nkeys > 0)
1061 : {
1062 : Assert(extra_data && extra_data[0]);
1063 3168 : res = execute_jsp_gin_node((JsonPathGinNode *) extra_data[0], check,
1064 : true);
1065 :
1066 : /* Should always recheck the result */
1067 3168 : if (res == GIN_TRUE)
1068 636 : res = GIN_MAYBE;
1069 : }
1070 : }
1071 : else
1072 0 : elog(ERROR, "unrecognized strategy number: %d", strategy);
1073 :
1074 63642 : PG_RETURN_GIN_TERNARY_VALUE(res);
1075 : }
1076 :
1077 : /*
1078 : *
1079 : * jsonb_path_ops GIN opclass support functions
1080 : *
1081 : * In a jsonb_path_ops index, the GIN keys are uint32 hashes, one per JSON
1082 : * value; but the JSON key(s) leading to each value are also included in its
1083 : * hash computation. This means we can only support containment queries,
1084 : * but the index can distinguish, for example, {"foo": 42} from {"bar": 42}
1085 : * since different hashes will be generated.
1086 : *
1087 : */
1088 :
1089 : Datum
1090 6216 : gin_extract_jsonb_path(PG_FUNCTION_ARGS)
1091 : {
1092 6216 : Jsonb *jb = PG_GETARG_JSONB_P(0);
1093 6216 : int32 *nentries = (int32 *) PG_GETARG_POINTER(1);
1094 6216 : int total = JB_ROOT_COUNT(jb);
1095 : JsonbIterator *it;
1096 : JsonbValue v;
1097 : JsonbIteratorToken r;
1098 : PathHashStack tail;
1099 : PathHashStack *stack;
1100 : GinEntries entries;
1101 :
1102 : /* If the root level is empty, we certainly have no keys */
1103 6216 : if (total == 0)
1104 : {
1105 720 : *nentries = 0;
1106 720 : PG_RETURN_POINTER(NULL);
1107 : }
1108 :
1109 : /* Otherwise, use 2 * root count as initial estimate of result size */
1110 5496 : init_gin_entries(&entries, 2 * total);
1111 :
1112 : /* We keep a stack of partial hashes corresponding to parent key levels */
1113 5496 : tail.parent = NULL;
1114 5496 : tail.hash = 0;
1115 5496 : stack = &tail;
1116 :
1117 5496 : it = JsonbIteratorInit(&jb->root);
1118 :
1119 74706 : while ((r = JsonbIteratorNext(&it, &v, false)) != WJB_DONE)
1120 : {
1121 : PathHashStack *parent;
1122 :
1123 69210 : switch (r)
1124 : {
1125 5658 : case WJB_BEGIN_ARRAY:
1126 : case WJB_BEGIN_OBJECT:
1127 : /* Push a stack level for this object */
1128 5658 : parent = stack;
1129 5658 : stack = (PathHashStack *) palloc(sizeof(PathHashStack));
1130 :
1131 : /*
1132 : * We pass forward hashes from outer nesting levels so that
1133 : * the hashes for nested values will include outer keys as
1134 : * well as their own keys.
1135 : *
1136 : * Nesting an array within another array will not alter
1137 : * innermost scalar element hash values, but that seems
1138 : * inconsequential.
1139 : */
1140 5658 : stack->hash = parent->hash;
1141 5658 : stack->parent = parent;
1142 5658 : break;
1143 28914 : case WJB_KEY:
1144 : /* mix this key into the current outer hash */
1145 28914 : JsonbHashScalarValue(&v, &stack->hash);
1146 : /* hash is now ready to incorporate the value */
1147 28914 : break;
1148 28980 : case WJB_ELEM:
1149 : case WJB_VALUE:
1150 : /* mix the element or value's hash into the prepared hash */
1151 28980 : JsonbHashScalarValue(&v, &stack->hash);
1152 : /* and emit an index entry */
1153 28980 : add_gin_entry(&entries, UInt32GetDatum(stack->hash));
1154 : /* reset hash for next key, value, or sub-object */
1155 28980 : stack->hash = stack->parent->hash;
1156 28980 : break;
1157 5658 : case WJB_END_ARRAY:
1158 : case WJB_END_OBJECT:
1159 : /* Pop the stack */
1160 5658 : parent = stack->parent;
1161 5658 : pfree(stack);
1162 5658 : stack = parent;
1163 : /* reset hash for next key, value, or sub-object */
1164 5658 : if (stack->parent)
1165 162 : stack->hash = stack->parent->hash;
1166 : else
1167 5496 : stack->hash = 0;
1168 5658 : break;
1169 0 : default:
1170 0 : elog(ERROR, "invalid JsonbIteratorNext rc: %d", (int) r);
1171 : }
1172 : }
1173 :
1174 5496 : *nentries = entries.count;
1175 :
