PostgreSQL Source Code  git master
jsonfuncs.c
Go to the documentation of this file.
1 /*-------------------------------------------------------------------------
2  *
3  * jsonfuncs.c
4  * Functions to process JSON data types.
5  *
6  * Portions Copyright (c) 1996-2019, PostgreSQL Global Development Group
7  * Portions Copyright (c) 1994, Regents of the University of California
8  *
9  * IDENTIFICATION
10  * src/backend/utils/adt/jsonfuncs.c
11  *
12  *-------------------------------------------------------------------------
13  */
14 
15 #include "postgres.h"
16 
17 #include <limits.h>
18 
19 #include "access/htup_details.h"
20 #include "catalog/pg_type.h"
21 #include "fmgr.h"
22 #include "funcapi.h"
23 #include "lib/stringinfo.h"
24 #include "mb/pg_wchar.h"
25 #include "miscadmin.h"
26 #include "utils/array.h"
27 #include "utils/builtins.h"
28 #include "utils/hsearch.h"
29 #include "utils/json.h"
30 #include "utils/jsonapi.h"
31 #include "utils/jsonb.h"
32 #include "utils/lsyscache.h"
33 #include "utils/memutils.h"
34 #include "utils/syscache.h"
35 #include "utils/typcache.h"
36 
37 /* Operations available for setPath */
38 #define JB_PATH_CREATE 0x0001
39 #define JB_PATH_DELETE 0x0002
40 #define JB_PATH_REPLACE 0x0004
41 #define JB_PATH_INSERT_BEFORE 0x0008
42 #define JB_PATH_INSERT_AFTER 0x0010
43 #define JB_PATH_CREATE_OR_INSERT \
44  (JB_PATH_INSERT_BEFORE | JB_PATH_INSERT_AFTER | JB_PATH_CREATE)
45 
46 /* state for json_object_keys */
47 typedef struct OkeysState
48 {
49  JsonLexContext *lex;
50  char **result;
51  int result_size;
52  int result_count;
53  int sent_count;
54 } OkeysState;
55 
56 /* state for iterate_json_values function */
57 typedef struct IterateJsonStringValuesState
58 {
59  JsonLexContext *lex;
60  JsonIterateStringValuesAction action; /* an action that will be applied
61  * to each json value */
62  void *action_state; /* any necessary context for iteration */
63  uint32 flags; /* what kind of elements from a json we want
64  * to iterate */
65 } IterateJsonStringValuesState;
66 
67 /* state for transform_json_string_values function */
68 typedef struct TransformJsonStringValuesState
69 {
70  JsonLexContext *lex;
71  StringInfo strval; /* resulting json */
72  JsonTransformStringValuesAction action; /* an action that will be applied
73  * to each json value */
74  void *action_state; /* any necessary context for transformation */
75 } TransformJsonStringValuesState;
76 
77 /* state for json_get* functions */
78 typedef struct GetState
79 {
80  JsonLexContext *lex;
81  text *tresult;
82  char *result_start;
83  bool normalize_results;
84  bool next_scalar;
85  int npath; /* length of each path-related array */
86  char **path_names; /* field name(s) being sought */
87  int *path_indexes; /* array index(es) being sought */
88  bool *pathok; /* is path matched to current depth? */
89  int *array_cur_index; /* current element index at each path
90  * level */
91 } GetState;
92 
93 /* state for json_array_length */
94 typedef struct AlenState
95 {
96  JsonLexContext *lex;
97  int count;
98 } AlenState;
99 
100 /* state for json_each */
101 typedef struct EachState
102 {
103  JsonLexContext *lex;
104  Tuplestorestate *tuple_store;
105  TupleDesc ret_tdesc;
106  MemoryContext tmp_cxt;
107  char *result_start;
108  bool normalize_results;
109  bool next_scalar;
110  char *normalized_scalar;
111 } EachState;
112 
113 /* state for json_array_elements */
114 typedef struct ElementsState
115 {
116  JsonLexContext *lex;
117  const char *function_name;
118  Tuplestorestate *tuple_store;
119  TupleDesc ret_tdesc;
120  MemoryContext tmp_cxt;
121  char *result_start;
122  bool normalize_results;
123  bool next_scalar;
124  char *normalized_scalar;
125 } ElementsState;
126 
127 /* state for get_json_object_as_hash */
128 typedef struct JHashState
129 {
130  JsonLexContext *lex;
131  const char *function_name;
132  HTAB *hash;
133  char *saved_scalar;
134  char *save_json_start;
135  JsonTokenType saved_token_type;
136 } JHashState;
137 
138 /* hashtable element */
139 typedef struct JsonHashEntry
140 {
141  char fname[NAMEDATALEN]; /* hash key (MUST BE FIRST) */
142  char *val;
143  JsonTokenType type;
144 } JsonHashEntry;
145 
146 /* structure to cache type I/O metadata needed for populate_scalar() */
147 typedef struct ScalarIOData
148 {
149  Oid typioparam;
150  FmgrInfo typiofunc;
151 } ScalarIOData;
152 
153 /* these two structures are used recursively */
154 typedef struct ColumnIOData ColumnIOData;
155 typedef struct RecordIOData RecordIOData;
156 
157 /* structure to cache metadata needed for populate_array() */
158 typedef struct ArrayIOData
159 {
160  ColumnIOData *element_info; /* metadata cache */
161  Oid element_type; /* array element type id */
162  int32 element_typmod; /* array element type modifier */
163 } ArrayIOData;
164 
165 /* structure to cache metadata needed for populate_composite() */
166 typedef struct CompositeIOData
167 {
168  /*
169  * We use pointer to a RecordIOData here because variable-length struct
170  * RecordIOData can't be used directly in ColumnIOData.io union
171  */
172  RecordIOData *record_io; /* metadata cache for populate_record() */
173  TupleDesc tupdesc; /* cached tuple descriptor */
174  /* these fields differ from target type only if domain over composite: */
175  Oid base_typid; /* base type id */
176  int32 base_typmod; /* base type modifier */
177  /* this field is used only if target type is domain over composite: */
178  void *domain_info; /* opaque cache for domain checks */
179 } CompositeIOData;
180 
181 /* structure to cache metadata needed for populate_domain() */
182 typedef struct DomainIOData
183 {
184  ColumnIOData *base_io; /* metadata cache */
185  Oid base_typid; /* base type id */
186  int32 base_typmod; /* base type modifier */
187  void *domain_info; /* opaque cache for domain checks */
188 } DomainIOData;
189 
190 /* enumeration type categories */
191 typedef enum TypeCat
192 {
193  TYPECAT_SCALAR = 's',
194  TYPECAT_ARRAY = 'a',
195  TYPECAT_COMPOSITE = 'c',
196  TYPECAT_COMPOSITE_DOMAIN = 'C',
197  TYPECAT_DOMAIN = 'd'
198 } TypeCat;
199 
200 /* these two are stolen from hstore / record_out, used in populate_record* */
201 
202 /* structure to cache record metadata needed for populate_record_field() */
203 struct ColumnIOData
204 {
205  Oid typid; /* column type id */
206  int32 typmod; /* column type modifier */
207  TypeCat typcat; /* column type category */
208  ScalarIOData scalar_io; /* metadata cache for directi conversion
209  * through input function */
210  union
211  {
212  ArrayIOData array;
213  CompositeIOData composite;
214  DomainIOData domain;
215  } io; /* metadata cache for various column type
216  * categories */
217 };
218 
219 /* structure to cache record metadata needed for populate_record() */
220 struct RecordIOData
221 {
222  Oid record_type;
223  int32 record_typmod;
224  int ncolumns;
225  ColumnIOData columns[FLEXIBLE_ARRAY_MEMBER];
226 };
227 
228 /* per-query cache for populate_recordset */
229 typedef struct PopulateRecordsetCache
230 {
231  Oid argtype; /* declared type of the record argument */
232  ColumnIOData c; /* metadata cache for populate_composite() */
233  MemoryContext fn_mcxt; /* where this is stored */
234 } PopulateRecordsetCache;
235 
236 /* per-call state for populate_recordset */
237 typedef struct PopulateRecordsetState
238 {
239  JsonLexContext *lex;
240  const char *function_name;
241  HTAB *json_hash;
242  char *saved_scalar;
243  char *save_json_start;
244  JsonTokenType saved_token_type;
245  Tuplestorestate *tuple_store;
246  HeapTupleHeader rec;
247  PopulateRecordsetCache *cache;
248 } PopulateRecordsetState;
249 
250 /* structure to cache metadata needed for populate_record_worker() */
251 typedef struct PopulateRecordCache
252 {
253  Oid argtype; /* declared type of the record argument */
254  ColumnIOData c; /* metadata cache for populate_composite() */
255 } PopulateRecordCache;
256 
257 /* common data for populate_array_json() and populate_array_dim_jsonb() */
258 typedef struct PopulateArrayContext
259 {
260  ArrayBuildState *astate; /* array build state */
261  ArrayIOData *aio; /* metadata cache */
262  MemoryContext acxt; /* array build memory context */
263  MemoryContext mcxt; /* cache memory context */
264  const char *colname; /* for diagnostics only */
265  int *dims; /* dimensions */
266  int *sizes; /* current dimension counters */
267  int ndims; /* number of dimensions */
268 } PopulateArrayContext;
269 
270 /* state for populate_array_json() */
271 typedef struct PopulateArrayState
272 {
273  JsonLexContext *lex; /* json lexer */
274  PopulateArrayContext *ctx; /* context */
275  char *element_start; /* start of the current array element */
276  char *element_scalar; /* current array element token if it is a
277  * scalar */
278  JsonTokenType element_type; /* current array element type */
279 } PopulateArrayState;
280 
281 /* state for json_strip_nulls */
282 typedef struct StripnullState
283 {
284  JsonLexContext *lex;
285  StringInfo strval;
286  bool skip_next_null;
287 } StripnullState;
288 
289 /* structure for generalized json/jsonb value passing */
290 typedef struct JsValue
291 {
292  bool is_json; /* json/jsonb */
293  union
294  {
295  struct
296  {
297  char *str; /* json string */
298  int len; /* json string length or -1 if null-terminated */
299  JsonTokenType type; /* json type */
300  } json; /* json value */
301 
302  JsonbValue *jsonb; /* jsonb value */
303  } val;
304 } JsValue;
305 
306 typedef struct JsObject
307 {
308  bool is_json; /* json/jsonb */
309  union
310  {
311  HTAB *json_hash;
312  JsonbContainer *jsonb_cont;
313  } val;
314 } JsObject;
315 
316 /* useful macros for testing JsValue properties */
317 #define JsValueIsNull(jsv) \
318  ((jsv)->is_json ? \
319  (!(jsv)->val.json.str || (jsv)->val.json.type == JSON_TOKEN_NULL) : \
320  (!(jsv)->val.jsonb || (jsv)->val.jsonb->type == jbvNull))
321 
322 #define JsValueIsString(jsv) \
323  ((jsv)->is_json ? (jsv)->val.json.type == JSON_TOKEN_STRING \
324  : ((jsv)->val.jsonb && (jsv)->val.jsonb->type == jbvString))
325 
326 #define JsObjectIsEmpty(jso) \
327  ((jso)->is_json \
328  ? hash_get_num_entries((jso)->val.json_hash) == 0 \
329  : ((jso)->val.jsonb_cont == NULL || \
330  JsonContainerSize((jso)->val.jsonb_cont) == 0))
331 
332 #define JsObjectFree(jso) \
333  do { \
334  if ((jso)->is_json) \
335  hash_destroy((jso)->val.json_hash); \
336  } while (0)
337 
338 /* semantic action functions for json_object_keys */
339 static void okeys_object_field_start(void *state, char *fname, bool isnull);
340 static void okeys_array_start(void *state);
341 static void okeys_scalar(void *state, char *token, JsonTokenType tokentype);
342 
343 /* semantic action functions for json_get* functions */
344 static void get_object_start(void *state);
345 static void get_object_end(void *state);
346 static void get_object_field_start(void *state, char *fname, bool isnull);
347 static void get_object_field_end(void *state, char *fname, bool isnull);
348 static void get_array_start(void *state);
349 static void get_array_end(void *state);
350 static void get_array_element_start(void *state, bool isnull);
351 static void get_array_element_end(void *state, bool isnull);
352 static void get_scalar(void *state, char *token, JsonTokenType tokentype);
353 
354 /* common worker function for json getter functions */
355 static Datum get_path_all(FunctionCallInfo fcinfo, bool as_text);
356 static text *get_worker(text *json, char **tpath, int *ipath, int npath,
357  bool normalize_results);
358 static Datum get_jsonb_path_all(FunctionCallInfo fcinfo, bool as_text);
359 
360 /* semantic action functions for json_array_length */
361 static void alen_object_start(void *state);
362 static void alen_scalar(void *state, char *token, JsonTokenType tokentype);
363 static void alen_array_element_start(void *state, bool isnull);
364 
365 /* common workers for json{b}_each* functions */
366 static Datum each_worker(FunctionCallInfo fcinfo, bool as_text);
367 static Datum each_worker_jsonb(FunctionCallInfo fcinfo, const char *funcname,
368  bool as_text);
369 
370 /* semantic action functions for json_each */
371 static void each_object_field_start(void *state, char *fname, bool isnull);
372 static void each_object_field_end(void *state, char *fname, bool isnull);
373 static void each_array_start(void *state);
374 static void each_scalar(void *state, char *token, JsonTokenType tokentype);
375 
376 /* common workers for json{b}_array_elements_* functions */
377 static Datum elements_worker(FunctionCallInfo fcinfo, const char *funcname,
378  bool as_text);
379 static Datum elements_worker_jsonb(FunctionCallInfo fcinfo, const char *funcname,
380  bool as_text);
381 
382 /* semantic action functions for json_array_elements */
383 static void elements_object_start(void *state);
384 static void elements_array_element_start(void *state, bool isnull);
385 static void elements_array_element_end(void *state, bool isnull);
386 static void elements_scalar(void *state, char *token, JsonTokenType tokentype);
387 
388 /* turn a json object into a hash table */
389 static HTAB *get_json_object_as_hash(char *json, int len, const char *funcname);
390 
391 /* semantic actions for populate_array_json */
392 static void populate_array_object_start(void *_state);
393 static void populate_array_array_end(void *_state);
394 static void populate_array_element_start(void *_state, bool isnull);
395 static void populate_array_element_end(void *_state, bool isnull);
396 static void populate_array_scalar(void *_state, char *token, JsonTokenType tokentype);
397 
398 /* semantic action functions for get_json_object_as_hash */
399 static void hash_object_field_start(void *state, char *fname, bool isnull);
400 static void hash_object_field_end(void *state, char *fname, bool isnull);
401 static void hash_array_start(void *state);
402 static void hash_scalar(void *state, char *token, JsonTokenType tokentype);
403 
404 /* semantic action functions for populate_recordset */
405 static void populate_recordset_object_field_start(void *state, char *fname, bool isnull);
406 static void populate_recordset_object_field_end(void *state, char *fname, bool isnull);
407 static void populate_recordset_scalar(void *state, char *token, JsonTokenType tokentype);
408 static void populate_recordset_object_start(void *state);
409 static void populate_recordset_object_end(void *state);
410 static void populate_recordset_array_start(void *state);
411 static void populate_recordset_array_element_start(void *state, bool isnull);
412 
413 /* semantic action functions for json_strip_nulls */
414 static void sn_object_start(void *state);
415 static void sn_object_end(void *state);
416 static void sn_array_start(void *state);
417 static void sn_array_end(void *state);
418 static void sn_object_field_start(void *state, char *fname, bool isnull);
419 static void sn_array_element_start(void *state, bool isnull);
420 static void sn_scalar(void *state, char *token, JsonTokenType tokentype);
421 
422 /* worker functions for populate_record, to_record, populate_recordset and to_recordset */
423 static Datum populate_recordset_worker(FunctionCallInfo fcinfo, const char *funcname,
424  bool is_json, bool have_record_arg);
425 static Datum populate_record_worker(FunctionCallInfo fcinfo, const char *funcname,
426  bool is_json, bool have_record_arg);
427 
428 /* helper functions for populate_record[set] */
429 static HeapTupleHeader populate_record(TupleDesc tupdesc, RecordIOData **record_p,
430  HeapTupleHeader defaultval, MemoryContext mcxt,
431  JsObject *obj);
432 static void JsValueToJsObject(JsValue *jsv, JsObject *jso);
433 static Datum populate_composite(CompositeIOData *io, Oid typid,
434  const char *colname, MemoryContext mcxt,
435  HeapTupleHeader defaultval, JsValue *jsv, bool isnull);
436 static Datum populate_scalar(ScalarIOData *io, Oid typid, int32 typmod, JsValue *jsv);
437 static void prepare_column_cache(ColumnIOData *column, Oid typid, int32 typmod,
438  MemoryContext mcxt, bool need_scalar);
439 static Datum populate_record_field(ColumnIOData *col, Oid typid, int32 typmod,
440  const char *colname, MemoryContext mcxt, Datum defaultval,
441  JsValue *jsv, bool *isnull);
442 static RecordIOData *allocate_record_info(MemoryContext mcxt, int ncolumns);
443 static bool JsObjectGetField(JsObject *obj, char *field, JsValue *jsv);
444 static void populate_recordset_record(PopulateRecordsetState *state, JsObject *obj);
445 static void populate_array_json(PopulateArrayContext *ctx, char *json, int len);
446 static void populate_array_dim_jsonb(PopulateArrayContext *ctx, JsonbValue *jbv,
447  int ndim);
448 static void populate_array_report_expected_array(PopulateArrayContext *ctx, int ndim);
449 static void populate_array_assign_ndims(PopulateArrayContext *ctx, int ndims);
450 static void populate_array_check_dimension(PopulateArrayContext *ctx, int ndim);
451 static void populate_array_element(PopulateArrayContext *ctx, int ndim, JsValue *jsv);
452 static Datum populate_array(ArrayIOData *aio, const char *colname,
453  MemoryContext mcxt, JsValue *jsv);
454 static Datum populate_domain(DomainIOData *io, Oid typid, const char *colname,
455  MemoryContext mcxt, JsValue *jsv, bool isnull);
456 
457 /* Worker that takes care of common setup for us */
458 static JsonbValue *findJsonbValueFromContainerLen(JsonbContainer *container,
459  uint32 flags,
460  char *key,
461  uint32 keylen);
462 
463 /* functions supporting jsonb_delete, jsonb_set and jsonb_concat */
464 static JsonbValue *IteratorConcat(JsonbIterator **it1, JsonbIterator **it2,
465  JsonbParseState **state);
466 static JsonbValue *setPath(JsonbIterator **it, Datum *path_elems,
467  bool *path_nulls, int path_len,
468  JsonbParseState **st, int level, Jsonb *newval,
469  int op_type);
470 static void setPathObject(JsonbIterator **it, Datum *path_elems,
471  bool *path_nulls, int path_len, JsonbParseState **st,
472  int level,
473  Jsonb *newval, uint32 npairs, int op_type);
474 static void setPathArray(JsonbIterator **it, Datum *path_elems,
475  bool *path_nulls, int path_len, JsonbParseState **st,
476  int level, Jsonb *newval, uint32 nelems, int op_type);
477 static void addJsonbToParseState(JsonbParseState **jbps, Jsonb *jb);
478 
479 /* function supporting iterate_json_values */
480 static void iterate_values_scalar(void *state, char *token, JsonTokenType tokentype);
481 static void iterate_values_object_field_start(void *state, char *fname, bool isnull);
482 
483 /* functions supporting transform_json_string_values */
484 static void transform_string_values_object_start(void *state);
485 static void transform_string_values_object_end(void *state);
486 static void transform_string_values_array_start(void *state);
487 static void transform_string_values_array_end(void *state);
488 static void transform_string_values_object_field_start(void *state, char *fname, bool isnull);
489 static void transform_string_values_array_element_start(void *state, bool isnull);
490 static void transform_string_values_scalar(void *state, char *token, JsonTokenType tokentype);
491 
492 /*
493  * SQL function json_object_keys
494  *
495  * Returns the set of keys for the object argument.
496  *
497  * This SRF operates in value-per-call mode. It processes the
498  * object during the first call, and the keys are simply stashed
499  * in an array, whose size is expanded as necessary. This is probably
500  * safe enough for a list of keys of a single object, since they are
501  * limited in size to NAMEDATALEN and the number of keys is unlikely to
502  * be so huge that it has major memory implications.