1176 5496 : PG_RETURN_POINTER(entries.buf);
1177 : }
1178 :
1179 : Datum
1180 420 : gin_extract_jsonb_query_path(PG_FUNCTION_ARGS)
1181 : {
1182 420 : int32 *nentries = (int32 *) PG_GETARG_POINTER(1);
1183 420 : StrategyNumber strategy = PG_GETARG_UINT16(2);
1184 420 : int32 *searchMode = (int32 *) PG_GETARG_POINTER(6);
1185 : Datum *entries;
1186 :
1187 420 : if (strategy == JsonbContainsStrategyNumber)
1188 : {
1189 : /* Query is a jsonb, so just apply gin_extract_jsonb_path ... */
1190 : entries = (Datum *)
1191 126 : DatumGetPointer(DirectFunctionCall2(gin_extract_jsonb_path,
1192 : PG_GETARG_DATUM(0),
1193 : PointerGetDatum(nentries)));
1194 :
1195 : /* ... although "contains {}" requires a full index scan */
1196 126 : if (*nentries == 0)
1197 12 : *searchMode = GIN_SEARCH_MODE_ALL;
1198 : }
1199 294 : else if (strategy == JsonbJsonpathPredicateStrategyNumber ||
1200 : strategy == JsonbJsonpathExistsStrategyNumber)
1201 294 : {
1202 294 : JsonPath *jp = PG_GETARG_JSONPATH_P(0);
1203 294 : Pointer **extra_data = (Pointer **) PG_GETARG_POINTER(4);
1204 :
1205 294 : entries = extract_jsp_query(jp, strategy, true, nentries, extra_data);
1206 :
1207 294 : if (!entries)
1208 48 : *searchMode = GIN_SEARCH_MODE_ALL;
1209 : }
1210 : else
1211 : {
1212 0 : elog(ERROR, "unrecognized strategy number: %d", strategy);
1213 : entries = NULL;
1214 : }
1215 :
1216 420 : PG_RETURN_POINTER(entries);
1217 : }
1218 :
1219 : Datum
1220 0 : gin_consistent_jsonb_path(PG_FUNCTION_ARGS)
1221 : {
1222 0 : bool *check = (bool *) PG_GETARG_POINTER(0);
1223 0 : StrategyNumber strategy = PG_GETARG_UINT16(1);
1224 :
1225 : /* Jsonb *query = PG_GETARG_JSONB_P(2); */
1226 0 : int32 nkeys = PG_GETARG_INT32(3);
1227 0 : Pointer *extra_data = (Pointer *) PG_GETARG_POINTER(4);
1228 0 : bool *recheck = (bool *) PG_GETARG_POINTER(5);
1229 0 : bool res = true;
1230 : int32 i;
1231 :
1232 0 : if (strategy == JsonbContainsStrategyNumber)
1233 : {
1234 : /*
1235 : * jsonb_path_ops is necessarily lossy, not only because of hash
1236 : * collisions but also because it doesn't preserve complete
1237 : * information about the structure of the JSON object. Besides, there
1238 : * are some special rules around the containment of raw scalars in
1239 : * arrays that are not handled here. So we must always recheck a
1240 : * match. However, if not all of the keys are present, the tuple
1241 : * certainly doesn't match.
1242 : */
1243 0 : *recheck = true;
1244 0 : for (i = 0; i < nkeys; i++)
1245 : {
1246 0 : if (!check[i])
1247 : {
1248 0 : res = false;
1249 0 : break;
1250 : }
1251 : }
1252 : }
1253 0 : else if (strategy == JsonbJsonpathPredicateStrategyNumber ||
1254 : strategy == JsonbJsonpathExistsStrategyNumber)
1255 : {
1256 0 : *recheck = true;
1257 :
1258 0 : if (nkeys > 0)
1259 : {
1260 : Assert(extra_data && extra_data[0]);
1261 0 : res = execute_jsp_gin_node((JsonPathGinNode *) extra_data[0], check,
1262 : false) != GIN_FALSE;
1263 : }
1264 : }
1265 : else
1266 0 : elog(ERROR, "unrecognized strategy number: %d", strategy);
1267 :
1268 0 : PG_RETURN_BOOL(res);
1269 : }
1270 :
1271 : Datum
1272 31188 : gin_triconsistent_jsonb_path(PG_FUNCTION_ARGS)
1273 : {
1274 31188 : GinTernaryValue *check = (GinTernaryValue *) PG_GETARG_POINTER(0);
1275 31188 : StrategyNumber strategy = PG_GETARG_UINT16(1);
1276 :
1277 : /* Jsonb *query = PG_GETARG_JSONB_P(2); */
1278 31188 : int32 nkeys = PG_GETARG_INT32(3);
1279 31188 : Pointer *extra_data = (Pointer *) PG_GETARG_POINTER(4);
1280 31188 : GinTernaryValue res = GIN_MAYBE;
1281 : int32 i;
1282 :
1283 31188 : if (strategy == JsonbContainsStrategyNumber)
1284 : {
1285 : /*
1286 : * Note that we never return GIN_TRUE, only GIN_MAYBE or GIN_FALSE;
1287 : * this corresponds to always forcing recheck in the regular
1288 : * consistent function, for the reasons listed there.
1289 : */
1290 6558 : for (i = 0; i < nkeys; i++)
1291 : {
1292 330 : if (check[i] == GIN_FALSE)
1293 : {
1294 84 : res = GIN_FALSE;
1295 84 : break;
1296 : }
1297 : }
1298 : }
1299 24876 : else if (strategy == JsonbJsonpathPredicateStrategyNumber ||
1300 : strategy == JsonbJsonpathExistsStrategyNumber)
1301 : {
1302 24876 : if (nkeys > 0)
1303 : {
1304 : Assert(extra_data && extra_data[0]);
1305 588 : res = execute_jsp_gin_node((JsonPathGinNode *) extra_data[0], check,
1306 : true);
1307 :
1308 : /* Should always recheck the result */
1309 588 : if (res == GIN_TRUE)
1310 420 : res = GIN_MAYBE;
1311 : }
1312 : }
1313 : else
1314 0 : elog(ERROR, "unrecognized strategy number: %d", strategy);
1315 :
1316 31188 : PG_RETURN_GIN_TERNARY_VALUE(res);
1317 : }
1318 :
1319 : /*
1320 : * Construct a jsonb_ops GIN key from a flag byte and a textual representation
1321 : * (which need not be null-terminated). This function is responsible
1322 : * for hashing overlength text representations; it will add the
1323 : * JGINFLAG_HASHED bit to the flag value if it does that.
1324 : */
1325 : static Datum
1326 64704 : make_text_key(char flag, const char *str, int len)
1327 : {
1328 : text *item;
1329 : char hashbuf[10];
1330 :
1331 64704 : if (len > JGIN_MAXLENGTH)
1332 : {
1333 : uint32 hashval;
1334 :
1335 0 : hashval = DatumGetUInt32(hash_any((const unsigned char *) str, len));
1336 0 : snprintf(hashbuf, sizeof(hashbuf), "%08x", hashval);
1337 0 : str = hashbuf;
1338 0 : len = 8;
1339 0 : flag |= JGINFLAG_HASHED;
1340 : }
1341 :
1342 : /*
1343 : * Now build the text Datum. For simplicity we build a 4-byte-header
1344 : * varlena text Datum here, but we expect it will get converted to short
1345 : * header format when stored in the index.
1346 : */
1347 64704 : item = (text *) palloc(VARHDRSZ + len + 1);
1348 64704 : SET_VARSIZE(item, VARHDRSZ + len + 1);
1349 :
1350 64704 : *VARDATA(item) = flag;
1351 :
1352 64704 : memcpy(VARDATA(item) + 1, str, len);
1353 :
1354 64704 : return PointerGetDatum(item);
1355 : }
1356 :
1357 : /*
1358 : * Create a textual representation of a JsonbValue that will serve as a GIN
1359 : * key in a jsonb_ops index. is_key is true if the JsonbValue is a key,
1360 : * or if it is a string array element (since we pretend those are keys,
1361 : * see jsonb.h).
1362 : */
1363 : static Datum
1364 64236 : make_scalar_key(const JsonbValue *scalarVal, bool is_key)
1365 : {
1366 : Datum item;
1367 : char *cstr;
1368 :
1369 64236 : switch (scalarVal->type)
1370 : {
1371 78 : case jbvNull:
1372 : Assert(!is_key);
1373 78 : item = make_text_key(JGINFLAG_NULL, "", 0);
1374 78 : break;
1375 5568 : case jbvBool:
1376 : Assert(!is_key);
1377 5568 : item = make_text_key(JGINFLAG_BOOL,
1378 5568 : scalarVal->val.boolean ? "t" : "f", 1);
1379 5568 : break;
1380 12966 : case jbvNumeric:
1381 : Assert(!is_key);
1382 :
1383 : /*
1384 : * A normalized textual representation, free of trailing zeroes,
1385 : * is required so that numerically equal values will produce equal
1386 : * strings.
1387 : *
1388 : * It isn't ideal that numerics are stored in a relatively bulky
1389 : * textual format. However, it's a notationally convenient way of
1390 : * storing a "union" type in the GIN B-Tree, and indexing Jsonb
1391 : * strings takes precedence.
1392 : */
1393 12966 : cstr = numeric_normalize(scalarVal->val.numeric);
1394 12966 : item = make_text_key(JGINFLAG_NUM, cstr, strlen(cstr));
1395 12966 : pfree(cstr);
1396 12966 : break;
1397 45624 : case jbvString:
1398 45624 : item = make_text_key(is_key ? JGINFLAG_KEY : JGINFLAG_STR,
1399 45624 : scalarVal->val.string.val,
1400 : scalarVal->val.string.len);
1401 45624 : break;
1402 0 : default:
1403 0 : elog(ERROR, "unrecognized jsonb scalar type: %d", scalarVal->type);
1404 : item = 0; /* keep compiler quiet */
1405 : break;
1406 : }
1407 :
1408 64236 : return item;
1409 : }
|