503  */
504 Datum
505 jsonb_object_keys(PG_FUNCTION_ARGS)
506 {
507  FuncCallContext *funcctx;
508  OkeysState *state;
509  int i;
510 
511  if (SRF_IS_FIRSTCALL())
512  {
513  MemoryContext oldcontext;
514  Jsonb *jb = PG_GETARG_JSONB_P(0);
515  bool skipNested = false;
516  JsonbIterator *it;
517  JsonbValue v;
518  JsonbIteratorToken r;
519 
520  if (JB_ROOT_IS_SCALAR(jb))
521  ereport(ERROR,
522  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
523  errmsg("cannot call %s on a scalar",
524  "jsonb_object_keys")));
525  else if (JB_ROOT_IS_ARRAY(jb))
526  ereport(ERROR,
527  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
528  errmsg("cannot call %s on an array",
529  "jsonb_object_keys")));
530 
531  funcctx = SRF_FIRSTCALL_INIT();
532  oldcontext = MemoryContextSwitchTo(funcctx->multi_call_memory_ctx);
533 
534  state = palloc(sizeof(OkeysState));
535 
536  state->result_size = JB_ROOT_COUNT(jb);
537  state->result_count = 0;
538  state->sent_count = 0;
539  state->result = palloc(state->result_size * sizeof(char *));
540 
541  it = JsonbIteratorInit(&jb->root);
542 
543  while ((r = JsonbIteratorNext(&it, &v, skipNested)) != WJB_DONE)
544  {
545  skipNested = true;
546 
547  if (r == WJB_KEY)
548  {
549  char *cstr;
550 
551  cstr = palloc(v.val.string.len + 1 * sizeof(char));
552  memcpy(cstr, v.val.string.val, v.val.string.len);
553  cstr[v.val.string.len] = '\0';
554  state->result[state->result_count++] = cstr;
555  }
556  }
557 
558  MemoryContextSwitchTo(oldcontext);
559  funcctx->user_fctx = (void *) state;
560  }
561 
562  funcctx = SRF_PERCALL_SETUP();
563  state = (OkeysState *) funcctx->user_fctx;
564 
565  if (state->sent_count < state->result_count)
566  {
567  char *nxt = state->result[state->sent_count++];
568 
569  SRF_RETURN_NEXT(funcctx, CStringGetTextDatum(nxt));
570  }
571 
572  /* cleanup to reduce or eliminate memory leaks */
573  for (i = 0; i < state->result_count; i++)
574  pfree(state->result[i]);
575  pfree(state->result);
576  pfree(state);
577 
578  SRF_RETURN_DONE(funcctx);
579 }
580 
581 
582 Datum
583 json_object_keys(PG_FUNCTION_ARGS)
584 {
585  FuncCallContext *funcctx;
586  OkeysState *state;
587  int i;
588 
589  if (SRF_IS_FIRSTCALL())
590  {
591  text *json = PG_GETARG_TEXT_PP(0);
592  JsonLexContext *lex = makeJsonLexContext(json, true);
593  JsonSemAction *sem;
594  MemoryContext oldcontext;
595 
596  funcctx = SRF_FIRSTCALL_INIT();
597  oldcontext = MemoryContextSwitchTo(funcctx->multi_call_memory_ctx);
598 
599  state = palloc(sizeof(OkeysState));
600  sem = palloc0(sizeof(JsonSemAction));
601 
602  state->lex = lex;
603  state->result_size = 256;
604  state->result_count = 0;
605  state->sent_count = 0;
606  state->result = palloc(256 * sizeof(char *));
607 
608  sem->semstate = (void *) state;
609  sem->array_start = okeys_array_start;
610  sem->scalar = okeys_scalar;
611  sem->object_field_start = okeys_object_field_start;
612  /* remainder are all NULL, courtesy of palloc0 above */
613 
614  pg_parse_json(lex, sem);
615  /* keys are now in state->result */
616 
617  pfree(lex->strval->data);
618  pfree(lex->strval);
619  pfree(lex);
620  pfree(sem);
621 
622  MemoryContextSwitchTo(oldcontext);
623  funcctx->user_fctx = (void *) state;
624  }
625 
626  funcctx = SRF_PERCALL_SETUP();
627  state = (OkeysState *) funcctx->user_fctx;
628 
629  if (state->sent_count < state->result_count)
630  {
631  char *nxt = state->result[state->sent_count++];
632 
633  SRF_RETURN_NEXT(funcctx, CStringGetTextDatum(nxt));
634  }
635 
636  /* cleanup to reduce or eliminate memory leaks */
637  for (i = 0; i < state->result_count; i++)
638  pfree(state->result[i]);
639  pfree(state->result);
640  pfree(state);
641 
642  SRF_RETURN_DONE(funcctx);
643 }
644 
645 static void
646 okeys_object_field_start(void *state, char *fname, bool isnull)
647 {
648  OkeysState *_state = (OkeysState *) state;
649 
650  /* only collecting keys for the top level object */
651  if (_state->lex->lex_level != 1)
652  return;
653 
654  /* enlarge result array if necessary */
655  if (_state->result_count >= _state->result_size)
656  {
657  _state->result_size *= 2;
658  _state->result = (char **)
659  repalloc(_state->result, sizeof(char *) * _state->result_size);
660  }
661 
662  /* save a copy of the field name */
663  _state->result[_state->result_count++] = pstrdup(fname);
664 }
665 
666 static void
667 okeys_array_start(void *state)
668 {
669  OkeysState *_state = (OkeysState *) state;
670 
671  /* top level must be a json object */
672  if (_state->lex->lex_level == 0)
673  ereport(ERROR,
674  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
675  errmsg("cannot call %s on an array",
676  "json_object_keys")));
677 }
678 
679 static void
680 okeys_scalar(void *state, char *token, JsonTokenType tokentype)
681 {
682  OkeysState *_state = (OkeysState *) state;
683 
684  /* top level must be a json object */
685  if (_state->lex->lex_level == 0)
686  ereport(ERROR,
687  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
688  errmsg("cannot call %s on a scalar",
689  "json_object_keys")));
690 }
691 
692 /*
693  * json and jsonb getter functions
694  * these implement the -> ->> #> and #>> operators
695  * and the json{b?}_extract_path*(json, text, ...) functions
696  */
697 
698 
699 Datum
700 json_object_field(PG_FUNCTION_ARGS)
701 {
702  text *json = PG_GETARG_TEXT_PP(0);
703  text *fname = PG_GETARG_TEXT_PP(1);
704  char *fnamestr = text_to_cstring(fname);
705  text *result;
706 
707  result = get_worker(json, &fnamestr, NULL, 1, false);
708 
709  if (result != NULL)
710  PG_RETURN_TEXT_P(result);
711  else
712  PG_RETURN_NULL();
713 }
714 
715 Datum
716 jsonb_object_field(PG_FUNCTION_ARGS)
717 {
718  Jsonb *jb = PG_GETARG_JSONB_P(0);
719  text *key = PG_GETARG_TEXT_PP(1);
720  JsonbValue *v;
721 
722  if (!JB_ROOT_IS_OBJECT(jb))
723  PG_RETURN_NULL();
724 
725  v = findJsonbValueFromContainerLen(&jb->root, JB_FOBJECT,
726  VARDATA_ANY(key),
727  VARSIZE_ANY_EXHDR(key));
728 
729  if (v != NULL)
730  PG_RETURN_JSONB_P(JsonbValueToJsonb(v));
731 
732  PG_RETURN_NULL();
733 }
734 
735 Datum
736 json_object_field_text(PG_FUNCTION_ARGS)
737 {
738  text *json = PG_GETARG_TEXT_PP(0);
739  text *fname = PG_GETARG_TEXT_PP(1);
740  char *fnamestr = text_to_cstring(fname);
741  text *result;
742 
743  result = get_worker(json, &fnamestr, NULL, 1, true);
744 
745  if (result != NULL)
746  PG_RETURN_TEXT_P(result);
747  else
748  PG_RETURN_NULL();
749 }
750 
751 Datum
752 jsonb_object_field_text(PG_FUNCTION_ARGS)
753 {
754  Jsonb *jb = PG_GETARG_JSONB_P(0);
755  text *key = PG_GETARG_TEXT_PP(1);
756  JsonbValue *v;
757 
758  if (!JB_ROOT_IS_OBJECT(jb))
759  PG_RETURN_NULL();
760 
761  v = findJsonbValueFromContainerLen(&jb->root, JB_FOBJECT,
762  VARDATA_ANY(key),
763  VARSIZE_ANY_EXHDR(key));
764 
765  if (v != NULL)
766  {
767  text *result = NULL;
768 
769  switch (v->type)
770  {
771  case jbvNull:
772  break;
773  case jbvBool:
774  result = cstring_to_text(v->val.boolean ? "true" : "false");
775  break;
776  case jbvString:
777  result = cstring_to_text_with_len(v->val.string.val, v->val.string.len);
778  break;
779  case jbvNumeric:
780  result = cstring_to_text(DatumGetCString(DirectFunctionCall1(numeric_out,
781  PointerGetDatum(v->val.numeric))));
782  break;
783  case jbvBinary:
784  {
785  StringInfo jtext = makeStringInfo();
786 
787  (void) JsonbToCString(jtext, v->val.binary.data, -1);
788  result = cstring_to_text_with_len(jtext->data, jtext->len);
789  }
790  break;
791  default:
792  elog(ERROR, "unrecognized jsonb type: %d", (int) v->type);
793  }
794 
795  if (result)
796  PG_RETURN_TEXT_P(result);
797  }
798 
799  PG_RETURN_NULL();
800 }
801 
802 Datum
803 json_array_element(PG_FUNCTION_ARGS)
804 {
805  text *json = PG_GETARG_TEXT_PP(0);
806  int element = PG_GETARG_INT32(1);
807  text *result;
808 
809  result = get_worker(json, NULL, &element, 1, false);
810 
811  if (result != NULL)
812  PG_RETURN_TEXT_P(result);
813  else
814  PG_RETURN_NULL();
815 }
816 
817 Datum
818 jsonb_array_element(PG_FUNCTION_ARGS)
819 {
820  Jsonb *jb = PG_GETARG_JSONB_P(0);
821  int element = PG_GETARG_INT32(1);
822  JsonbValue *v;
823 
824  if (!JB_ROOT_IS_ARRAY(jb))
825  PG_RETURN_NULL();
826 
827  /* Handle negative subscript */
828  if (element < 0)
829  {
830  uint32 nelements = JB_ROOT_COUNT(jb);
831 
832  if (-element > nelements)
833  PG_RETURN_NULL();
834  else
835  element += nelements;
836  }
837 
838  v = getIthJsonbValueFromContainer(&jb->root, element);
839  if (v != NULL)
840  PG_RETURN_JSONB_P(JsonbValueToJsonb(v));
841 
842  PG_RETURN_NULL();
843 }
844 
845 Datum
846 json_array_element_text(PG_FUNCTION_ARGS)
847 {
848  text *json = PG_GETARG_TEXT_PP(0);
849  int element = PG_GETARG_INT32(1);
850  text *result;
851 
852  result = get_worker(json, NULL, &element, 1, true);
853 
854  if (result != NULL)
855  PG_RETURN_TEXT_P(result);
856  else
857  PG_RETURN_NULL();
858 }
859 
860 Datum
861 jsonb_array_element_text(PG_FUNCTION_ARGS)
862 {
863  Jsonb *jb = PG_GETARG_JSONB_P(0);
864  int element = PG_GETARG_INT32(1);
865  JsonbValue *v;
866 
867  if (!JB_ROOT_IS_ARRAY(jb))
868  PG_RETURN_NULL();
869 
870  /* Handle negative subscript */
871  if (element < 0)
872  {
873  uint32 nelements = JB_ROOT_COUNT(jb);
874 
875  if (-element > nelements)
876  PG_RETURN_NULL();
877  else
878  element += nelements;
879  }
880 
881  v = getIthJsonbValueFromContainer(&jb->root, element);
882  if (v != NULL)
883  {
884  text *result = NULL;
885 
886  switch (v->type)
887  {
888  case jbvNull:
889  break;
890  case jbvBool:
891  result = cstring_to_text(v->val.boolean ? "true" : "false");
892  break;
893  case jbvString:
894  result = cstring_to_text_with_len(v->val.string.val, v->val.string.len);
895  break;
896  case jbvNumeric:
897  result = cstring_to_text(DatumGetCString(DirectFunctionCall1(numeric_out,
898  PointerGetDatum(v->val.numeric))));
899  break;
900  case jbvBinary:
901  {
902  StringInfo jtext = makeStringInfo();
903 
904  (void) JsonbToCString(jtext, v->val.binary.data, -1);
905  result = cstring_to_text_with_len(jtext->data, jtext->len);
906  }
907  break;
908  default:
909  elog(ERROR, "unrecognized jsonb type: %d", (int) v->type);
910  }
911 
912  if (result)
913  PG_RETURN_TEXT_P(result);
914  }
915 
916  PG_RETURN_NULL();
917 }
918 
919 Datum
920 json_extract_path(PG_FUNCTION_ARGS)
921 {
922  return get_path_all(fcinfo, false);
923 }
924 
925 Datum
926 json_extract_path_text(PG_FUNCTION_ARGS)
927 {
928  return get_path_all(fcinfo, true);
929 }
930 
931 /*
932  * common routine for extract_path functions
933  */
934 static Datum
935 get_path_all(FunctionCallInfo fcinfo, bool as_text)
936 {
937  text *json = PG_GETARG_TEXT_PP(0);
938  ArrayType *path = PG_GETARG_ARRAYTYPE_P(1);
939  text *result;
940  Datum *pathtext;
941  bool *pathnulls;
942  int npath;
943  char **tpath;
944  int *ipath;
945  int i;
946 
947  /*
948  * If the array contains any null elements, return NULL, on the grounds
949  * that you'd have gotten NULL if any RHS value were NULL in a nested
950  * series of applications of the -> operator. (Note: because we also
951  * return NULL for error cases such as no-such-field, this is true
952  * regardless of the contents of the rest of the array.)
953  */
954  if (array_contains_nulls(path))
955  PG_RETURN_NULL();
956 
957  deconstruct_array(path, TEXTOID, -1, false, 'i',
958  &pathtext, &pathnulls, &npath);
959 
960  tpath = palloc(npath * sizeof(char *));
961  ipath = palloc(npath * sizeof(int));
962 
963  for (i = 0; i < npath; i++)
964  {
965  Assert(!pathnulls[i]);
966  tpath[i] = TextDatumGetCString(pathtext[i]);
967 
968  /*
969  * we have no idea at this stage what structure the document is so
970  * just convert anything in the path that we can to an integer and set
971  * all the other integers to INT_MIN which will never match.
972  */
973  if (*tpath[i] != '\0')
974  {
975  long ind;
976  char *endptr;
977 
978  errno = 0;
979  ind = strtol(tpath[i], &endptr, 10);
980  if (*endptr == '\0' && errno == 0 && ind <= INT_MAX && ind >= INT_MIN)
981  ipath[i] = (int) ind;
982  else
983  ipath[i] = INT_MIN;
984  }
985  else
986  ipath[i] = INT_MIN;
987  }
988 
989  result = get_worker(json, tpath, ipath, npath, as_text);
990 
991  if (result != NULL)
992  PG_RETURN_TEXT_P(result);
993  else
994  PG_RETURN_NULL();
995 }
996 
997 /*
998  * get_worker
999  *
1000  * common worker for all the json getter functions
1001  *
1002  * json: JSON object (in text form)
1003  * tpath[]: field name(s) to extract
1004  * ipath[]: array index(es) (zero-based) to extract, accepts negatives
1005  * npath: length of tpath[] and/or ipath[]
1006  * normalize_results: true to de-escape string and null scalars
1007  *
1008  * tpath can be NULL, or any one tpath[] entry can be NULL, if an object
1009  * field is not to be matched at that nesting level. Similarly, ipath can
1010  * be NULL, or any one ipath[] entry can be INT_MIN if an array element is
1011  * not to be matched at that nesting level (a json datum should never be
1012  * large enough to have -INT_MIN elements due to MaxAllocSize restriction).
1013  */
1014 static text *
1015 get_worker(text *json,
1016  char **tpath,
1017  int *ipath,
1018  int npath,
1019  bool normalize_results)
1020 {
1021  JsonLexContext *lex = makeJsonLexContext(json, true);
1022  JsonSemAction *sem = palloc0(sizeof(JsonSemAction));
1023  GetState *state = palloc0(sizeof(GetState));
1024 
1025  Assert(npath >= 0);
1026 
1027  state->lex = lex;
1028  /* is it "_as_text" variant? */
1029  state->normalize_results = normalize_results;
1030  state->npath = npath;
1031  state->path_names = tpath;
1032  state->path_indexes = ipath;
1033  state->pathok = palloc0(sizeof(bool) * npath);
1034  state->array_cur_index = palloc(sizeof(int) * npath);
1035 
1036  if (npath > 0)
1037  state->pathok[0] = true;
1038 
1039  sem->semstate = (void *) state;
1040 
1041  /*
1042  * Not all variants need all the semantic routines. Only set the ones that
1043  * are actually needed for maximum efficiency.
1044  */
1045  sem->scalar = get_scalar;
1046  if (npath == 0)
1047  {
1048  sem->object_start = get_object_start;
1049  sem->object_end = get_object_end;
1050  sem->array_start = get_array_start;
1051  sem->array_end = get_array_end;
1052  }
1053  if (tpath != NULL)
1054  {
1055  sem->object_field_start = get_object_field_start;
1056  sem->object_field_end = get_object_field_end;
1057  }
1058  if (ipath != NULL)
1059  {
1060  sem->array_start = get_array_start;
1061  sem->array_element_start = get_array_element_start;
1062  sem->array_element_end = get_array_element_end;
1063  }
1064 
1065  pg_parse_json(lex, sem);
1066 
1067  return state->tresult;
1068 }
1069 
1070 static void
1071 get_object_start(void *state)
1072 {
1073  GetState *_state = (GetState *) state;
1074  int lex_level = _state->lex->lex_level;
1075 
1076  if (lex_level == 0 && _state->npath == 0)
1077  {
1078  /*
1079  * Special case: we should match the entire object. We only need this
1080  * at outermost level because at nested levels the match will have
1081  * been started by the outer field or array element callback.
1082  */
1083  _state->result_start = _state->lex->token_start;
1084  }
1085 }
1086 
1087 static void
1088 get_object_end(void *state)
1089 {
1090  GetState *_state = (GetState *) state;
1091  int lex_level = _state->lex->lex_level;
1092 
1093  if (lex_level == 0 && _state->npath == 0)
1094  {
1095  /* Special case: return the entire object */
1096  char *start = _state->result_start;
1097  int len = _state->lex->prev_token_terminator - start;
1098 
1099  _state->tresult = cstring_to_text_with_len(start, len);
1100  }
1101 }
1102 
1103 static void
1104 get_object_field_start(void *state, char *fname, bool isnull)
1105 {
1106  GetState *_state = (GetState *) state;
1107  bool get_next = false;
1108  int lex_level = _state->lex->lex_level;
1109 
1110  if (lex_level <= _state->npath &&
1111  _state->pathok[lex_level - 1] &&
1112  _state->path_names != NULL &&
1113  _state->path_names[lex_level - 1] != NULL &&
1114  strcmp(fname, _state->path_names[lex_level - 1]) == 0)
1115  {
1116  if (lex_level < _state->npath)
1117  {
1118  /* if not at end of path just mark path ok */
1119  _state->pathok[lex_level] = true;
1120  }
1121  else
1122  {
1123  /* end of path, so we want this value */
1124  get_next = true;
1125  }
1126  }
1127 
1128  if (get_next)
1129  {
1130  /* this object overrides any previous matching object */
1131  _state->tresult = NULL;
1132  _state->result_start = NULL;
1133 
1134  if (_state->normalize_results &&
1135  _state->lex->token_type == JSON_TOKEN_STRING)
1136  {
1137  /* for as_text variants, tell get_scalar to set it for us */
1138  _state->next_scalar = true;
1139  }
1140  else
1141  {
1142  /* for non-as_text variants, just note the json starting point */
1143  _state->result_start = _state->lex->token_start;
1144  }
1145  }
1146 }
1147 
1148 static void
1149 get_object_field_end(void *state, char *fname, bool isnull)
1150 {
1151  GetState *_state = (GetState *) state;
1152  bool get_last = false;
1153  int lex_level = _state->lex->lex_level;
1154 
1155  /* same tests as in get_object_field_start */
1156  if (lex_level <= _state->npath &&
1157  _state->pathok[lex_level - 1] &&
1158  _state->path_names != NULL &&
1159  _state->path_names[lex_level - 1] != NULL &&
1160  strcmp(fname, _state->path_names[lex_level - 1]) == 0)
1161  {
1162  if (lex_level < _state->npath)
1163  {
1164  /* done with this field so reset pathok */
1165  _state->pathok[lex_level] = false;
1166  }
1167  else
1168  {
1169  /* end of path, so we want this value */
1170  get_last = true;
1171  }
1172  }
1173 
1174  /* for as_text scalar case, our work is already done */
1175  if (get_last && _state->result_start != NULL)
1176  {
1177  /*
1178  * make a text object from the string from the previously noted json
1179  * start up to the end of the previous token (the lexer is by now
1180  * ahead of us on whatever came after what we're interested in).
1181  */
1182  if (isnull && _state->normalize_results)
1183  _state->tresult = (text *) NULL;
1184  else
1185  {
1186  char *start = _state->result_start;
1187  int len = _state->lex->prev_token_terminator - start;
1188 
1189  _state->tresult = cstring_to_text_with_len(start, len);
1190  }
1191 
1192  /* this should be unnecessary but let's do it for cleanliness: */
1193  _state->result_start = NULL;
1194  }
1195 }
1196 
1197 static void
1198 get_array_start(void *state)
1199 {
1200  GetState *_state = (GetState *) state;
1201  int lex_level = _state->lex->lex_level;
1202 
1203  if (lex_level < _state->npath)
1204  {
1205  /* Initialize counting of elements in this array */
1206  _state->array_cur_index[lex_level] = -1;
1207 
1208  /* INT_MIN value is reserved to represent invalid subscript */
1209  if (_state->path_indexes[lex_level] < 0 &&
1210  _state->path_indexes[lex_level] != INT_MIN)
1211  {
1212  /* Negative subscript -- convert to positive-wise subscript */
1213  int nelements = json_count_array_elements(_state->lex);
1214 
1215  if (-_state->path_indexes[lex_level] <= nelements)
1216  _state->path_indexes[lex_level] += nelements;
1217  }
1218  }
1219  else if (lex_level == 0 && _state->npath == 0)
1220  {
1221  /*
1222  * Special case: we should match the entire array. We only need this
1223  * at the outermost level because at nested levels the match will have
1224  * been started by the outer field or array element callback.
1225  */
1226  _state->result_start = _state->lex->token_start;
1227  }
1228 }
1229 
1230 static void
1231 get_array_end(void *state)
1232 {
1233  GetState *_state = (GetState *) state;
1234  int lex_level = _state->lex->lex_level;
1235 
1236  if (lex_level == 0 && _state->npath == 0)
1237  {
1238  /* Special case: return the entire array */
1239  char *start = _state->result_start;
1240  int len = _state->lex->prev_token_terminator - start;
1241 
1242  _state->tresult = cstring_to_text_with_len(start, len);
1243  }
1244 }
1245 
1246 static void
1247 get_array_element_start(void *state, bool isnull)
1248 {
1249  GetState *_state = (GetState *) state;
1250  bool get_next = false;
1251  int lex_level = _state->lex->lex_level;
1252 
1253  /* Update array element counter */
1254  if (lex_level <= _state->npath)
1255  _state->array_cur_index[lex_level - 1]++;
1256 
1257  if (lex_level <= _state->npath &&
1258  _state->pathok[lex_level - 1] &&
1259  _state->path_indexes != NULL &&
1260  _state->array_cur_index[lex_level - 1] == _state->path_indexes[lex_level - 1])
1261  {
1262  if (lex_level < _state->npath)
1263  {
1264  /* if not at end of path just mark path ok */
1265  _state->pathok[lex_level] = true;
1266  }
1267  else
1268  {
1269  /* end of path, so we want this value */
1270  get_next = true;
1271  }
1272  }
1273 
1274  /* same logic as for objects */
1275  if (get_next)
1276  {
1277  _state->tresult = NULL;
1278  _state->result_start = NULL;
1279 
1280  if (_state->normalize_results &&
1281  _state->lex->token_type == JSON_TOKEN_STRING)
1282  {
1283  _state->next_scalar = true;
1284  }
1285  else
1286  {
1287  _state->result_start = _state->lex->token_start;
1288  }
1289  }
1290 }
1291 
1292 static void
1293 get_array_element_end(void *state, bool isnull)
1294 {
1295  GetState *_state = (GetState *) state;
1296  bool get_last = false;
1297  int lex_level = _state->lex->lex_level;
1298 
1299  /* same tests as in get_array_element_start */
1300  if (lex_level <= _state->npath &&
1301  _state->pathok[lex_level - 1] &&
1302  _state->path_indexes != NULL &&
1303  _state->array_cur_index[lex_level - 1] == _state->path_indexes[lex_level - 1])
1304  {
1305  if (lex_level < _state->npath)
1306  {
1307  /* done with this element so reset pathok */
1308  _state->pathok[lex_level] = false;
1309  }
1310  else
1311  {
1312  /* end of path, so we want this value */
1313  get_last = true;
1314  }
1315  }
1316 
1317  /* same logic as for objects */
1318  if (get_last && _state->result_start != NULL)
1319  {
1320  if (isnull && _state->normalize_results)
1321  _state->tresult = (text *) NULL;
1322  else
1323  {
1324  char *start = _state->result_start;
1325  int len = _state->lex->prev_token_terminator - start;
1326 
1327  _state->tresult = cstring_to_text_with_len(start, len);
1328  }
1329 
1330  _state->result_start = NULL;
1331  }
1332 }
1333 
1334 static void
1335 get_scalar(void *state, char *token, JsonTokenType tokentype)
1336 {
1337  GetState *_state = (GetState *) state;
1338  int lex_level = _state->lex->lex_level;
1339 
1340  /* Check for whole-object match */
1341  if (lex_level == 0 && _state->npath == 0)
1342  {
1343  if (_state->normalize_results && tokentype == JSON_TOKEN_STRING)
1344  {
1345  /* we want the de-escaped string */
1346  _state->next_scalar = true;
1347  }
1348  else if (_state->normalize_results && tokentype == JSON_TOKEN_NULL)
1349  {
1350  _state->tresult = (text *) NULL;
1351  }
1352  else
1353  {
1354  /*
1355  * This is a bit hokey: we will suppress whitespace after the
1356  * scalar token, but not whitespace before it. Probably not worth
1357  * doing our own space-skipping to avoid that.
1358  */
1359  char *start = _state->lex->input;
1360  int len = _state->lex->prev_token_terminator - start;
1361 
1362  _state->tresult = cstring_to_text_with_len(start, len);
1363  }
1364  }
1365 
1366  if (_state->next_scalar)
1367  {
1368  /* a de-escaped text value is wanted, so supply it */
1369  _state->tresult = cstring_to_text(token);
1370  /* make sure the next call to get_scalar doesn't overwrite it */
1371  _state->next_scalar = false;
1372  }
1373 }
1374 
1375 Datum
1376 jsonb_extract_path(PG_FUNCTION_ARGS)
1377 {
1378  return get_jsonb_path_all(fcinfo, false);
1379 }
1380 
1381 Datum
1382 jsonb_extract_path_text(PG_FUNCTION_ARGS)
1383 {
1384  return get_jsonb_path_all(fcinfo, true);
1385 }
1386 
1387 static Datum
1388 get_jsonb_path_all(FunctionCallInfo fcinfo, bool as_text)
1389 {
1390  Jsonb *jb = PG_GETARG_JSONB_P(0);
1391  ArrayType *path = PG_GETARG_ARRAYTYPE_P(1);
1392  Jsonb *res;
1393  Datum *pathtext;
1394  bool *pathnulls;
1395  int npath;
1396  int i;
1397  bool have_object = false,
1398  have_array = false;
1399  JsonbValue *jbvp = NULL;
1400  JsonbValue tv;
1401  JsonbContainer *container;
1402 
1403  /*
1404  * If the array contains any null elements, return NULL, on the grounds
1405  * that you'd have gotten NULL if any RHS value were NULL in a nested
1406  * series of applications of the -> operator. (Note: because we also
1407  * return NULL for error cases such as no-such-field, this is true
1408  * regardless of the contents of the rest of the array.)
1409  */
1410  if (array_contains_nulls(path))
1411  PG_RETURN_NULL();
1412 
1413  deconstruct_array(path, TEXTOID, -1, false, 'i',
1414  &pathtext, &pathnulls, &npath);
1415 
1416  /* Identify whether we have object, array, or scalar at top-level */
1417  container = &jb->root;
1418 
1419  if (JB_ROOT_IS_OBJECT(jb))
1420  have_object = true;
1421  else if (JB_ROOT_IS_ARRAY(jb) && !JB_ROOT_IS_SCALAR(jb))
1422  have_array = true;
1423  else
1424  {
1425  Assert(JB_ROOT_IS_ARRAY(jb) && JB_ROOT_IS_SCALAR(jb));
1426  /* Extract the scalar value, if it is what we'll return */
1427  if (npath <= 0)
1428  jbvp = getIthJsonbValueFromContainer(container, 0);
1429  }
1430 
1431  /*
1432  * If the array is empty, return the entire LHS object, on the grounds
1433  * that we should do zero field or element extractions. For the
1434  * non-scalar case we can just hand back the object without much work. For
1435  * the scalar case, fall through and deal with the value below the loop.
1436  * (This inconsistency arises because there's no easy way to generate a
1437  * JsonbValue directly for root-level containers.)
1438  */
1439  if (npath <= 0 && jbvp == NULL)
1440  {
1441  if (as_text)
1442  {
1443  PG_RETURN_TEXT_P(cstring_to_text(JsonbToCString(NULL,
1444  container,
1445  VARSIZE(jb))));
1446  }
1447  else
1448  {
1449  /* not text mode - just hand back the jsonb */
1450  PG_RETURN_JSONB_P(jb);
1451  }
1452  }
1453 
1454  for (i = 0; i < npath; i++)
1455  {
1456  if (have_object)
1457  {
1458  jbvp = findJsonbValueFromContainerLen(container,
1459  JB_FOBJECT,
1460  VARDATA(pathtext[i]),
1461  VARSIZE(pathtext[i]) - VARHDRSZ);
1462  }
1463  else if (have_array)
1464  {
1465  long lindex;
1466  uint32 index;
1467  char *indextext = TextDatumGetCString(pathtext[i]);
1468  char *endptr;
1469 
1470  errno = 0;
1471  lindex = strtol(indextext, &endptr, 10);
1472  if (endptr == indextext || *endptr != '\0' || errno != 0 ||
1473  lindex > INT_MAX || lindex < INT_MIN)
1474  PG_RETURN_NULL();
1475 
1476  if (lindex >= 0)
1477  {
1478  index = (uint32) lindex;
1479  }
1480  else
1481  {
1482  /* Handle negative subscript */
1483  uint32 nelements;
1484 
1485  /* Container must be array, but make sure */
1486  if (!JsonContainerIsArray(container))
1487  elog(ERROR, "not a jsonb array");
1488 
1489  nelements = JsonContainerSize(container);
1490 
1491  if (-lindex > nelements)
1492  PG_RETURN_NULL();
1493  else
1494  index = nelements + lindex;
1495  }
1496 
1497  jbvp = getIthJsonbValueFromContainer(container, index);
1498  }
1499  else
1500  {
1501  /* scalar, extraction yields a null */
1502  PG_RETURN_NULL();
1503  }
1504 
1505  if (jbvp == NULL)
1506  PG_RETURN_NULL();
1507  else if (i == npath - 1)
1508  break;
1509 
1510  if (jbvp->type == jbvBinary)
1511  {
1512  JsonbIterator *it = JsonbIteratorInit((JsonbContainer *) jbvp->val.binary.data);
1513  JsonbIteratorToken r;
1514 
1515  r = JsonbIteratorNext(&it, &tv, true);
1516  container = (JsonbContainer *) jbvp->val.binary.data;
1517  have_object = r == WJB_BEGIN_OBJECT;
1518  have_array = r == WJB_BEGIN_ARRAY;
1519  }
1520  else
1521  {
1522  have_object = jbvp->type == jbvObject;
1523  have_array = jbvp->type == jbvArray;
1524  }
1525  }
1526 
1527  if (as_text)
1528  {
1529  /* special-case outputs for string and null values */
1530  if (jbvp->type == jbvString)
1531  PG_RETURN_TEXT_P(cstring_to_text_with_len(jbvp->val.string.val,
1532  jbvp->val.string.len));
1533  if (jbvp->type == jbvNull)
1534  PG_RETURN_NULL();
1535  }
1536 
1537  res = JsonbValueToJsonb(jbvp);
1538 
1539  if (as_text)
1540  {
1541  PG_RETURN_TEXT_P(cstring_to_text(JsonbToCString(NULL,
1542  &res->root,
1543  VARSIZE(res))));
1544  }
1545  else
1546  {
1547  /* not text mode - just hand back the jsonb */
1548  PG_RETURN_JSONB_P(res);
1549  }
1550 }
1551 
1552 /*
1553  * SQL function json_array_length(json) -> int
1554  */
1555 Datum
1556 json_array_length(PG_FUNCTION_ARGS)
1557 {
1558  text *json = PG_GETARG_TEXT_PP(0);
1559  AlenState *state;
1560  JsonLexContext *lex;
1561  JsonSemAction *sem;
1562 
1563  lex = makeJsonLexContext(json, false);
1564  state = palloc0(sizeof(AlenState));
1565  sem = palloc0(sizeof(JsonSemAction));
1566 
1567  /* palloc0 does this for us */
1568 #if 0
1569  state->count = 0;
1570 #endif
1571  state->lex = lex;
1572 
1573  sem->semstate = (void *) state;
1574  sem->object_start = alen_object_start;
1575  sem->scalar = alen_scalar;
1576  sem->array_element_start = alen_array_element_start;
1577 
1578  pg_parse_json(lex, sem);
1579 
1580  PG_RETURN_INT32(state->count);
1581 }
1582 
1583 Datum
1584 jsonb_array_length(PG_FUNCTION_ARGS)
1585 {
1586  Jsonb *jb = PG_GETARG_JSONB_P(0);
1587 
1588  if (JB_ROOT_IS_SCALAR(jb))
1589  ereport(ERROR,
1590  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1591  errmsg("cannot get array length of a scalar")));
1592  else if (!JB_ROOT_IS_ARRAY(jb))
1593  ereport(ERROR,
1594  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1595  errmsg("cannot get array length of a non-array")));
1596 
1597  PG_RETURN_INT32(JB_ROOT_COUNT(jb));
1598 }
1599 
1600 /*
1601  * These next two checks ensure that the json is an array (since it can't be
1602  * a scalar or an object).
1603  */
1604 
1605 static void
1606 alen_object_start(void *state)
1607 {
1608  AlenState *_state = (AlenState *) state;
1609 
1610  /* json structure check */
1611  if (_state->lex->lex_level == 0)
1612  ereport(ERROR,
1613  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1614  errmsg("cannot get array length of a non-array")));
1615 }
1616 
1617 static void
1618 alen_scalar(void *state, char *token, JsonTokenType tokentype)
1619 {
1620  AlenState *_state = (AlenState *) state;
1621 
1622  /* json structure check */
1623  if (_state->lex->lex_level == 0)
1624  ereport(ERROR,
1625  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1626  errmsg("cannot get array length of a scalar")));
1627 }
1628 
1629 static void
1630 alen_array_element_start(void *state, bool isnull)
1631 {
1632  AlenState *_state = (AlenState *) state;
1633 
1634  /* just count up all the level 1 elements */
1635  if (_state->lex->lex_level == 1)
1636  _state->count++;
1637 }
1638 
1639 /*
1640  * SQL function json_each and json_each_text
1641  *
1642  * decompose a json object into key value pairs.
1643  *
1644  * Unlike json_object_keys() these SRFs operate in materialize mode,
1645  * stashing results into a Tuplestore object as they go.
1646  * The construction of tuples is done using a temporary memory context
1647  * that is cleared out after each tuple is built.
1648  */
1649 Datum
1650 json_each(PG_FUNCTION_ARGS)
1651 {
1652  return each_worker(fcinfo, false);
1653 }
1654 
1655 Datum
1656 jsonb_each(PG_FUNCTION_ARGS)
1657 {
1658  return each_worker_jsonb(fcinfo, "jsonb_each", false);
1659 }
1660 
1661 Datum
1662 json_each_text(PG_FUNCTION_ARGS)
1663 {
1664  return each_worker(fcinfo, true);
1665 }
1666 
1667 Datum
1668 jsonb_each_text(PG_FUNCTION_ARGS)
1669 {
1670  return each_worker_jsonb(fcinfo, "jsonb_each_text", true);
1671 }
1672 
1673 static Datum
1674 each_worker_jsonb(FunctionCallInfo fcinfo, const char *funcname, bool as_text)
1675 {
1676  Jsonb *jb = PG_GETARG_JSONB_P(0);
1677  ReturnSetInfo *rsi;
1678  Tuplestorestate *tuple_store;
1679  TupleDesc tupdesc;
1680  TupleDesc ret_tdesc;
1681  MemoryContext old_cxt,
1682  tmp_cxt;
1683  bool skipNested = false;
1684  JsonbIterator *it;
1685  JsonbValue v;
1686  JsonbIteratorToken r;
1687 
1688  if (!JB_ROOT_IS_OBJECT(jb))
1689  ereport(ERROR,
1690  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1691  errmsg("cannot call %s on a non-object",
1692  funcname)));
1693 
1694  rsi = (ReturnSetInfo *) fcinfo->resultinfo;
1695 
1696  if (!rsi || !IsA(rsi, ReturnSetInfo) ||
1697  (rsi->allowedModes & SFRM_Materialize) == 0 ||
1698  rsi->expectedDesc == NULL)
1699  ereport(ERROR,
1700  (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1701  errmsg("set-valued function called in context that "
1702  "cannot accept a set")));
1703 
1704  rsi->returnMode = SFRM_Materialize;
1705 
1706  if (get_call_result_type(fcinfo, NULL, &tupdesc) != TYPEFUNC_COMPOSITE)
1707  ereport(ERROR,
1708  (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1709  errmsg("function returning record called in context "
1710  "that cannot accept type record")));
1711 
1712  old_cxt = MemoryContextSwitchTo(rsi->econtext->ecxt_per_query_memory);
1713 
1714  ret_tdesc = CreateTupleDescCopy(tupdesc);
1715  BlessTupleDesc(ret_tdesc);
1716  tuple_store =
1717  tuplestore_begin_heap(rsi->allowedModes & SFRM_Materialize_Random,
1718  false, work_mem);
1719 
1720  MemoryContextSwitchTo(old_cxt);
1721 
1722  tmp_cxt = AllocSetContextCreate(CurrentMemoryContext,
1723  "jsonb_each temporary cxt",
1724  ALLOCSET_DEFAULT_SIZES);
1725 
1726  it = JsonbIteratorInit(&jb->root);
1727 
1728  while ((r = JsonbIteratorNext(&it, &v, skipNested)) != WJB_DONE)
1729  {
1730  skipNested = true;
1731 
1732  if (r == WJB_KEY)
1733  {
1734  text *key;
1735  HeapTuple tuple;
1736  Datum values[2];
1737  bool nulls[2] = {false, false};
1738 
1739  /* Use the tmp context so we can clean up after each tuple is done */
1740  old_cxt = MemoryContextSwitchTo(tmp_cxt);
1741 
1742  key = cstring_to_text_with_len(v.val.string.val, v.val.string.len);
1743 
1744  /*
1745  * The next thing the iterator fetches should be the value, no
1746  * matter what shape it is.
1747  */
1748  r = JsonbIteratorNext(&it, &v, skipNested);
1749  Assert(r != WJB_DONE);
1750 
1751  values[0] = PointerGetDatum(key);
1752 
1753  if (as_text)
1754  {
1755  if (v.type == jbvNull)
1756  {
1757  /* a json null is an sql null in text mode */
1758  nulls[1] = true;
1759  values[1] = (Datum) NULL;
1760  }
1761  else
1762  {
1763  text *sv;
1764 
1765  if (v.type == jbvString)
1766  {
1767  /* In text mode, scalar strings should be dequoted */
1768  sv = cstring_to_text_with_len(v.val.string.val, v.val.string.len);
1769  }
1770  else
1771  {
1772  /* Turn anything else into a json string */
1773  StringInfo jtext = makeStringInfo();
1774  Jsonb *jb = JsonbValueToJsonb(&v);
1775 
1776  (void) JsonbToCString(jtext, &jb->root, 0);
1777  sv = cstring_to_text_with_len(jtext->data, jtext->len);
1778  }
1779 
1780  values[1] = PointerGetDatum(sv);
1781  }
1782  }
1783  else
1784  {
1785  /* Not in text mode, just return the Jsonb */
1786  Jsonb *val = JsonbValueToJsonb(&v);
1787 
1788  values[1] = PointerGetDatum(val);
1789  }
1790 
1791  tuple = heap_form_tuple(ret_tdesc, values, nulls);
1792 
1793  tuplestore_puttuple(tuple_store, tuple);
1794 
1795  /* clean up and switch back */
1796  MemoryContextSwitchTo(old_cxt);
1797  MemoryContextReset(tmp_cxt);
1798  }
1799  }
1800 
1801  MemoryContextDelete(tmp_cxt);
1802 
1803  rsi->setResult = tuple_store;
1804  rsi->setDesc = ret_tdesc;
1805 
1806  PG_RETURN_NULL();
1807 }
1808 
1809 
1810 static Datum
1811 each_worker(FunctionCallInfo fcinfo, bool as_text)
1812 {
1813  text *json = PG_GETARG_TEXT_PP(0);
1814  JsonLexContext *lex;
1815  JsonSemAction *sem;
1816  ReturnSetInfo *rsi;
1817  MemoryContext old_cxt;
1818  TupleDesc tupdesc;
1819  EachState *state;
1820 
1821  lex = makeJsonLexContext(json, true);
1822  state = palloc0(sizeof(EachState));
1823  sem = palloc0(sizeof(JsonSemAction));
1824 
1825  rsi = (ReturnSetInfo *) fcinfo->resultinfo;
1826 
1827  if (!rsi || !IsA(rsi, ReturnSetInfo) ||
1828  (rsi->allowedModes & SFRM_Materialize) == 0 ||
1829  rsi->expectedDesc == NULL)
1830  ereport(ERROR,
1831  (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1832  errmsg("set-valued function called in context that "
1833  "cannot accept a set")));
1834 
1835  rsi->returnMode = SFRM_Materialize;
1836 
1837  (void) get_call_result_type(fcinfo, NULL, &tupdesc);
1838 
1839  /* make these in a sufficiently long-lived memory context */
1840  old_cxt = MemoryContextSwitchTo(rsi->econtext->ecxt_per_query_memory);
1841 
1842  state->ret_tdesc = CreateTupleDescCopy(tupdesc);
1843  BlessTupleDesc(state->ret_tdesc);
1844  state->tuple_store =
1845  tuplestore_begin_heap(rsi->allowedModes & SFRM_Materialize_Random,
1846  false, work_mem);
1847 
1848  MemoryContextSwitchTo(old_cxt);
1849 
1850  sem->semstate = (void *) state;
1851  sem->array_start = each_array_start;
1852  sem->scalar = each_scalar;
1853  sem->object_field_start = each_object_field_start;
1854  sem->object_field_end = each_object_field_end;
1855 
1856  state->normalize_results = as_text;
1857  state->next_scalar = false;
1858  state->lex = lex;
1859  state->tmp_cxt = AllocSetContextCreate(CurrentMemoryContext,
1860  "json_each temporary cxt",
1861  ALLOCSET_DEFAULT_SIZES);
1862 
1863  pg_parse_json(lex, sem);
1864 
1865  MemoryContextDelete(state->tmp_cxt);
1866 
1867  rsi->setResult = state->tuple_store;
1868  rsi->setDesc = state->ret_tdesc;
1869 
1870  PG_RETURN_NULL();
1871 }
1872 
1873 
1874 static void
1875 each_object_field_start(void *state, char *fname, bool isnull)
1876 {
1877  EachState *_state = (EachState *) state;
1878 
1879  /* save a pointer to where the value starts */
1880  if (_state->lex->lex_level == 1)
1881  {
1882  /*
1883  * next_scalar will be reset in the object_field_end handler, and
1884  * since we know the value is a scalar there is no danger of it being
1885  * on while recursing down the tree.
1886  */
1887  if (_state->normalize_results && _state->lex->token_type == JSON_TOKEN_STRING)
1888  _state->next_scalar = true;
1889  else
1890  _state->result_start = _state->lex->token_start;
1891  }
1892 }
1893 
1894 static void
1895 each_object_field_end(void *state, char *fname, bool isnull)
1896 {
1897  EachState *_state = (EachState *) state;
1898  MemoryContext old_cxt;
1899  int len;
1900  text *val;
1901  HeapTuple tuple;
1902  Datum values[2];
1903  bool nulls[2] = {false, false};
1904 
1905  /* skip over nested objects */
1906  if (_state->lex->lex_level != 1)
1907  return;
1908 
1909  /* use the tmp context so we can clean up after each tuple is done */
1910  old_cxt = MemoryContextSwitchTo(_state->tmp_cxt);
1911 
1912  values[0] = CStringGetTextDatum(fname);
1913 
1914  if (isnull && _state->normalize_results)
1915  {
1916  nulls[1] = true;
1917  values[1] = (Datum) 0;
1918  }
1919  else if (_state->next_scalar)
1920  {
1921  values[1] = CStringGetTextDatum(_state->normalized_scalar);
1922  _state->next_scalar = false;
1923  }
1924  else
1925  {
1926  len = _state->lex->prev_token_terminator - _state->result_start;
1927  val = cstring_to_text_with_len(_state->result_start, len);
1928  values[1] = PointerGetDatum(val);
1929  }
1930 
1931  tuple = heap_form_tuple(_state->ret_tdesc, values, nulls);
1932 
1933  tuplestore_puttuple(_state->tuple_store, tuple);
1934 
1935  /* clean up and switch back */
1936  MemoryContextSwitchTo(old_cxt);
1937  MemoryContextReset(_state->tmp_cxt);
1938 }
1939 
1940 static void
1941 each_array_start(void *state)
1942 {
1943  EachState *_state = (EachState *) state;
1944 
1945  /* json structure check */
1946  if (_state->lex->lex_level == 0)
1947  ereport(ERROR,
1948  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1949  errmsg("cannot deconstruct an array as an object")));
1950 }
1951 
1952 static void
1953 each_scalar(void *state, char *token, JsonTokenType tokentype)
1954 {
1955  EachState *_state = (EachState *) state;
1956 
1957  /* json structure check */
1958  if (_state->lex->lex_level == 0)
1959  ereport(ERROR,
1960  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1961  errmsg("cannot deconstruct a scalar")));
1962 
1963  /* supply de-escaped value if required */
1964  if (_state->next_scalar)
1965  _state->normalized_scalar = token;
1966 }
1967 
1968 /*
1969  * SQL functions json_array_elements and json_array_elements_text
1970  *
1971  * get the elements from a json array
1972  *
1973  * a lot of this processing is similar to the json_each* functions
1974  */
1975 
1976 Datum
1977 jsonb_array_elements(PG_FUNCTION_ARGS)
1978 {
1979  return elements_worker_jsonb(fcinfo, "jsonb_array_elements", false);
1980 }
1981 
1982 Datum
1983 jsonb_array_elements_text(PG_FUNCTION_ARGS)
1984 {
1985  return elements_worker_jsonb(fcinfo, "jsonb_array_elements_text", true);
1986 }
1987 
1988 static Datum
1989 elements_worker_jsonb(FunctionCallInfo fcinfo, const char *funcname,
1990  bool as_text)
1991 {
1992  Jsonb *jb = PG_GETARG_JSONB_P(0);
1993  ReturnSetInfo *rsi;
1994  Tuplestorestate *tuple_store;
1995  TupleDesc tupdesc;
1996  TupleDesc ret_tdesc;
1997  MemoryContext old_cxt,
1998  tmp_cxt;
1999  bool skipNested = false;
2000  JsonbIterator *it;
2001  JsonbValue v;
2002  JsonbIteratorToken r;
2003 
2004  if (JB_ROOT_IS_SCALAR(jb))
2005  ereport(ERROR,
2006  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
2007  errmsg("cannot extract elements from a scalar")));
2008  else if (!JB_ROOT_IS_ARRAY(jb))
2009  ereport(ERROR,
2010  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
2011  errmsg("cannot extract elements from an object")));
2012 
2013  rsi = (ReturnSetInfo *) fcinfo->resultinfo;
2014 
2015  if (!rsi || !IsA(rsi, ReturnSetInfo) ||
2016  (rsi->allowedModes & SFRM_Materialize) == 0 ||
2017  rsi->expectedDesc == NULL)
2018  ereport(ERROR,
2019  (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
2020  errmsg("set-valued function called in context that "
2021  "cannot accept a set")));
2022 
2023  rsi->returnMode = SFRM_Materialize;
2024 
2025  /* it's a simple type, so don't use get_call_result_type() */
2026  tupdesc = rsi->expectedDesc;
2027 
2028  old_cxt = MemoryContextSwitchTo(rsi->econtext->ecxt_per_query_memory);
2029 
2030  ret_tdesc = CreateTupleDescCopy(tupdesc);
2031  BlessTupleDesc(ret_tdesc);
2032  tuple_store =
2033  tuplestore_begin_heap(rsi->allowedModes & SFRM_Materialize_Random,
2034  false, work_mem);
2035 
2036  MemoryContextSwitchTo(old_cxt);
2037 
2038  tmp_cxt = AllocSetContextCreate(CurrentMemoryContext,
2039  "jsonb_array_elements temporary cxt",
2040  ALLOCSET_DEFAULT_SIZES);
2041 
2042  it = JsonbIteratorInit(&jb->root);
2043 
2044  while ((r = JsonbIteratorNext(&it, &v, skipNested)) != WJB_DONE)
2045  {
2046  skipNested = true;
2047 
2048  if (r == WJB_ELEM)
2049  {
2050  HeapTuple tuple;
2051  Datum values[1];
2052  bool nulls[1] = {false};
2053 
2054  /* use the tmp context so we can clean up after each tuple is done */
2055  old_cxt = MemoryContextSwitchTo(tmp_cxt);
2056 
2057  if (!as_text)
2058  {
2059  Jsonb *val = JsonbValueToJsonb(&v);
2060 
2061  values[0] = PointerGetDatum(val);
2062  }
2063  else
2064  {
2065  if (v.type == jbvNull)
2066  {
2067  /* a json null is an sql null in text mode */
2068  nulls[0] = true;
2069  values[0] = (Datum) NULL;
2070  }
2071  else
2072  {
2073  text *sv;
2074 
2075  if (v.type == jbvString)
2076  {
2077  /* in text mode scalar strings should be dequoted */
2078  sv = cstring_to_text_with_len(v.val.string.val, v.val.string.len);
2079  }
2080  else
2081  {
2082  /* turn anything else into a json string */
2083  StringInfo jtext = makeStringInfo();
2084  Jsonb *jb = JsonbValueToJsonb(&v);
2085 
2086  (void) JsonbToCString(jtext, &jb->root, 0);
2087  sv = cstring_to_text_with_len(jtext->data, jtext->len);
2088  }
2089 
2090  values[0] = PointerGetDatum(sv);
2091  }
2092  }
2093 
2094  tuple = heap_form_tuple(ret_tdesc, values, nulls);
2095 
2096  tuplestore_puttuple(tuple_store, tuple);
2097 
2098  /* clean up and switch back */
2099  MemoryContextSwitchTo(old_cxt);
2100  MemoryContextReset(tmp_cxt);
2101  }
2102  }
2103 
2104  MemoryContextDelete(tmp_cxt);
2105 
2106  rsi->setResult = tuple_store;
2107  rsi->setDesc = ret_tdesc;
2108 
2109  PG_RETURN_NULL();
2110 }
2111 
2112 Datum
2113 json_array_elements(PG_FUNCTION_ARGS)
2114 {
2115  return elements_worker(fcinfo, "json_array_elements", false);
2116 }
2117 
2118 Datum
2119 json_array_elements_text(PG_FUNCTION_ARGS)
2120 {
2121  return elements_worker(fcinfo, "json_array_elements_text", true);
2122 }
2123 
2124 static Datum
2125 elements_worker(FunctionCallInfo fcinfo, const char *funcname, bool as_text)
2126 {
2127  text *json = PG_GETARG_TEXT_PP(0);
2128 
2129  /* elements only needs escaped strings when as_text */
2130  JsonLexContext *lex = makeJsonLexContext(json, as_text);
2131  JsonSemAction *sem;
2132  ReturnSetInfo *rsi;
2133  MemoryContext old_cxt;
2134  TupleDesc tupdesc;
2135  ElementsState *state;
2136 
2137  state = palloc0(sizeof(ElementsState));
2138  sem = palloc0(sizeof(JsonSemAction));
2139 
2140  rsi = (ReturnSetInfo *) fcinfo->resultinfo;
2141 
2142  if (!rsi || !IsA(rsi, ReturnSetInfo) ||
2143  (rsi->allowedModes & SFRM_Materialize) == 0 ||
2144  rsi->expectedDesc == NULL)
2145  ereport(ERROR,
2146  (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
2147  errmsg("set-valued function called in context that "
2148  "cannot accept a set")));
2149 
2150  rsi->returnMode = SFRM_Materialize;
2151 
2152  /* it's a simple type, so don't use get_call_result_type() */
2153  tupdesc = rsi->expectedDesc;
2154 
2155  /* make these in a sufficiently long-lived memory context */
2156  old_cxt = MemoryContextSwitchTo(rsi->econtext->ecxt_per_query_memory);
2157 
2158  state->ret_tdesc = CreateTupleDescCopy(tupdesc);
2159  BlessTupleDesc(state->ret_tdesc);
2160  state->tuple_store =
2161  tuplestore_begin_heap(rsi->allowedModes & SFRM_Materialize_Random,
2162  false, work_mem);
2163 
2164  MemoryContextSwitchTo(old_cxt);
2165 
2166  sem->semstate = (void *) state;
2167  sem->object_start = elements_object_start;
2168  sem->scalar = elements_scalar;
2169  sem->array_element_start = elements_array_element_start;
2170  sem->array_element_end = elements_array_element_end;
2171 
2172  state->function_name = funcname;
2173  state->normalize_results = as_text;
2174  state->next_scalar = false;
2175  state->lex = lex;
2176  state->tmp_cxt = AllocSetContextCreate(CurrentMemoryContext,
2177  "json_array_elements temporary cxt",
2178  ALLOCSET_DEFAULT_SIZES);
2179 
2180  pg_parse_json(lex, sem);
2181 
2182  MemoryContextDelete(state->tmp_cxt);
2183 
2184  rsi->setResult = state->tuple_store;
2185  rsi->setDesc = state->ret_tdesc;
2186 
2187  PG_RETURN_NULL();
2188 }
2189 
2190 static void
2191 elements_array_element_start(void *state, bool isnull)
2192 {
2193  ElementsState *_state = (ElementsState *) state;
2194 
2195  /* save a pointer to where the value starts */
2196  if (_state->lex->lex_level == 1)
2197  {
2198  /*
2199  * next_scalar will be reset in the array_element_end handler, and
2200  * since we know the value is a scalar there is no danger of it being
2201  * on while recursing down the tree.
2202  */
2203  if (_state->normalize_results && _state->lex->token_type == JSON_TOKEN_STRING)
2204  _state->next_scalar = true;
2205  else
2206  _state->result_start = _state->lex->token_start;
2207  }
2208 }
2209 
2210 static void
2211 elements_array_element_end(void *state, bool isnull)
2212 {
2213  ElementsState *_state = (ElementsState *) state;
2214  MemoryContext old_cxt;
2215  int len;
2216  text *val;
2217  HeapTuple tuple;
2218  Datum values[1];
2219  bool nulls[1] = {false};
2220 
2221  /* skip over nested objects */
2222  if (_state->lex->lex_level != 1)
2223  return;
2224 
2225  /* use the tmp context so we can clean up after each tuple is done */
2226  old_cxt = MemoryContextSwitchTo(_state->tmp_cxt);
2227 
2228  if (isnull && _state->normalize_results)
2229  {
2230  nulls[0] = true;
2231  values[0] = (Datum) NULL;
2232  }
2233  else if (_state->next_scalar)
2234  {
2235  values[0] = CStringGetTextDatum(_state->normalized_scalar);
2236  _state->next_scalar = false;
2237  }
2238  else
2239  {
2240  len = _state->lex->prev_token_terminator - _state->result_start;
2241  val = cstring_to_text_with_len(_state->result_start, len);
2242  values[0] = PointerGetDatum(val);
2243  }
2244 
2245  tuple = heap_form_tuple(_state->ret_tdesc, values, nulls);
2246 
2247  tuplestore_puttuple(_state->tuple_store, tuple);
2248 
2249  /* clean up and switch back */
2250  MemoryContextSwitchTo(old_cxt);
2251  MemoryContextReset(_state->tmp_cxt);
2252 }
2253 
2254 static void
2255 elements_object_start(void *state)
2256 {
2257  ElementsState *_state = (ElementsState *) state;
2258 
2259  /* json structure check */
2260  if (_state->lex->lex_level == 0)
2261  ereport(ERROR,
2262  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
2263  errmsg("cannot call %s on a non-array",
2264  _state->function_name)));
2265 }
2266 
2267 static void
2268 elements_scalar(void *state, char *token, JsonTokenType tokentype)
2269 {
2270  ElementsState *_state = (ElementsState *) state;
2271 
2272  /* json structure check */
2273  if (_state->lex->lex_level == 0)
2274  ereport(ERROR,
2275  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
2276  errmsg("cannot call %s on a scalar",
2277  _state->function_name)));
2278 
2279  /* supply de-escaped value if required */
2280  if (_state->next_scalar)
2281  _state->normalized_scalar = token;
2282 }
2283 
2284 /*
2285  * SQL function json_populate_record
2286  *
2287  * set fields in a record from the argument json
2288  *
2289  * Code adapted shamelessly from hstore's populate_record
2290  * which is in turn partly adapted from record_out.
2291  *
2292  * The json is decomposed into a hash table, in which each
2293  * field in the record is then looked up by name. For jsonb
2294  * we fetch the values direct from the object.
2295  */
2296 Datum
2297 jsonb_populate_record(PG_FUNCTION_ARGS)
2298 {
2299  return populate_record_worker(fcinfo, "jsonb_populate_record",
2300  false, true);
2301 }
2302 
2303 Datum
2304 jsonb_to_record(PG_FUNCTION_ARGS)
2305 {
2306  return populate_record_worker(fcinfo, "jsonb_to_record",
2307  false, false);
2308 }
2309 
2310 Datum
2311 json_populate_record(PG_FUNCTION_ARGS)
2312 {
2313  return populate_record_worker(fcinfo, "json_populate_record",
2314  true, true);
2315 }
2316 
2317 Datum
2318 json_to_record(PG_FUNCTION_ARGS)
2319 {
2320  return populate_record_worker(fcinfo, "json_to_record",
2321  true, false);
2322 }
2323 
2324 /* helper function for diagnostics */
2325 static void
2326 populate_array_report_expected_array(PopulateArrayContext *ctx, int ndim)
2327 {
2328  if (ndim <= 0)
2329  {
2330  if (ctx->colname)
2331  ereport(ERROR,
2332  (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
2333  errmsg("expected JSON array"),
2334  errhint("See the value of key \"%s\".", ctx->colname)));
2335  else
2336  ereport(ERROR,
2337  (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
2338  errmsg("expected JSON array")));
2339  }
2340  else
2341  {
2342  StringInfoData indices;
2343  int i;
2344 
2345  initStringInfo(&indices);
2346 
2347  Assert(ctx->ndims > 0 && ndim < ctx->ndims);
2348 
2349  for (i = 0; i < ndim; i++)
2350  appendStringInfo(&indices, "[%d]", ctx->sizes[i]);
2351 
2352  if (ctx->colname)
2353  ereport(ERROR,
2354  (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
2355  errmsg("expected JSON array"),
2356  errhint("See the array element %s of key \"%s\".",
2357  indices.data, ctx->colname)));
2358  else
2359  ereport(ERROR,
2360  (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
2361  errmsg("expected JSON array"),
2362  errhint("See the array element %s.",
2363  indices.data)));
2364  }
2365 }
2366 
2367 /* set the number of dimensions of the populated array when it becomes known */
2368 static void
2369 populate_array_assign_ndims(PopulateArrayContext *ctx, int ndims)
2370 {
2371  int i;
2372 
2373  Assert(ctx->ndims <= 0);
2374 
2375  if (ndims <= 0)
2376  populate_array_report_expected_array(ctx, ndims);
2377 
2378  ctx->ndims = ndims;
2379  ctx->dims = palloc(sizeof(int) * ndims);
2380  ctx->sizes = palloc0(sizeof(int) * ndims);
2381 
2382  for (i = 0; i < ndims; i++)
2383  ctx->dims[i] = -1; /* dimensions are unknown yet */
2384 }
2385 
2386 /* check the populated subarray dimension */
2387 static void
2388 populate_array_check_dimension(PopulateArrayContext *ctx, int ndim)
2389 {
2390  int dim = ctx->sizes[ndim]; /* current dimension counter */
2391 
2392  if (ctx->dims[ndim] == -1)
2393  ctx->dims[ndim] = dim; /* assign dimension if not yet known */
2394  else if (ctx->dims[ndim] != dim)
2395  ereport(ERROR,
2396  (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
2397  errmsg("malformed JSON array"),
2398  errdetail("Multidimensional arrays must have "
2399  "sub-arrays with matching dimensions.")));
2400 
2401  /* reset the current array dimension size counter */
2402  ctx->sizes[ndim] = 0;
2403 
2404  /* increment the parent dimension counter if it is a nested sub-array */
2405  if (ndim > 0)
2406  ctx->sizes[ndim - 1]++;
2407 }
2408 
2409 static void
2410 populate_array_element(PopulateArrayContext *ctx, int ndim, JsValue *jsv)
2411 {
2412  Datum element;
2413  bool element_isnull;
2414 
2415  /* populate the array element */
2416  element = populate_record_field(ctx->aio->element_info,
2417  ctx->aio->element_type,
2418  ctx->aio->element_typmod,
2419  NULL, ctx->mcxt, PointerGetDatum(NULL),
2420  jsv, &element_isnull);
2421 
2422  accumArrayResult(ctx->astate, element, element_isnull,
2423  ctx->aio->element_type, ctx->acxt);
2424 
2425  Assert(ndim > 0);
2426  ctx->sizes[ndim - 1]++; /* increment current dimension counter */
2427 }
2428 
2429 /* json object start handler for populate_array_json() */
2430 static void
2431 populate_array_object_start(void *_state)
2432 {
2433  PopulateArrayState *state = (PopulateArrayState *) _state;
2434  int ndim = state->lex->lex_level;
2435 
2436  if (state->ctx->ndims <= 0)
2437  populate_array_assign_ndims(state->ctx, ndim);
2438  else if (ndim < state->ctx->ndims)
2439  populate_array_report_expected_array(state->ctx, ndim);
2440 }
2441 
2442 /* json array end handler for populate_array_json() */
2443 static void
2444 populate_array_array_end(void *_state)
2445 {
2446  PopulateArrayState *state = (PopulateArrayState *) _state;
2447  PopulateArrayContext *ctx = state->ctx;
2448  int ndim = state->lex->lex_level;
2449 
2450  if (ctx->ndims <= 0)
2451  populate_array_assign_ndims(ctx, ndim + 1);
2452 
2453  if (ndim < ctx->ndims)
2454  populate_array_check_dimension(ctx, ndim);
2455 }
2456 
2457 /* json array element start handler for populate_array_json() */
2458 static void
2459 populate_array_element_start(void *_state, bool isnull)
2460 {
2461  PopulateArrayState *state = (PopulateArrayState *) _state;
2462  int ndim = state->lex->lex_level;
2463 
2464  if (state->ctx->ndims <= 0 || ndim == state->ctx->ndims)
2465  {
2466  /* remember current array element start */
2467  state->element_start = state->lex->token_start;
2468  state->element_type = state->lex->token_type;
2469  state->element_scalar = NULL;
2470  }
2471 }
2472 
2473 /* json array element end handler for populate_array_json() */
2474 static void
2475 populate_array_element_end(void *_state, bool isnull)
2476 {
2477  PopulateArrayState *state = (PopulateArrayState *) _state;
2478  PopulateArrayContext *ctx = state->ctx;
2479  int ndim = state->lex->lex_level;
2480 
2481  Assert(ctx->ndims > 0);
2482 
2483  if (ndim == ctx->ndims)
2484  {
2485  JsValue jsv;
2486 
2487  jsv.is_json = true;
2488  jsv.val.json.type = state->element_type;
2489 
2490  if (isnull)
2491  {
2492  Assert(jsv.val.json.type == JSON_TOKEN_NULL);
2493  jsv.val.json.str = NULL;
2494  jsv.val.json.len = 0;
2495  }
2496  else if (state->element_scalar)
2497  {
2498  jsv.val.json.str = state->element_scalar;
2499  jsv.val.json.len = -1; /* null-terminated */
2500  }
2501  else
2502  {
2503  jsv.val.json.str = state->element_start;
2504  jsv.val.json.len = (state->lex->prev_token_terminator -
2505  state->element_start) * sizeof(char);
2506  }
2507 
2508  populate_array_element(ctx, ndim, &jsv);
2509  }
2510 }
2511 
2512 /* json scalar handler for populate_array_json() */
2513 static void
2514 populate_array_scalar(void *_state, char *token, JsonTokenType tokentype)
2515 {
2516  PopulateArrayState *state = (PopulateArrayState *) _state;
2517  PopulateArrayContext *ctx = state->ctx;
2518  int ndim = state->lex->lex_level;
2519 
2520  if (ctx->ndims <= 0)
2521  populate_array_assign_ndims(ctx, ndim);
2522  else if (ndim < ctx->ndims)
2523  populate_array_report_expected_array(ctx, ndim);
2524 
2525  if (ndim == ctx->ndims)
2526  {
2527  /* remember the scalar element token */
2528  state->element_scalar = token;
2529  /* element_type must already be set in populate_array_element_start() */
2530  Assert(state->element_type == tokentype);
2531  }
2532 }
2533 
2534 /* parse a json array and populate array */
2535 static void
2536 populate_array_json(PopulateArrayContext *ctx, char *json, int len)
2537 {
2538  PopulateArrayState state;
2539  JsonSemAction sem;
2540 
2541  state.lex = makeJsonLexContextCstringLen(json, len, true);
2542  state.ctx = ctx;
2543 
2544  memset(&sem, 0, sizeof(sem));
2545  sem.semstate = (void *) &state;
2546  sem.object_start = populate_array_object_start;
2547  sem.array_end = populate_array_array_end;
2548  sem.array_element_start = populate_array_element_start;
2549  sem.array_element_end = populate_array_element_end;
2550  sem.scalar = populate_array_scalar;
2551 
2552  pg_parse_json(state.lex, &sem);
2553 
2554  /* number of dimensions should be already known */
2555  Assert(ctx->ndims > 0 && ctx->dims);
2556 
2557  pfree(state.lex);
2558 }
2559 
2560 /*
2561  * populate_array_dim_jsonb() -- Iterate recursively through jsonb sub-array
2562  * elements and accumulate result using given ArrayBuildState.
2563  */
2564 static void
2565 populate_array_dim_jsonb(PopulateArrayContext *ctx, /* context */
2566  JsonbValue *jbv, /* jsonb sub-array */
2567  int ndim) /* current dimension */
2568 {
2569  JsonbContainer *jbc = jbv->val.binary.data;
2570  JsonbIterator *it;
2571  JsonbIteratorToken tok;
2572  JsonbValue val;
2573  JsValue jsv;
2574 
2575  check_stack_depth();
2576 
2577  if (jbv->type != jbvBinary || !JsonContainerIsArray(jbc))
2578  populate_array_report_expected_array(ctx, ndim - 1);
2579 
2580  Assert(!JsonContainerIsScalar(jbc));
2581 
2582  it = JsonbIteratorInit(jbc);
2583 
2584  tok = JsonbIteratorNext(&it, &val, true);
2585  Assert(tok == WJB_BEGIN_ARRAY);
2586 
2587  tok = JsonbIteratorNext(&it, &val, true);
2588 
2589  /*
2590  * If the number of dimensions is not yet known and we have found end of
2591  * the array, or the first child element is not an array, then assign the
2592  * number of dimensions now.
2593  */
2594  if (ctx->ndims <= 0 &&
2595  (tok == WJB_END_ARRAY ||
2596  (tok == WJB_ELEM &&
2597  (val.type != jbvBinary ||
2598  !JsonContainerIsArray(val.val.binary.data)))))
2599  populate_array_assign_ndims(ctx, ndim);
2600 
2601  jsv.is_json = false;
2602  jsv.val.jsonb = &val;
2603 
2604  /* process all the array elements */
2605  while (tok == WJB_ELEM)
2606  {
2607  /*
2608  * Recurse only if the dimensions of dimensions is still unknown or if
2609  * it is not the innermost dimension.
2610  */
2611  if (ctx->ndims > 0 && ndim >= ctx->ndims)
2612  populate_array_element(ctx, ndim, &jsv);
2613  else
2614  {
2615  /* populate child sub-array */
2616  populate_array_dim_jsonb(ctx, &val, ndim + 1);
2617 
2618  /* number of dimensions should be already known */
2619  Assert(ctx->ndims > 0 && ctx->dims);
2620 
2621  populate_array_check_dimension(ctx, ndim);
2622  }
2623 
2624  tok = JsonbIteratorNext(&it, &val, true);
2625  }
2626 
2627  Assert(tok == WJB_END_ARRAY);
2628 
2629  /* free iterator, iterating until WJB_DONE */
2630  tok = JsonbIteratorNext(&it, &val, true);
2631  Assert(tok == WJB_DONE && !it);
2632 }
2633 
2634 /* recursively populate an array from json/jsonb */
2635 static Datum
2636 populate_array(ArrayIOData *aio,
2637  const char *colname,
2638  MemoryContext mcxt,
2639  JsValue *jsv)
2640 {
2641  PopulateArrayContext ctx;
2642  Datum result;
2643  int *lbs;
2644  int i;
2645 
2646  ctx.aio = aio;
2647  ctx.mcxt = mcxt;
2648  ctx.acxt = CurrentMemoryContext;
2649  ctx.astate = initArrayResult(aio->element_type, ctx.acxt, true);
2650  ctx.colname = colname;
2651  ctx.ndims = 0; /* unknown yet */
2652  ctx.dims = NULL;
2653  ctx.sizes = NULL;
2654 
2655  if (jsv->is_json)
2656  populate_array_json(&ctx, jsv->val.json.str,
2657  jsv->val.json.len >= 0 ? jsv->val.json.len
2658  : strlen(jsv->val.json.str));
2659  else
2660  {
2661  populate_array_dim_jsonb(&ctx, jsv->val.jsonb, 1);
2662  ctx.dims[0] = ctx.sizes[0];
2663  }
2664 
2665  Assert(ctx.ndims > 0);
2666 
2667  lbs = palloc(sizeof(int) * ctx.ndims);
2668 
2669  for (i = 0; i < ctx.ndims; i++)
2670  lbs[i] = 1;
2671 
2672  result = makeMdArrayResult(ctx.astate, ctx.ndims, ctx.dims, lbs,
2673  ctx.acxt, true);
2674 
2675  pfree(ctx.dims);
2676  pfree(ctx.sizes);
2677  pfree(lbs);
2678 
2679  return result;
2680 }
2681 
2682 static void
2683 JsValueToJsObject(JsValue *jsv, JsObject *jso)
2684 {
2685  jso->is_json = jsv->is_json;
2686 
2687  if (jsv->is_json)
2688  {
2689  /* convert plain-text json into a hash table */
2690  jso->val.json_hash =
2691  get_json_object_as_hash(jsv->val.json.str,
2692  jsv->val.json.len >= 0
2693  ? jsv->val.json.len
2694  : strlen(jsv->val.json.str),
2695  "populate_composite");
2696  }
2697  else
2698  {
2699  JsonbValue *jbv = jsv->val.jsonb;
2700 
2701  if (jbv->type == jbvBinary &&
2702  JsonContainerIsObject(jbv->val.binary.data))
2703  {
2704  jso->val.jsonb_cont = jbv->val.binary.data;
2705  }
2706  else
2707  {
2708  bool is_scalar;
2709 
2710  is_scalar = IsAJsonbScalar(jbv) ||
2711  (jbv->type == jbvBinary &&
2712  JsonContainerIsScalar(jbv->val.binary.data));
2713  ereport(ERROR,
2714  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
2715  is_scalar
2716  ? errmsg("cannot call %s on a scalar",
2717  "populate_composite")
2718  : errmsg("cannot call %s on an array",
2719  "populate_composite")));
2720  }
2721  }
2722 }
2723 
2724 /* acquire or update cached tuple descriptor for a composite type */
2725 static void
2726 update_cached_tupdesc(CompositeIOData *io, MemoryContext mcxt)
2727 {
2728  if (!io->tupdesc ||
2729  io->tupdesc->tdtypeid != io->base_typid ||
2730  io->tupdesc->tdtypmod != io->base_typmod)
2731  {
2732  TupleDesc tupdesc = lookup_rowtype_tupdesc(io->base_typid,
2733  io->base_typmod);
2734  MemoryContext oldcxt;
2735 
2736  if (io->tupdesc)
2737  FreeTupleDesc(io->tupdesc);
2738 
2739  /* copy tuple desc without constraints into cache memory context */
2740  oldcxt = MemoryContextSwitchTo(mcxt);
2741  io->tupdesc = CreateTupleDescCopy(tupdesc);
2742  MemoryContextSwitchTo(oldcxt);
2743 
2744  ReleaseTupleDesc(tupdesc);
2745  }
2746 }
2747 
2748 /* recursively populate a composite (row type) value from json/jsonb */
2749 static Datum
2750 populate_composite(CompositeIOData *io,
2751  Oid typid,
2752  const char *colname,
2753  MemoryContext mcxt,
2754  HeapTupleHeader defaultval,
2755  JsValue *jsv,
2756  bool isnull)
2757 {
2758  Datum result;
2759 
2760  /* acquire/update cached tuple descriptor */
2761  update_cached_tupdesc(io, mcxt);
2762 
2763  if (isnull)
2764  result = (Datum) 0;
2765  else
2766  {
2767  HeapTupleHeader tuple;
2768  JsObject jso;
2769 
2770  /* prepare input value */
2771  JsValueToJsObject(jsv, &jso);
2772 
2773  /* populate resulting record tuple */
2774  tuple = populate_record(io->tupdesc, &io->record_io,
2775  defaultval, mcxt, &jso);
2776  result = HeapTupleHeaderGetDatum(tuple);
2777 
2778  JsObjectFree(&jso);
2779  }
2780 
2781  /*
2782  * If it's domain over composite, check domain constraints. (This should
2783  * probably get refactored so that we can see the TYPECAT value, but for
2784  * now, we can tell by comparing typid to base_typid.)
2785  */
2786  if (typid != io->base_typid && typid != RECORDOID)
2787  domain_check(result, isnull, typid, &io->domain_info, mcxt);
2788 
2789  return result;
2790 }
2791 
2792 /* populate non-null scalar value from json/jsonb value */
2793 static Datum
2794 populate_scalar(ScalarIOData *io, Oid typid, int32 typmod, JsValue *jsv)
2795 {
2796  Datum res;
2797  char *str = NULL;
2798  char *json = NULL;
2799 
2800  if (jsv->is_json)
2801  {
2802  int len = jsv->val.json.len;
2803 
2804  json = jsv->val.json.str;
2805  Assert(json);
2806  if (len >= 0)
2807  {
2808  /* Need to copy non-null-terminated string */
2809  str = palloc(len + 1 * sizeof(char));
2810  memcpy(str, json, len);
2811  str[len] = '\0';
2812  }
2813  else
2814  str = json; /* string is already null-terminated */
2815 
2816  /* If converting to json/jsonb, make string into valid JSON literal */
2817  if ((typid == JSONOID || typid == JSONBOID) &&
2818  jsv->val.json.type == JSON_TOKEN_STRING)
2819  {
2820  StringInfoData buf;
2821 
2822  initStringInfo(&buf);
2823  escape_json(&buf, str);
2824  /* free temporary buffer */
2825  if (str != json)
2826  pfree(str);
2827  str = buf.data;
2828  }
2829  }
2830  else
2831  {
2832  JsonbValue *jbv = jsv->val.jsonb;
2833 
2834  if (typid == JSONBOID)
2835  {
2836  Jsonb *jsonb = JsonbValueToJsonb(jbv); /* directly use jsonb */
2837 
2838  return JsonbPGetDatum(jsonb);
2839  }
2840  /* convert jsonb to string for typio call */
2841  else if (typid == JSONOID && jbv->type != jbvBinary)
2842  {
2843  /*
2844  * Convert scalar jsonb (non-scalars are passed here as jbvBinary)
2845  * to json string, preserving quotes around top-level strings.
2846  */
2847  Jsonb *jsonb = JsonbValueToJsonb(jbv);
2848 
2849  str = JsonbToCString(NULL, &jsonb->root, VARSIZE(jsonb));
2850  }
2851  else if (jbv->type == jbvString) /* quotes are stripped */
2852  str = pnstrdup(jbv->val.string.val, jbv->val.string.len);
2853  else if (jbv->type == jbvBool)
2854  str = pstrdup(jbv->val.boolean ? "true" : "false");
2855  else if (jbv->type == jbvNumeric)
2856  str = DatumGetCString(DirectFunctionCall1(numeric_out,
2857  PointerGetDatum(jbv->val.numeric)));
2858  else if (jbv->type == jbvBinary)
2859  str = JsonbToCString(NULL, jbv->val.binary.data,
2860  jbv->val.binary.len);
2861  else
2862  elog(ERROR, "unrecognized jsonb type: %d", (int) jbv->type);
2863  }
2864 
2865  res = InputFunctionCall(&io->typiofunc, str, io->typioparam, typmod);
2866 
2867  /* free temporary buffer */
2868  if (str != json)
2869  pfree(str);
2870 
2871  return res;
2872 }
2873 
2874 static Datum
2875 populate_domain(DomainIOData *io,
2876  Oid typid,
2877  const char *colname,
2878  MemoryContext mcxt,
2879  JsValue *jsv,
2880  bool isnull)
2881 {
2882  Datum res;
2883 
2884  if (isnull)
2885  res = (Datum) 0;
2886  else
2887  {
2888  res = populate_record_field(io->base_io,
2889  io->base_typid, io->base_typmod,
2890  colname, mcxt, PointerGetDatum(NULL),
2891  jsv, &isnull);
2892  Assert(!isnull);
2893  }
2894 
2895  domain_check(res, isnull, typid, &io->domain_info, mcxt);
2896 
2897  return res;
2898 }
2899 
2900 /* prepare column metadata cache for the given type */
2901 static void
2902 prepare_column_cache(ColumnIOData *column,
2903  Oid typid,
2904  int32 typmod,
2905  MemoryContext mcxt,
2906  bool need_scalar)
2907 {
2908  HeapTuple tup;
2909  Form_pg_type type;
2910 
2911  column->typid = typid;
2912  column->typmod = typmod;
2913 
2914  tup = SearchSysCache1(TYPEOID, ObjectIdGetDatum(typid));
2915  if (!HeapTupleIsValid(tup))
2916  elog(ERROR, "cache lookup failed for type %u", typid);
2917 
2918  type = (Form_pg_type) GETSTRUCT(tup);
2919 
2920  if (type->typtype == TYPTYPE_DOMAIN)
2921  {
2922  /*
2923  * We can move directly to the bottom base type; domain_check() will
2924  * take care of checking all constraints for a stack of domains.
2925  */
2926  Oid base_typid;
2927  int32 base_typmod = typmod;
2928 
2929  base_typid = getBaseTypeAndTypmod(typid, &base_typmod);
2930  if (get_typtype(base_typid) == TYPTYPE_COMPOSITE)
2931  {
2932  /* domain over composite has its own code path */
2933  column->typcat = TYPECAT_COMPOSITE_DOMAIN;
2934  column->io.composite.record_io = NULL;
2935  column->io.composite.tupdesc = NULL;
2936  column->io.composite.base_typid = base_typid;
2937  column->io.composite.base_typmod = base_typmod;
2938  column->io.composite.domain_info = NULL;
2939  }
2940  else
2941  {
2942  /* domain over anything else */
2943  column->typcat = TYPECAT_DOMAIN;
2944  column->io.domain.base_typid = base_typid;
2945  column->io.domain.base_typmod = base_typmod;
2946  column->io.domain.base_io =
2947  MemoryContextAllocZero(mcxt, sizeof(ColumnIOData));
2948  column->io.domain.domain_info = NULL;
2949  }
2950  }
2951  else if (type->typtype == TYPTYPE_COMPOSITE || typid == RECORDOID)
2952  {
2953  column->typcat = TYPECAT_COMPOSITE;
2954  column->io.composite.record_io = NULL;
2955  column->io.composite.tupdesc = NULL;
2956  column->io.composite.base_typid = typid;
2957  column->io.composite.base_typmod = typmod;
2958  column->io.composite.domain_info = NULL;
2959  }
2960  else if (type->typlen == -1 && OidIsValid(type->typelem))
2961  {
2962  column->typcat = TYPECAT_ARRAY;
2963  column->io.array.element_info = MemoryContextAllocZero(mcxt,
2964  sizeof(ColumnIOData));
2965  column->io.array.element_type = type->typelem;
2966  /* array element typemod stored in attribute's typmod */
2967  column->io.array.element_typmod = typmod;
2968  }
2969  else
2970  {
2971  column->typcat = TYPECAT_SCALAR;
2972  need_scalar = true;
2973  }
2974 
2975  /* caller can force us to look up scalar_io info even for non-scalars */
2976  if (need_scalar)
2977  {
2978  Oid typioproc;
2979 
2980  getTypeInputInfo(typid, &typioproc, &column->scalar_io.typioparam);
2981  fmgr_info_cxt(typioproc, &column->scalar_io.typiofunc, mcxt);
2982  }
2983 
2984  ReleaseSysCache(tup);
2985 }
2986 
2987 /* recursively populate a record field or an array element from a json/jsonb value */
2988 static Datum
2989 populate_record_field(ColumnIOData *col,
2990  Oid typid,
2991  int32 typmod,
2992  const char *colname,
2993  MemoryContext mcxt,
2994  Datum defaultval,
2995  JsValue *jsv,
2996  bool *isnull)
2997 {
2998  TypeCat typcat;
2999 
3000  check_stack_depth();
3001 
3002  /*
3003  * Prepare column metadata cache for the given type. Force lookup of the
3004  * scalar_io data so that the json string hack below will work.
3005  */
3006  if (col->typid != typid || col->typmod != typmod)
3007  prepare_column_cache(col, typid, typmod, mcxt, true);
3008 
3009  *isnull = JsValueIsNull(jsv);
3010 
3011  typcat = col->typcat;
3012 
3013  /* try to convert json string to a non-scalar type through input function */
3014  if (JsValueIsString(jsv) &&
3015  (typcat == TYPECAT_ARRAY ||
3016  typcat == TYPECAT_COMPOSITE ||
3017  typcat == TYPECAT_COMPOSITE_DOMAIN))
3018  typcat = TYPECAT_SCALAR;
3019 
3020  /* we must perform domain checks for NULLs, otherwise exit immediately */
3021  if (*isnull &&
3022  typcat != TYPECAT_DOMAIN &&
3023  typcat != TYPECAT_COMPOSITE_DOMAIN)
3024  return (Datum) 0;
3025 
3026  switch (typcat)
3027  {
3028  case TYPECAT_SCALAR:
3029  return populate_scalar(&col->scalar_io, typid, typmod, jsv);
3030 
3031  case TYPECAT_ARRAY:
3032  return populate_array(&col->io.array, colname, mcxt, jsv);
3033 
3034  case TYPECAT_COMPOSITE:
3035  case TYPECAT_COMPOSITE_DOMAIN:
3036  return populate_composite(&col->io.composite, typid,
3037  colname, mcxt,
3038  DatumGetPointer(defaultval)
3039  ? DatumGetHeapTupleHeader(defaultval)
3040  : NULL,
3041  jsv, *isnull);
3042 
3043  case TYPECAT_DOMAIN:
3044  return populate_domain(&col->io.domain, typid, colname, mcxt,
3045  jsv, *isnull);
3046 
3047  default:
3048  elog(ERROR, "unrecognized type category '%c'", typcat);
3049  return (Datum) 0;
3050  }
3051 }
3052 
3053 static RecordIOData *
3054 allocate_record_info(MemoryContext mcxt, int ncolumns)
3055 {
3056  RecordIOData *data = (RecordIOData *)
3057  MemoryContextAlloc(mcxt,
3058  offsetof(RecordIOData, columns) +
3059  ncolumns * sizeof(ColumnIOData));
3060 
3061  data->record_type = InvalidOid;
3062  data->record_typmod = 0;
3063  data->ncolumns = ncolumns;
3064  MemSet(data->columns, 0, sizeof(ColumnIOData) * ncolumns);
3065 
3066  return data;
3067 }
3068 
3069 static bool
3070 JsObjectGetField(JsObject *obj, char *field, JsValue *jsv)
3071 {
3072  jsv->is_json = obj->is_json;
3073 
3074  if (jsv->is_json)
3075  {
3076  JsonHashEntry *hashentry = hash_search(obj->val.json_hash, field,
3077  HASH_FIND, NULL);
3078 
3079  jsv->val.json.type = hashentry ? hashentry->type : JSON_TOKEN_NULL;
3080  jsv->val.json.str = jsv->val.json.type == JSON_TOKEN_NULL ? NULL :
3081  hashentry->val;
3082  jsv->val.json.len = jsv->val.json.str ? -1 : 0; /* null-terminated */
3083 
3084  return hashentry != NULL;
3085  }
3086  else
3087  {
3088  jsv->val.jsonb = !obj->val.jsonb_cont ? NULL :
3089  findJsonbValueFromContainerLen(obj->val.jsonb_cont, JB_FOBJECT,
3090  field, strlen(field));
3091 
3092  return jsv->val.jsonb != NULL;
3093  }
3094 }
3095 
3096 /* populate a record tuple from json/jsonb value */
3097 static HeapTupleHeader
3098 populate_record(TupleDesc tupdesc,
3099  RecordIOData **record_p,
3100  HeapTupleHeader defaultval,
3101  MemoryContext mcxt,
3102  JsObject *obj)
3103 {
3104  RecordIOData *record = *record_p;
3105  Datum *values;
3106  bool *nulls;
3107  HeapTuple res;
3108  int ncolumns = tupdesc->natts;
3109  int i;
3110 
3111  /*
3112  * if the input json is empty, we can only skip the rest if we were passed
3113  * in a non-null record, since otherwise there may be issues with domain
3114  * nulls.
3115  */
3116  if (defaultval && JsObjectIsEmpty(obj))
3117  return defaultval;
3118 
3119  /* (re)allocate metadata cache */
3120  if (record == NULL ||
3121  record->ncolumns != ncolumns)
3122  *record_p = record = allocate_record_info(mcxt, ncolumns);
3123 
3124  /* invalidate metadata cache if the record type has changed */
3125  if (record->record_type != tupdesc->tdtypeid ||
3126  record->record_typmod != tupdesc->tdtypmod)
3127  {
3128  MemSet(record, 0, offsetof(RecordIOData, columns) +
3129  ncolumns * sizeof(ColumnIOData));
3130  record->record_type = tupdesc->tdtypeid;
3131  record->record_typmod = tupdesc->tdtypmod;
3132  record->ncolumns = ncolumns;
3133  }
3134 
3135  values = (Datum *) palloc(ncolumns * sizeof(Datum));
3136  nulls = (bool *) palloc(ncolumns * sizeof(bool));
3137 
3138  if (defaultval)
3139  {
3140  HeapTupleData tuple;
3141 
3142  /* Build a temporary HeapTuple control structure */
3143  tuple.t_len = HeapTupleHeaderGetDatumLength(defaultval);
3144  ItemPointerSetInvalid(&(tuple.t_self));
3145  tuple.t_tableOid = InvalidOid;
3146  tuple.t_data = defaultval;
3147 
3148  /* Break down the tuple into fields */
3149  heap_deform_tuple(&tuple, tupdesc, values, nulls);
3150  }
3151  else
3152  {
3153  for (i = 0; i < ncolumns; ++i)
3154  {
3155  values[i] = (Datum) 0;
3156  nulls[i] = true;
3157  }
3158  }
3159 
3160  for (i = 0; i < ncolumns; ++i)
3161  {
3162  Form_pg_attribute att = TupleDescAttr(tupdesc, i);
3163  char *colname = NameStr(att->attname);
3164  JsValue field = {0};
3165  bool found;
3166 
3167  /* Ignore dropped columns in datatype */
3168  if (att->attisdropped)
3169  {
3170  nulls[i] = true;
3171  continue;
3172  }
3173 
3174  found = JsObjectGetField(obj, colname, &field);
3175 
3176  /*
3177  * we can't just skip here if the key wasn't found since we might have
3178  * a domain to deal with. If we were passed in a non-null record
3179  * datum, we assume that the existing values are valid (if they're
3180  * not, then it's not our fault), but if we were passed in a null,
3181  * then every field which we don't populate needs to be run through
3182  * the input function just in case it's a domain type.
3183  */
3184  if (defaultval && !found)
3185  continue;
3186 
3187  values[i] = populate_record_field(&record->columns[i],
3188  att->atttypid,
3189  att->atttypmod,
3190  colname,
3191  mcxt,
3192  nulls[i] ? (Datum) 0 : values[i],
3193  &field,
3194  &nulls[i]);
3195  }
3196 
3197  res = heap_form_tuple(tupdesc, values, nulls);
3198 
3199  pfree(values);
3200  pfree(nulls);
3201 
3202  return res->t_data;
3203 }
3204 
3205 /*
3206  * common worker for json{b}_populate_record() and json{b}_to_record()
3207  * is_json and have_record_arg identify the specific function
3208  */
3209 static Datum
3210 populate_record_worker(FunctionCallInfo fcinfo, const char *funcname,
3211  bool is_json, bool have_record_arg)
3212 {
3213  int json_arg_num = have_record_arg ? 1 : 0;
3214  JsValue jsv = {0};
3215  HeapTupleHeader rec;
3216  Datum rettuple;
3217  JsonbValue jbv;
3218  MemoryContext fnmcxt = fcinfo->flinfo->fn_mcxt;
3219  PopulateRecordCache *cache = fcinfo->flinfo->fn_extra;
3220 
3221  /*
3222  * If first time through, identify input/result record type. Note that
3223  * this stanza looks only at fcinfo context, which can't change during the
3224  * query; so we may not be able to fully resolve a RECORD input type yet.
3225  */
3226  if (!cache)
3227  {
3228  fcinfo->flinfo->fn_extra = cache =
3229  MemoryContextAllocZero(fnmcxt, sizeof(*cache));
3230 
3231  if (have_record_arg)
3232  {
3233  /*
3234  * json{b}_populate_record case: result type will be same as first
3235  * argument's.
3236  */
3237  cache->argtype = get_fn_expr_argtype(fcinfo->flinfo, 0);
3238  prepare_column_cache(&cache->c,
3239  cache->argtype, -1,
3240  fnmcxt, false);
3241  if (cache->c.typcat != TYPECAT_COMPOSITE &&
3242  cache->c.typcat != TYPECAT_COMPOSITE_DOMAIN)
3243  ereport(ERROR,
3244  (errcode(ERRCODE_DATATYPE_MISMATCH),
3245  errmsg("first argument of %s must be a row type",
3246  funcname)));
3247  }
3248  else
3249  {
3250  /*
3251  * json{b}_to_record case: result type is specified by calling
3252  * query. Here it is syntactically impossible to specify the
3253  * target type as domain-over-composite.
3254  */
3255  TupleDesc tupdesc;
3256  MemoryContext old_cxt;
3257 
3258  if (get_call_result_type(fcinfo, NULL, &tupdesc) != TYPEFUNC_COMPOSITE)
3259  ereport(ERROR,
3260  (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3261  errmsg("function returning record called in context "
3262  "that cannot accept type record"),
3263  errhint("Try calling the function in the FROM clause "
3264  "using a column definition list.")));
3265 
3266  Assert(tupdesc);
3267  cache->argtype = tupdesc->tdtypeid;
3268 
3269  /* Save identified tupdesc */
3270  old_cxt = MemoryContextSwitchTo(fnmcxt);
3271  cache->c.io.composite.tupdesc = CreateTupleDescCopy(tupdesc);
3272  cache->c.io.composite.base_typid = tupdesc->tdtypeid;
3273  cache->c.io.composite.base_typmod = tupdesc->tdtypmod;
3274  MemoryContextSwitchTo(old_cxt);
3275  }
3276  }
3277 
3278  /* Collect record arg if we have one */
3279  if (have_record_arg && !PG_ARGISNULL(0))
3280  {
3281  rec = PG_GETARG_HEAPTUPLEHEADER(0);
3282 
3283  /*
3284  * When declared arg type is RECORD, identify actual record type from
3285  * the tuple itself. Note the lookup_rowtype_tupdesc call in
3286  * update_cached_tupdesc will fail if we're unable to do this.
3287  */
3288  if (cache->argtype == RECORDOID)
3289  {
3290  cache->c.io.composite.base_typid = HeapTupleHeaderGetTypeId(rec);
3291  cache->c.io.composite.base_typmod = HeapTupleHeaderGetTypMod(rec);
3292  }
3293  }
3294  else
3295  rec = NULL;
3296 
3297  /* If no JSON argument, just return the record (if any) unchanged */
3298  if (PG_ARGISNULL(json_arg_num))
3299  {
3300  if (rec)
3301  PG_RETURN_POINTER(rec);
3302  else
3303  PG_RETURN_NULL();
3304  }
3305 
3306  jsv.is_json = is_json;
3307 
3308  if (is_json)
3309  {
3310  text *json = PG_GETARG_TEXT_PP(json_arg_num);
3311 
3312  jsv.val.json.str = VARDATA_ANY(json);
3313  jsv.val.json.len = VARSIZE_ANY_EXHDR(json);
3314  jsv.val.json.type = JSON_TOKEN_INVALID; /* not used in
3315  * populate_composite() */
3316  }
3317  else
3318  {
3319  Jsonb *jb = PG_GETARG_JSONB_P(json_arg_num);
3320 
3321  jsv.val.jsonb = &jbv;
3322 
3323  /* fill binary jsonb value pointing to jb */
3324  jbv.type = jbvBinary;
3325  jbv.val.binary.data = &jb->root;
3326  jbv.val.binary.len = VARSIZE(jb) - VARHDRSZ;
3327  }
3328 
3329  rettuple = populate_composite(&cache->c.io.composite, cache->argtype,
3330  NULL, fnmcxt, rec, &jsv, false);
3331 
3332  PG_RETURN_DATUM(rettuple);
3333 }
3334 
3335 /*
3336  * get_json_object_as_hash
3337  *
3338  * decompose a json object into a hash table.
3339  */
3340 static HTAB *
3341 get_json_object_as_hash(char *json, int len, const char *funcname)
3342 {
3343  HASHCTL ctl;
3344  HTAB *tab;
3345  JHashState *state;
3346  JsonLexContext *lex = makeJsonLexContextCstringLen(json, len, true);
3347  JsonSemAction *sem;
3348 
3349  memset(&ctl, 0, sizeof(ctl));
3350  ctl.keysize = NAMEDATALEN;
3351  ctl.entrysize = sizeof(JsonHashEntry);
3352  ctl.hcxt = CurrentMemoryContext;
3353  tab = hash_create("json object hashtable",
3354  100,
3355  &ctl,
3356  HASH_ELEM | HASH_CONTEXT);
3357 
3358  state = palloc0(sizeof(JHashState));
3359  sem = palloc0(sizeof(JsonSemAction));
3360 
3361  state->function_name = funcname;
3362  state->hash = tab;
3363  state->lex = lex;
3364 
3365  sem->semstate = (void *) state;
3366  sem->array_start = hash_array_start;
3367  sem->scalar = hash_scalar;
3368  sem->object_field_start = hash_object_field_start;
3369  sem->object_field_end = hash_object_field_end;
3370 
3371  pg_parse_json(lex, sem);
3372 
3373  return tab;
3374 }
3375 
3376 static void
3377 hash_object_field_start(void *state, char *fname, bool isnull)
3378 {
3379  JHashState *_state = (JHashState *) state;
3380 
3381  if (_state->lex->lex_level > 1)
3382  return;
3383 
3384  /* remember token type */
3385  _state->saved_token_type = _state->lex->token_type;
3386 
3387  if (_state->lex->token_type == JSON_TOKEN_ARRAY_START ||
3388  _state->lex->token_type == JSON_TOKEN_OBJECT_START)
3389  {
3390  /* remember start position of the whole text of the subobject */
3391  _state->save_json_start = _state->lex->token_start;
3392  }
3393  else
3394  {
3395  /* must be a scalar */
3396  _state->save_json_start = NULL;
3397  }
3398 }
3399 
3400 static void
3401 hash_object_field_end(void *state, char *fname, bool isnull)
3402 {
3403  JHashState *_state = (JHashState *) state;
3404  JsonHashEntry *hashentry;
3405  bool found;
3406 
3407  /*
3408  * Ignore nested fields.
3409  */
3410  if (_state->lex->lex_level > 1)
3411  return;
3412 
3413  /*
3414  * Ignore field names >= NAMEDATALEN - they can't match a record field.
3415  * (Note: without this test, the hash code would truncate the string at
3416  * NAMEDATALEN-1, and could then match against a similarly-truncated
3417  * record field name. That would be a reasonable behavior, but this code
3418  * has previously insisted on exact equality, so we keep this behavior.)
3419  */
3420  if (strlen(fname) >= NAMEDATALEN)
3421  return;
3422 
3423  hashentry = hash_search(_state->hash, fname, HASH_ENTER, &found);
3424 
3425  /*
3426  * found being true indicates a duplicate. We don't do anything about
3427  * that, a later field with the same name overrides the earlier field.
3428  */
3429 
3430  hashentry->type = _state->saved_token_type;
3431  Assert(isnull == (hashentry->type == JSON_TOKEN_NULL));
3432 
3433  if (_state->save_json_start != NULL)
3434  {
3435  int len = _state->lex->prev_token_terminator - _state->save_json_start;
3436  char *val = palloc((len + 1) * sizeof(char));
3437 
3438  memcpy(val, _state->save_json_start, len);
3439  val[len] = '\0';
3440  hashentry->val = val;
3441  }
3442  else
3443  {
3444  /* must have had a scalar instead */
3445  hashentry->val = _state->saved_scalar;
3446  }
3447 }
3448 
3449 static void
3450 hash_array_start(void *state)
3451 {
3452  JHashState *_state = (JHashState *) state;
3453 
3454  if (_state->lex->lex_level == 0)
3455  ereport(ERROR,
3456  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
3457  errmsg("cannot call %s on an array", _state->function_name)));
3458 }
3459 
3460 static void
3461 hash_scalar(void *state, char *token, JsonTokenType tokentype)
3462 {
3463  JHashState *_state = (JHashState *) state;
3464 
3465  if (_state->lex->lex_level == 0)
3466  ereport(ERROR,
3467  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
3468  errmsg("cannot call %s on a scalar", _state->function_name)));
3469 
3470  if (_state->lex->lex_level == 1)
3471  {
3472  _state->saved_scalar = token;
3473  /* saved_token_type must already be set in hash_object_field_start() */
3474  Assert(_state->saved_token_type == tokentype);
3475  }
3476 }
3477 
3478 
3479 /*
3480  * SQL function json_populate_recordset
3481  *
3482  * set fields in a set of records from the argument json,
3483  * which must be an array of objects.
3484  *
3485  * similar to json_populate_record, but the tuple-building code
3486  * is pushed down into the semantic action handlers so it's done
3487  * per object in the array.
3488  */
3489 Datum
3490 jsonb_populate_recordset(PG_FUNCTION_ARGS)
3491 {
3492  return populate_recordset_worker(fcinfo, "jsonb_populate_recordset",
3493  false, true);
3494 }
3495 
3496 Datum
3497 jsonb_to_recordset(PG_FUNCTION_ARGS)
3498 {
3499  return populate_recordset_worker(fcinfo, "jsonb_to_recordset",
3500  false, false);
3501 }
3502 
3503 Datum
3504 json_populate_recordset(PG_FUNCTION_ARGS)
3505 {
3506  return populate_recordset_worker(fcinfo, "json_populate_recordset",
3507  true, true);
3508 }
3509 
3510 Datum
3511 json_to_recordset(PG_FUNCTION_ARGS)
3512 {
3513  return populate_recordset_worker(fcinfo, "json_to_recordset",
3514  true, false);
3515 }
3516 
3517 static void
3518 populate_recordset_record(PopulateRecordsetState *state, JsObject *obj)
3519 {
3520  PopulateRecordsetCache *cache = state->cache;
3521  HeapTupleHeader tuphead;
3522  HeapTupleData tuple;
3523 
3524  /* acquire/update cached tuple descriptor */
3525  update_cached_tupdesc(&cache->c.io.composite, cache->fn_mcxt);
3526 
3527  /* replace record fields from json */
3528  tuphead = populate_record(cache->c.io.composite.tupdesc,
3529  &cache->c.io.composite.record_io,
3530  state->rec,
3531  cache->fn_mcxt,
3532  obj);
3533 
3534  /* if it's domain over composite, check domain constraints */
3535  if (cache->c.typcat == TYPECAT_COMPOSITE_DOMAIN)
3536  domain_check(HeapTupleHeaderGetDatum(tuphead), false,
3537  cache->argtype,
3538  &cache->c.io.composite.domain_info,
3539  cache->fn_mcxt);
3540 
3541  /* ok, save into tuplestore */
3542  tuple.t_len = HeapTupleHeaderGetDatumLength(tuphead);
3543  ItemPointerSetInvalid(&(tuple.t_self));
3544  tuple.t_tableOid = InvalidOid;
3545  tuple.t_data = tuphead;
3546 
3547  tuplestore_puttuple(state->tuple_store, &tuple);
3548 }
3549 
3550 /*
3551  * common worker for json{b}_populate_recordset() and json{b}_to_recordset()
3552  * is_json and have_record_arg identify the specific function
3553  */
3554 static Datum
3555 populate_recordset_worker(FunctionCallInfo fcinfo, const char *funcname,
3556  bool is_json, bool have_record_arg)
3557 {
3558  int json_arg_num = have_record_arg ? 1 : 0;
3559  ReturnSetInfo *rsi;
3560  MemoryContext old_cxt;
3561  HeapTupleHeader rec;
3562  PopulateRecordsetCache *cache = fcinfo->flinfo->fn_extra;
3563  PopulateRecordsetState *state;
3564 
3565  rsi = (ReturnSetInfo *) fcinfo->resultinfo;
3566 
3567  if (!rsi || !IsA(rsi, ReturnSetInfo) ||
3568  (rsi->allowedModes & SFRM_Materialize) == 0)
3569  ereport(ERROR,
3570  (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3571  errmsg("set-valued function called in context that "
3572  "cannot accept a set")));
3573 
3574  rsi->returnMode = SFRM_Materialize;
3575 
3576  /*
3577  * If first time through, identify input/result record type. Note that
3578  * this stanza looks only at fcinfo context, which can't change during the
3579  * query; so we may not be able to fully resolve a RECORD input type yet.
3580  */
3581  if (!cache)
3582  {
3583  fcinfo->flinfo->fn_extra = cache =
3584  MemoryContextAllocZero(fcinfo->flinfo->fn_mcxt, sizeof(*cache));
3585  cache->fn_mcxt = fcinfo->flinfo->fn_mcxt;
3586 
3587  if (have_record_arg)
3588  {
3589  /*
3590  * json{b}_populate_recordset case: result type will be same as
3591  * first argument's.
3592  */
3593  cache->argtype = get_fn_expr_argtype(fcinfo->flinfo, 0);
3594  prepare_column_cache(&cache->c,
3595  cache->argtype, -1,
3596  cache->fn_mcxt, false);
3597  if (cache->c.typcat != TYPECAT_COMPOSITE &&
3598  cache->c.typcat != TYPECAT_COMPOSITE_DOMAIN)
3599  ereport(ERROR,
3600  (errcode(ERRCODE_DATATYPE_MISMATCH),
3601  errmsg("first argument of %s must be a row type",
3602  funcname)));
3603  }
3604  else
3605  {
3606  /*
3607  * json{b}_to_recordset case: result type is specified by calling
3608  * query. Here it is syntactically impossible to specify the
3609  * target type as domain-over-composite.
3610  */
3611  TupleDesc tupdesc;
3612 
3613  if (get_call_result_type(fcinfo, NULL, &tupdesc) != TYPEFUNC_COMPOSITE)
3614  ereport(ERROR,
3615  (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3616  errmsg("function returning record called in context "
3617  "that cannot accept type record"),
3618  errhint("Try calling the function in the FROM clause "
3619  "using a column definition list.")));
3620 
3621  Assert(tupdesc);
3622  cache->argtype = tupdesc->tdtypeid;
3623 
3624  /* Save identified tupdesc */
3625  old_cxt = MemoryContextSwitchTo(cache->fn_mcxt);
3626  cache->c.io.composite.tupdesc = CreateTupleDescCopy(tupdesc);
3627  cache->c.io.composite.base_typid = tupdesc->tdtypeid;
3628  cache->c.io.composite.base_typmod = tupdesc->tdtypmod;
3629  MemoryContextSwitchTo(old_cxt);
3630  }
3631  }
3632 
3633  /* Collect record arg if we have one */
3634  if (have_record_arg && !PG_ARGISNULL(0))
3635  {
3636  rec = PG_GETARG_HEAPTUPLEHEADER(0);
3637 
3638  /*
3639  * When declared arg type is RECORD, identify actual record type from
3640  * the tuple itself. Note the lookup_rowtype_tupdesc call in
3641  * update_cached_tupdesc will fail if we're unable to do this.
3642  */
3643  if (cache->argtype == RECORDOID)
3644  {
3645  cache->c.io.composite.base_typid = HeapTupleHeaderGetTypeId(rec);
3646  cache->c.io.composite.base_typmod = HeapTupleHeaderGetTypMod(rec);
3647  }
3648  }
3649  else
3650  rec = NULL;
3651 
3652  /* if the json is null send back an empty set */
3653  if (PG_ARGISNULL(json_arg_num))
3654  PG_RETURN_NULL();
3655 
3656  /*
3657  * Forcibly update the cached tupdesc, to ensure we have the right tupdesc
3658  * to return even if the JSON contains no rows.
3659  */
3660  update_cached_tupdesc(&cache->c.io.composite, cache->fn_mcxt);
3661 
3662  state = palloc0(sizeof(PopulateRecordsetState));
3663 
3664  /* make tuplestore in a sufficiently long-lived memory context */
3665  old_cxt = MemoryContextSwitchTo(rsi->econtext->ecxt_per_query_memory);
3666  state->tuple_store = tuplestore_begin_heap(rsi->allowedModes &
3667  SFRM_Materialize_Random,
3668  false, work_mem);
3669  MemoryContextSwitchTo(old_cxt);
3670 
3671  state->function_name = funcname;
3672  state->cache = cache;
3673  state->rec = rec;
3674 
3675  if (is_json)
3676  {
3677  text *json = PG_GETARG_TEXT_PP(json_arg_num);
3678  JsonLexContext *lex;
3679  JsonSemAction *sem;
3680 
3681  sem = palloc0(sizeof(JsonSemAction));
3682 
3683  lex = makeJsonLexContext(json, true);
3684 
3685  sem->semstate = (void *) state;
3686  sem->array_start = populate_recordset_array_start;
3687  sem->array_element_start = populate_recordset_array_element_start;
3688  sem->scalar = populate_recordset_scalar;
3689  sem->object_field_start = populate_recordset_object_field_start;
3690  sem->object_field_end = populate_recordset_object_field_end;
3691  sem->object_start = populate_recordset_object_start;
3692  sem->object_end = populate_recordset_object_end;
3693 
3694  state->lex = lex;
3695 
3696  pg_parse_json(lex, sem);
3697  }
3698  else
3699  {
3700  Jsonb *jb = PG_GETARG_JSONB_P(json_arg_num);
3701  JsonbIterator *it;
3702  JsonbValue v;
3703  bool skipNested = false;
3704  JsonbIteratorToken r;
3705 
3706  if (JB_ROOT_IS_SCALAR(jb) || !JB_ROOT_IS_ARRAY(jb))
3707  ereport(ERROR,
3708  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
3709  errmsg("cannot call %s on a non-array",
3710  funcname)));
3711 
3712  it = JsonbIteratorInit(&jb->root);
3713 
3714  while ((r = JsonbIteratorNext(&it, &v, skipNested)) != WJB_DONE)
3715  {
3716  skipNested = true;
3717 
3718  if (r == WJB_ELEM)
3719  {
3720  JsObject obj;
3721 
3722  if (v.type != jbvBinary ||
3723  !JsonContainerIsObject(v.val.binary.data))
3724  ereport(ERROR,
3725  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
3726  errmsg("argument of %s must be an array of objects",
3727  funcname)));
3728 
3729  obj.is_json = false;
3730  obj.val.jsonb_cont = v.val.binary.data;
3731 
3732  populate_recordset_record(state, &obj);
3733  }
3734  }
3735  }
3736 
3737  /*
3738  * Note: we must copy the cached tupdesc because the executor will free
3739  * the passed-back setDesc, but we want to hang onto the cache in case
3740  * we're called again in the same query.
3741  */
3742  rsi->setResult = state->tuple_store;
3743  rsi->setDesc = CreateTupleDescCopy(cache->c.io.composite.tupdesc);
3744 
3745  PG_RETURN_NULL();
3746 }
3747 
3748 static void
3749 populate_recordset_object_start(void *state)
3750 {
3751  PopulateRecordsetState *_state = (PopulateRecordsetState *) state;
3752  int lex_level = _state->lex->lex_level;
3753  HASHCTL ctl;
3754 
3755  /* Reject object at top level: we must have an array at level 0 */
3756  if (lex_level == 0)
3757  ereport(ERROR,
3758  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
3759  errmsg("cannot call %s on an object",
3760  _state->function_name)));
3761 
3762  /* Nested objects require no special processing */
3763  if (lex_level > 1)
3764  return;
3765 
3766  /* Object at level 1: set up a new hash table for this object */
3767  memset(&ctl, 0, sizeof(ctl));
3768  ctl.keysize = NAMEDATALEN;
3769  ctl.entrysize = sizeof(JsonHashEntry);
3770  ctl.hcxt = CurrentMemoryContext;
3771  _state->json_hash = hash_create("json object hashtable",
3772  100,
3773  &ctl,
3774  HASH_ELEM | HASH_CONTEXT);
3775 }
3776 
3777 static void
3778 populate_recordset_object_end(void *state)
3779 {
3780  PopulateRecordsetState *_state = (PopulateRecordsetState *) state;
3781  JsObject obj;
3782 
3783  /* Nested objects require no special processing */
3784  if (_state->lex->lex_level > 1)
3785  return;
3786 
3787  obj.is_json = true;
3788  obj.val.json_hash = _state->json_hash;
3789 
3790  /* Otherwise, construct and return a tuple based on this level-1 object */
3791  populate_recordset_record(_state, &obj);
3792 
3793  /* Done with hash for this object */
3794  hash_destroy(_state->json_hash);
3795  _state->json_hash = NULL;
3796 }
3797 
3798 static void
3799 populate_recordset_array_element_start(void *state, bool isnull)
3800 {
3801  PopulateRecordsetState *_state = (PopulateRecordsetState *) state;
3802 
3803  if (_state->lex->lex_level == 1 &&
3804  _state->lex->token_type != JSON_TOKEN_OBJECT_START)
3805  ereport(ERROR,
3806  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
3807  errmsg("argument of %s must be an array of objects",
3808  _state->function_name)));
3809 }
3810 
3811 static void
3812 populate_recordset_array_start(void *state)
3813 {
3814  /* nothing to do */
3815 }
3816 
3817 static void
3818 populate_recordset_scalar(void *state, char *token, JsonTokenType tokentype)
3819 {
3820  PopulateRecordsetState *_state = (PopulateRecordsetState *) state;
3821 
3822  if (_state->lex->lex_level == 0)
3823  ereport(ERROR,
3824  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
3825  errmsg("cannot call %s on a scalar",
3826  _state->function_name)));
3827 
3828  if (_state->lex->lex_level == 2)
3829  _state->saved_scalar = token;
3830 }
3831 
3832 static void
3833 populate_recordset_object_field_start(void *state, char *fname, bool isnull)
3834 {
3835  PopulateRecordsetState *_state = (PopulateRecordsetState *) state;
3836 
3837  if (_state->lex->lex_level > 2)
3838  return;
3839 
3840  _state->saved_token_type = _state->lex->token_type;
3841 
3842  if (_state->lex->token_type == JSON_TOKEN_ARRAY_START ||
3843  _state->lex->token_type == JSON_TOKEN_OBJECT_START)
3844  {
3845  _state->save_json_start = _state->lex->token_start;
3846  }
3847  else
3848  {
3849  _state->save_json_start = NULL;
3850  }
3851 }
3852 
3853 static void
3854 populate_recordset_object_field_end(void *state, char *fname, bool isnull)
3855 {
3856  PopulateRecordsetState *_state = (PopulateRecordsetState *) state;
3857  JsonHashEntry *hashentry;
3858  bool found;
3859 
3860  /*
3861  * Ignore nested fields.
3862  */
3863  if (_state->lex->lex_level > 2)
3864  return;
3865 
3866  /*
3867  * Ignore field names >= NAMEDATALEN - they can't match a record field.
3868  * (Note: without this test, the hash code would truncate the string at
3869  * NAMEDATALEN-1, and could then match against a similarly-truncated
3870  * record field name. That would be a reasonable behavior, but this code
3871  * has previously insisted on exact equality, so we keep this behavior.)
3872  */
3873  if (strlen(fname) >= NAMEDATALEN)
3874  return;
3875 
3876  hashentry = hash_search(_state->json_hash, fname, HASH_ENTER, &found);
3877 
3878  /*
3879  * found being true indicates a duplicate. We don't do anything about
3880  * that, a later field with the same name overrides the earlier field.
3881  */
3882 
3883  hashentry->type = _state->saved_token_type;
3884  Assert(isnull == (hashentry->type == JSON_TOKEN_NULL));
3885 
3886  if (_state->save_json_start != NULL)
3887  {
3888  int len = _state->lex->prev_token_terminator - _state->save_json_start;
3889  char *val = palloc((len + 1) * sizeof(char));
3890 
3891  memcpy(val, _state->save_json_start, len);
3892  val[len] = '\0';
3893  hashentry->val = val;
3894  }
3895  else
3896  {
3897  /* must have had a scalar instead */
3898  hashentry->val = _state->saved_scalar;
3899  }
3900 }
3901 
3902 /*
3903  * findJsonbValueFromContainer() wrapper that sets up JsonbValue key string.
3904  */
3905 static JsonbValue *
3906 findJsonbValueFromContainerLen(JsonbContainer *container, uint32 flags,
3907  char *key, uint32 keylen)
3908 {
3909  JsonbValue k;
3910 
3911  k.type = jbvString;
3912  k.val.string.val = key;
3913  k.val.string.len = keylen;
3914 
3915  return findJsonbValueFromContainer(container, flags, &k);
3916 }
3917 
3918 /*
3919  * Semantic actions for json_strip_nulls.
3920  *
3921  * Simply repeat the input on the output unless we encounter
3922  * a null object field. State for this is set when the field
3923  * is started and reset when the scalar action (which must be next)
3924  * is called.
3925  */
3926 
3927 static void
3928 sn_object_start(void *state)
3929 {
3930  StripnullState *_state = (StripnullState *) state;
3931 
3932  appendStringInfoCharMacro(_state->strval, '{');
3933 }
3934 
3935 static void
3936 sn_object_end(void *state)
3937 {
3938  StripnullState *_state = (StripnullState *) state;
3939 
3940  appendStringInfoCharMacro(_state->strval, '}');
3941 }
3942 
3943 static void
3944 sn_array_start(void *state)
3945 {
3946  StripnullState *_state = (StripnullState *) state;
3947 
3948  appendStringInfoCharMacro(_state->strval, '[');
3949 }
3950 
3951 static void
3952 sn_array_end(void *state)
3953 {
3954  StripnullState *_state = (StripnullState *) state;
3955 
3956  appendStringInfoCharMacro(_state->strval, ']');
3957 }
3958 
3959 static void
3960 sn_object_field_start(void *state, char *fname, bool isnull)
3961 {
3962  StripnullState *_state = (StripnullState *) state;
3963 
3964  if (isnull)
3965  {
3966  /*
3967  * The next thing must be a scalar or isnull couldn't be true, so
3968  * there is no danger of this state being carried down into a nested
3969  * object or array. The flag will be reset in the scalar action.
3970  */
3971  _state->skip_next_null = true;
3972  return;
3973  }
3974 
3975  if (_state->strval->data[_state->strval->len - 1] != '{')
3976  appendStringInfoCharMacro(_state->strval, ',');
3977 
3978  /*
3979  * Unfortunately we don't have the quoted and escaped string any more, so
3980  * we have to re-escape it.
3981  */
3982  escape_json(_state->strval, fname);
3983 
3984  appendStringInfoCharMacro(_state->strval, ':');
3985 }
3986 
3987 static void
3988 sn_array_element_start(void *state, bool isnull)
3989 {
3990  StripnullState *_state = (StripnullState *) state;
3991 
3992  if (_state->strval->data[_state->strval->len - 1] != '[')
3993  appendStringInfoCharMacro(_state->strval, ',');
3994 }
3995 
3996 static void
3997 sn_scalar(void *state, char *token, JsonTokenType tokentype)
3998 {
3999  StripnullState *_state = (StripnullState *) state;
4000 
4001  if (_state->skip_next_null)
4002  {
4003  Assert(tokentype == JSON_TOKEN_NULL);
4004  _state->skip_next_null = false;
4005  return;
4006  }
4007 
4008  if (tokentype == JSON_TOKEN_STRING)
4009  escape_json(_state->strval, token);
4010  else
4011  appendStringInfoString(_state->strval, token);
4012 }
4013 
4014 /*
4015  * SQL function json_strip_nulls(json) -> json
4016  */
4017 Datum
4018 json_strip_nulls(PG_FUNCTION_ARGS)
4019 {
4020  text *json = PG_GETARG_TEXT_PP(0);
4021  StripnullState *state;
4022  JsonLexContext *lex;
4023  JsonSemAction *sem;
4024 
4025  lex = makeJsonLexContext(json, true);
4026  state = palloc0(sizeof(StripnullState));
4027  sem = palloc0(sizeof(JsonSemAction));
4028 
4029  state->strval = makeStringInfo();
4030  state->skip_next_null = false;
4031  state->lex = lex;
4032 
4033  sem->semstate = (void *) state;
4034  sem->object_start = sn_object_start;
4035  sem->object_end = sn_object_end;
4036  sem->array_start = sn_array_start;
4037  sem->array_end = sn_array_end;
4038  sem->scalar = sn_scalar;
4039  sem->array_element_start = sn_array_element_start;
4040  sem->object_field_start = sn_object_field_start;
4041 
4042  pg_parse_json(lex, sem);
4043 
4044  PG_RETURN_TEXT_P(cstring_to_text_with_len(state->strval->data,
4045  state->strval->len));
4046 
4047 }
4048 
4049 /*
4050  * SQL function jsonb_strip_nulls(jsonb) -> jsonb
4051  */
4052 Datum
4053 jsonb_strip_nulls(PG_FUNCTION_ARGS)
4054 {
4055  Jsonb *jb = PG_GETARG_JSONB_P(0);
4056  JsonbIterator *it;
4057  JsonbParseState *parseState = NULL;
4058  JsonbValue *res = NULL;
4059  JsonbValue v,
4060  k;
4061  JsonbIteratorToken type;
4062  bool last_was_key = false;
4063 
4064  if (JB_ROOT_IS_SCALAR(jb))
4065  PG_RETURN_POINTER(jb);
4066 
4067  it = JsonbIteratorInit(&jb->root);
4068 
4069  while ((type = JsonbIteratorNext(&it, &v, false)) != WJB_DONE)
4070  {
4071  Assert(!(type == WJB_KEY && last_was_key));
4072 
4073  if (type == WJB_KEY)
4074  {
4075  /* stash the key until we know if it has a null value */
4076  k = v;
4077  last_was_key = true;
4078  continue;
4079  }
4080 
4081  if (last_was_key)
4082  {
4083  /* if the last element was a key this one can't be */
4084  last_was_key = false;
4085 
4086  /* skip this field if value is null */
4087  if (type == WJB_VALUE && v.type == jbvNull)
4088  continue;
4089 
4090  /* otherwise, do a delayed push of the key */
4091  (void) pushJsonbValue(&parseState, WJB_KEY, &k);
4092  }
4093 
4094  if (type == WJB_VALUE || type == WJB_ELEM)
4095  res = pushJsonbValue(&parseState, type, &v);
4096  else
4097  res = pushJsonbValue(&parseState, type, NULL);
4098  }
4099 
4100  Assert(res != NULL);
4101 
4102  PG_RETURN_POINTER(JsonbValueToJsonb(res));
4103 }
4104 
4105 /*
4106  * Add values from the jsonb to the parse state.
4107  *
4108  * If the parse state container is an object, the jsonb is pushed as
4109  * a value, not a key.
4110  *
4111  * This needs to be done using an iterator because pushJsonbValue doesn't
4112  * like getting jbvBinary values, so we can't just push jb as a whole.
4113  */
4114 static void
4115 addJsonbToParseState(JsonbParseState **jbps, Jsonb *jb)
4116 {
4117  JsonbIterator *it;
4118  JsonbValue *o = &(*jbps)->contVal;
4119  JsonbValue v;
4120  JsonbIteratorToken type;
4121 
4122  it = JsonbIteratorInit(&jb->root);
4123 
4124  Assert(o->type == jbvArray || o->type == jbvObject);
4125 
4126  if (JB_ROOT_IS_SCALAR(jb))
4127  {
4128  (void) JsonbIteratorNext(&it, &v, false); /* skip array header */
4129  Assert(v.type == jbvArray);
4130  (void) JsonbIteratorNext(&it, &v, false); /* fetch scalar value */
4131 
4132  switch (o->type)
4133  {
4134  case jbvArray:
4135  (void) pushJsonbValue(jbps, WJB_ELEM, &v);
4136  break;
4137  case jbvObject:
4138  (void) pushJsonbValue(jbps, WJB_VALUE, &v);
4139  break;
4140  default:
4141  elog(ERROR, "unexpected parent of nested structure");
4142  }
4143  }
4144  else
4145  {
4146  while ((type = JsonbIteratorNext(&it, &v, false)) != WJB_DONE)
4147  {
4148  if (type == WJB_KEY || type == WJB_VALUE || type == WJB_ELEM)
4149  (void) pushJsonbValue(jbps, type, &v);
4150  else
4151  (void) pushJsonbValue(jbps, type, NULL);
4152  }
4153  }
4154 
4155 }
4156 
4157 /*
4158  * SQL function jsonb_pretty (jsonb)
4159  *
4160  * Pretty-printed text for the jsonb
4161  */
4162 Datum
4163 jsonb_pretty(PG_FUNCTION_ARGS)
4164 {
4165  Jsonb *jb = PG_GETARG_JSONB_P(0);
4166  StringInfo str = makeStringInfo();
4167 
4168  JsonbToCStringIndent(str, &jb->root, VARSIZE(jb));
4169 
4170  PG_RETURN_TEXT_P(cstring_to_text_with_len(str->data, str->len));
4171 }
4172 
4173 /*
4174  * SQL function jsonb_concat (jsonb, jsonb)
4175  *
4176  * function for || operator
4177  */
4178 Datum
4179 jsonb_concat(PG_FUNCTION_ARGS)
4180 {
4181  Jsonb *jb1 = PG_GETARG_JSONB_P(0);
4182  Jsonb *jb2 = PG_GETARG_JSONB_P(1);
4183  JsonbParseState *state = NULL;
4184  JsonbValue *res;
4185  JsonbIterator *it1,
4186  *it2;
4187 
4188  /*
4189  * If one of the jsonb is empty, just return the other if it's not scalar
4190  * and both are of the same kind. If it's a scalar or they are of
4191  * different kinds we need to perform the concatenation even if one is
4192  * empty.
4193  */
4194  if (JB_ROOT_IS_OBJECT(jb1) == JB_ROOT_IS_OBJECT(jb2))
4195  {
4196  if (JB_ROOT_COUNT(jb1) == 0 && !JB_ROOT_IS_SCALAR(jb2))
4197  PG_RETURN_JSONB_P(jb2);
4198  else if (JB_ROOT_COUNT(jb2) == 0 && !JB_ROOT_IS_SCALAR(jb1))
4199  PG_RETURN_JSONB_P(jb1);
4200  }
4201 
4202  it1 = JsonbIteratorInit(&jb1->root);
4203  it2 = JsonbIteratorInit(&jb2->root);
4204 
4205  res = IteratorConcat(&it1, &it2, &state);
4206 
4207  Assert(res != NULL);
4208 
4209  PG_RETURN_JSONB_P(JsonbValueToJsonb(res));
4210 }
4211 
4212 
4213 /*
4214  * SQL function jsonb_delete (jsonb, text)
4215  *
4216  * return a copy of the jsonb with the indicated item
4217  * removed.
4218  */
4219 Datum
4220 jsonb_delete(PG_FUNCTION_ARGS)
4221 {
4222  Jsonb *in = PG_GETARG_JSONB_P(0);
4223  text *key = PG_GETARG_TEXT_PP(1);
4224  char *keyptr = VARDATA_ANY(key);
4225  int keylen = VARSIZE_ANY_EXHDR(key);
4226  JsonbParseState *state = NULL;
4227  JsonbIterator *it;
4228  JsonbValue v,
4229  *res = NULL;
4230  bool skipNested = false;
4231  JsonbIteratorToken r;
4232 
4233  if (JB_ROOT_IS_SCALAR(in))
4234  ereport(ERROR,
4235  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4236  errmsg("cannot delete from scalar")));
4237 
4238  if (JB_ROOT_COUNT(in) == 0)
4239  PG_RETURN_JSONB_P(in);
4240 
4241  it = JsonbIteratorInit(&in->root);
4242 
4243  while ((r = JsonbIteratorNext(&it, &v, skipNested)) != WJB_DONE)
4244  {
4245  skipNested = true;
4246 
4247  if ((r == WJB_ELEM || r == WJB_KEY) &&
4248  (v.type == jbvString && keylen == v.val.string.len &&
4249  memcmp(keyptr, v.val.string.val, keylen) == 0))
4250  {
4251  /* skip corresponding value as well */
4252  if (r == WJB_KEY)
4253  (void) JsonbIteratorNext(&it, &v, true);
4254 
4255  continue;
4256  }
4257 
4258  res = pushJsonbValue(&state, r, r < WJB_BEGIN_ARRAY ? &v : NULL);
4259  }
4260 
4261  Assert(res != NULL);
4262 
4263  PG_RETURN_JSONB_P(JsonbValueToJsonb(res));
4264 }
4265 
4266 /*
4267  * SQL function jsonb_delete (jsonb, variadic text[])
4268  *
4269  * return a copy of the jsonb with the indicated items
4270  * removed.
4271  */
4272 Datum
4273 jsonb_delete_array(PG_FUNCTION_ARGS)
4274 {
4275  Jsonb *in = PG_GETARG_JSONB_P(0);
4276  ArrayType *keys = PG_GETARG_ARRAYTYPE_P(1);
4277  Datum *keys_elems;
4278  bool *keys_nulls;
4279  int keys_len;
4280  JsonbParseState *state = NULL;
4281  JsonbIterator *it;
4282  JsonbValue v,
4283  *res = NULL;
4284  bool skipNested = false;
4285  JsonbIteratorToken r;
4286 
4287  if (ARR_NDIM(keys) > 1)
4288  ereport(ERROR,
4289  (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
4290  errmsg("wrong number of array subscripts")));
4291 
4292  if (JB_ROOT_IS_SCALAR(in))
4293  ereport(ERROR,
4294  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4295  errmsg("cannot delete from scalar")));
4296 
4297  if (JB_ROOT_COUNT(in) == 0)
4298  PG_RETURN_JSONB_P(in);
4299 
4300  deconstruct_array(keys, TEXTOID, -1, false, 'i',
4301  &keys_elems, &keys_nulls, &keys_len);
4302 
4303  if (keys_len == 0)
4304  PG_RETURN_JSONB_P(in);
4305 
4306  it = JsonbIteratorInit(&in->root);
4307 
4308  while ((r = JsonbIteratorNext(&it, &v, skipNested)) != WJB_DONE)
4309  {
4310  skipNested = true;
4311 
4312  if ((r == WJB_ELEM || r == WJB_KEY) && v.type == jbvString)
4313  {
4314  int i;
4315  bool found = false;
4316 
4317  for (i = 0; i < keys_len; i++)
4318  {
4319  char *keyptr;
4320  int keylen;
4321 
4322  if (keys_nulls[i])
4323  continue;
4324 
4325  keyptr = VARDATA_ANY(keys_elems[i]);
4326  keylen = VARSIZE_ANY_EXHDR(keys_elems[i]);
4327  if (keylen == v.val.string.len &&
4328  memcmp(keyptr, v.val.string.val, keylen) == 0)
4329  {
4330  found = true;
4331  break;
4332  }
4333  }
4334  if (found)
4335  {
4336  /* skip corresponding value as well */
4337  if (r == WJB_KEY)
4338  (void) JsonbIteratorNext(&it, &v, true);
4339 
4340  continue;
4341  }
4342  }
4343 
4344  res = pushJsonbValue(&state, r, r < WJB_BEGIN_ARRAY ? &v : NULL);
4345  }
4346 
4347  Assert(res != NULL);
4348 
4349  PG_RETURN_JSONB_P(JsonbValueToJsonb(res));
4350 }
4351 
4352 /*
4353  * SQL function jsonb_delete (jsonb, int)
4354  *
4355  * return a copy of the jsonb with the indicated item
4356  * removed. Negative int means count back from the
4357  * end of the items.
4358  */
4359 Datum
4360 jsonb_delete_idx(PG_FUNCTION_ARGS)
4361 {
4362  Jsonb *in = PG_GETARG_JSONB_P(0);
4363  int idx = PG_GETARG_INT32(1);
4364  JsonbParseState *state = NULL;
4365  JsonbIterator *it;
4366  uint32 i = 0,
4367  n;
4368  JsonbValue v,
4369  *res = NULL;
4370  JsonbIteratorToken r;
4371 
4372  if (JB_ROOT_IS_SCALAR(in))
4373  ereport(ERROR,
4374  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4375  errmsg("cannot delete from scalar")));
4376 
4377  if (JB_ROOT_IS_OBJECT(in))
4378  ereport(ERROR,
4379  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4380  errmsg("cannot delete from object using integer index")));
4381 
4382  if (JB_ROOT_COUNT(in) == 0)
4383  PG_RETURN_JSONB_P(in);
4384 
4385  it = JsonbIteratorInit(&in->root);
4386 
4387  r = JsonbIteratorNext(&it, &v, false);
4388  Assert(r == WJB_BEGIN_ARRAY);
4389  n = v.val.array.nElems;
4390 
4391  if (idx < 0)
4392  {
4393  if (-idx > n)
4394  idx = n;
4395  else
4396  idx = n + idx;
4397  }
4398 
4399  if (idx >= n)
4400  PG_RETURN_JSONB_P(in);
4401 
4402  pushJsonbValue(&state, r, NULL);
4403 
4404  while ((r = JsonbIteratorNext(&it, &v, true)) != WJB_DONE)
4405  {
4406  if (r == WJB_ELEM)
4407  {
4408  if (i++ == idx)
4409  continue;
4410  }
4411 
4412  res = pushJsonbValue(&state, r, r < WJB_BEGIN_ARRAY ? &v : NULL);
4413  }
4414 
4415  Assert(res != NULL);
4416 
4417  PG_RETURN_JSONB_P(JsonbValueToJsonb(res));
4418 }
4419 
4420 /*
4421  * SQL function jsonb_set(jsonb, text[], jsonb, boolean)
4422  *
4423  */
4424 Datum
4425 jsonb_set(PG_FUNCTION_ARGS)
4426 {
4427  Jsonb *in = PG_GETARG_JSONB_P(0);
4428  ArrayType *path = PG_GETARG_ARRAYTYPE_P(1);
4429  Jsonb *newval = PG_GETARG_JSONB_P(2);
4430  bool create = PG_GETARG_BOOL(3);
4431  JsonbValue *res = NULL;
4432  Datum *path_elems;
4433  bool *path_nulls;
4434  int path_len;
4435  JsonbIterator *it;
4436  JsonbParseState *st = NULL;
4437 
4438  if (ARR_NDIM(path) > 1)
4439  ereport(ERROR,
4440  (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
4441  errmsg("wrong number of array subscripts")));
4442 
4443  if (JB_ROOT_IS_SCALAR(in))
4444  ereport(ERROR,
4445  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4446  errmsg("cannot set path in scalar")));
4447 
4448  if (JB_ROOT_COUNT(in) == 0 && !create)
4449  PG_RETURN_JSONB_P(in);
4450 
4451  deconstruct_array(path, TEXTOID, -1, false, 'i',
4452  &path_elems, &path_nulls, &path_len);
4453 
4454  if (path_len == 0)
4455  PG_RETURN_JSONB_P(in);
4456 
4457  it = JsonbIteratorInit(&in->root);
4458 
4459  res = setPath(&it, path_elems, path_nulls, path_len, &st,
4460  0, newval, create ? JB_PATH_CREATE : JB_PATH_REPLACE);
4461 
4462  Assert(res != NULL);
4463 
4464  PG_RETURN_JSONB_P(JsonbValueToJsonb(res));
4465 }
4466 
4467 
4468 /*
4469  * SQL function jsonb_delete_path(jsonb, text[])
4470  */
4471 Datum
4472 jsonb_delete_path(PG_FUNCTION_ARGS)
4473 {
4474  Jsonb *in = PG_GETARG_JSONB_P(0);
4475  ArrayType *path = PG_GETARG_ARRAYTYPE_P(1);
4476  JsonbValue *res = NULL;
4477  Datum *path_elems;
4478  bool *path_nulls;
4479  int path_len;
4480  JsonbIterator *it;
4481  JsonbParseState *st = NULL;
4482 
4483  if (ARR_NDIM(path) > 1)
4484  ereport(ERROR,
4485  (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
4486  errmsg("wrong number of array subscripts")));
4487 
4488  if (JB_ROOT_IS_SCALAR(in))
4489  ereport(ERROR,
4490  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4491  errmsg("cannot delete path in scalar")));
4492 
4493  if (JB_ROOT_COUNT(in) == 0)
4494  PG_RETURN_JSONB_P(in);
4495 
4496  deconstruct_array(path, TEXTOID, -1, false, 'i',
4497  &path_elems, &path_nulls, &path_len);
4498 
4499  if (path_len == 0)
4500  PG_RETURN_JSONB_P(in);
4501 
4502  it = JsonbIteratorInit(&in->root);
4503 
4504  res = setPath(&it, path_elems, path_nulls, path_len, &st,
4505  0, NULL, JB_PATH_DELETE);
4506 
4507  Assert(res != NULL);
4508 
4509  PG_RETURN_JSONB_P(JsonbValueToJsonb(res));
4510 }
4511 
4512 /*
4513  * SQL function jsonb_insert(jsonb, text[], jsonb, boolean)
4514  *
4515  */
4516 Datum
4517 jsonb_insert(PG_FUNCTION_ARGS)
4518 {
4519  Jsonb *in = PG_GETARG_JSONB_P(0);
4520  ArrayType *path = PG_GETARG_ARRAYTYPE_P(1);
4521  Jsonb *newval = PG_GETARG_JSONB_P(2);
4522  bool after = PG_GETARG_BOOL(3);
4523  JsonbValue *res = NULL;
4524  Datum *path_elems;
4525  bool *path_nulls;
4526  int path_len;
4527  JsonbIterator *it;
4528  JsonbParseState *st = NULL;
4529 
4530  if (ARR_NDIM(path) > 1)
4531  ereport(ERROR,
4532  (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
4533  errmsg("wrong number of array subscripts")));
4534 
4535  if (JB_ROOT_IS_SCALAR(in))
4536  ereport(ERROR,
4537  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4538  errmsg("cannot set path in scalar")));
4539 
4540  deconstruct_array(path, TEXTOID, -1, false, 'i',
4541  &path_elems, &path_nulls, &path_len);
4542 
4543  if (path_len == 0)
4544  PG_RETURN_JSONB_P(in);
4545 
4546  it = JsonbIteratorInit(&in->root);
4547 
4548  res = setPath(&it, path_elems, path_nulls, path_len, &st, 0, newval,
4549  after ? JB_PATH_INSERT_AFTER : JB_PATH_INSERT_BEFORE);
4550 
4551  Assert(res != NULL);
4552 
4553  PG_RETURN_JSONB_P(JsonbValueToJsonb(res));
4554 }
4555 
4556 /*
4557  * Iterate over all jsonb objects and merge them into one.
4558  * The logic of this function copied from the same hstore function,
4559  * except the case, when it1 & it2 represents jbvObject.
4560  * In that case we just append the content of it2 to it1 without any
4561  * verifications.
4562  */
4563 static JsonbValue *
4564 IteratorConcat(JsonbIterator **it1, JsonbIterator **it2,
4565  JsonbParseState **state)
4566 {
4567  JsonbValue v1,
4568  v2,
4569  *res = NULL;
4570  JsonbIteratorToken r1,
4571  r2,
4572  rk1,
4573  rk2;
4574 
4575  r1 = rk1 = JsonbIteratorNext(it1, &v1, false);
4576  r2 = rk2 = JsonbIteratorNext(it2, &v2, false);
4577 
4578  /*
4579  * Both elements are objects.
4580  */
4581  if (rk1 == WJB_BEGIN_OBJECT && rk2 == WJB_BEGIN_OBJECT)
4582  {
4583  /*
4584  * Append the all tokens from v1 to res, except last WJB_END_OBJECT
4585  * (because res will not be finished yet).
4586  */
4587  pushJsonbValue(state, r1, NULL);
4588  while ((r1 = JsonbIteratorNext(it1, &v1, true)) != WJB_END_OBJECT)
4589  pushJsonbValue(state, r1, &v1);
4590 
4591  /*
4592  * Append the all tokens from v2 to res, include last WJB_END_OBJECT
4593  * (the concatenation will be completed).
4594  */
4595  while ((r2 = JsonbIteratorNext(it2, &v2, true)) != WJB_DONE)
4596  res = pushJsonbValue(state, r2, r2 != WJB_END_OBJECT ? &v2 : NULL);
4597  }
4598 
4599  /*
4600  * Both elements are arrays (either can be scalar).
4601  */
4602  else if (rk1 == WJB_BEGIN_ARRAY && rk2 == WJB_BEGIN_ARRAY)
4603  {
4604  pushJsonbValue(state, r1, NULL);
4605 
4606  while ((r1 = JsonbIteratorNext(it1, &v1, true)) != WJB_END_ARRAY)
4607  {
4608  Assert(r1 == WJB_ELEM);
4609  pushJsonbValue(state, r1, &v1);
4610  }
4611 
4612  while ((r2 = JsonbIteratorNext(it2, &v2, true)) != WJB_END_ARRAY)
4613  {
4614  Assert(r2 == WJB_ELEM);
4615  pushJsonbValue(state, WJB_ELEM, &v2);
4616  }
4617 
4618  res = pushJsonbValue(state, WJB_END_ARRAY, NULL /* signal to sort */ );
4619  }
4620  /* have we got array || object or object || array? */
4621  else if (((rk1 == WJB_BEGIN_ARRAY && !(*it1)->isScalar) && rk2 == WJB_BEGIN_OBJECT) ||
4622  (rk1 == WJB_BEGIN_OBJECT && (rk2 == WJB_BEGIN_ARRAY && !(*it2)->isScalar)))
4623  {
4624 
4625  JsonbIterator **it_array = rk1 == WJB_BEGIN_ARRAY ? it1 : it2;
4626  JsonbIterator **it_object = rk1 == WJB_BEGIN_OBJECT ? it1 : it2;
4627 
4628  bool prepend = (rk1 == WJB_BEGIN_OBJECT);
4629 
4630  pushJsonbValue(state, WJB_BEGIN_ARRAY, NULL);
4631 
4632  if (prepend)
4633  {
4634  pushJsonbValue(state, WJB_BEGIN_OBJECT, NULL);
4635  while ((r1 = JsonbIteratorNext(it_object, &v1, true)) != WJB_DONE)
4636  pushJsonbValue(state, r1, r1 != WJB_END_OBJECT ? &v1 : NULL);
4637 
4638  while ((r2 = JsonbIteratorNext(it_array, &v2, true)) != WJB_DONE)
4639  res = pushJsonbValue(state, r2, r2 != WJB_END_ARRAY ? &v2 : NULL);
4640  }
4641  else
4642  {
4643  while ((r1 = JsonbIteratorNext(it_array, &v1, true)) != WJB_END_ARRAY)
4644  pushJsonbValue(state, r1, &v1);
4645 
4646  pushJsonbValue(state, WJB_BEGIN_OBJECT, NULL);
4647  while ((r2 = JsonbIteratorNext(it_object, &v2, true)) != WJB_DONE)
4648  pushJsonbValue(state, r2, r2 != WJB_END_OBJECT ? &v2 : NULL);
4649 
4650  res = pushJsonbValue(state, WJB_END_ARRAY, NULL);
4651  }
4652  }
4653  else
4654  {
4655  /*
4656  * This must be scalar || object or object || scalar, as that's all
4657  * that's left. Both of these make no sense, so error out.
4658  */
4659  ereport(ERROR,
4660  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4661  errmsg("invalid concatenation of jsonb objects")));
4662  }
4663 
4664  return res;
4665 }
4666 
4667 /*
4668  * Do most of the heavy work for jsonb_set/jsonb_insert
4669  *
4670  * If JB_PATH_DELETE bit is set in op_type, the element is to be removed.
4671  *
4672  * If any bit mentioned in JB_PATH_CREATE_OR_INSERT is set in op_type,
4673  * we create the new value if the key or array index does not exist.
4674  *
4675  * Bits JB_PATH_INSERT_BEFORE and JB_PATH_INSERT_AFTER in op_type
4676  * behave as JB_PATH_CREATE if new value is inserted in JsonbObject.
4677  *
4678  * All path elements before the last must already exist
4679  * whatever bits in op_type are set, or nothing is done.
4680  */
4681 static JsonbValue *
4682 setPath(JsonbIterator **it, Datum *path_elems,
4683  bool *path_nulls, int path_len,
4684  JsonbParseState **st, int level, Jsonb *newval, int op_type)
4685 {
4686  JsonbValue v;
4687  JsonbIteratorToken r;
4688  JsonbValue *res;
4689 
4690  check_stack_depth();
4691 
4692  if (path_nulls[level])
4693  ereport(ERROR,
4694  (errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED),
4695  errmsg("path element at position %d is null",
4696  level + 1)));
4697 
4698  r = JsonbIteratorNext(it, &v, false);
4699 
4700  switch (r)
4701  {
4702  case WJB_BEGIN_ARRAY:
4703  (void) pushJsonbValue(st, r, NULL);
4704  setPathArray(it, path_elems, path_nulls, path_len, st, level,
4705  newval, v.val.array.nElems, op_type);
4706  r = JsonbIteratorNext(it, &v, false);
4707  Assert(r == WJB_END_ARRAY);
4708  res = pushJsonbValue(st, r, NULL);
4709  break;
4710  case WJB_BEGIN_OBJECT:
4711  (void) pushJsonbValue(st, r, NULL);
4712  setPathObject(it, path_elems, path_nulls, path_len, st, level,
4713  newval, v.val.object.nPairs, op_type);
4714  r = JsonbIteratorNext(it, &v, true);
4715  Assert(r == WJB_END_OBJECT);
4716  res = pushJsonbValue(st, r, NULL);
4717  break;
4718  case WJB_ELEM:
4719  case WJB_VALUE:
4720  res = pushJsonbValue(st, r, &v);
4721  break;
4722  default:
4723  elog(ERROR, "unrecognized iterator result: %d", (int) r);
4724  res = NULL; /* keep compiler quiet */
4725  break;
4726  }
4727 
4728  return res;
4729 }
4730 
4731 /*
4732  * Object walker for setPath
4733  */
4734 static void
4735 setPathObject(JsonbIterator **it, Datum *path_elems, bool *path_nulls,
4736  int path_len, JsonbParseState **st, int level,
4737  Jsonb *newval, uint32 npairs, int op_type)
4738 {
4739  JsonbValue v;
4740  int i;
4741  JsonbValue k;
4742  bool done = false;
4743 
4744  if (level >= path_len || path_nulls[level])
4745  done = true;
4746 
4747  /* empty object is a special case for create */
4748  if ((npairs == 0) && (op_type & JB_PATH_CREATE_OR_INSERT) &&
4749  (level == path_len - 1))
4750  {
4751  JsonbValue newkey;
4752 
4753  newkey.type = jbvString;
4754  newkey.val.string.len = VARSIZE_ANY_EXHDR(path_elems[level]);
4755  newkey.val.string.val = VARDATA_ANY(path_elems[level]);
4756 
4757  (void) pushJsonbValue(st, WJB_KEY, &newkey);
4758  addJsonbToParseState(st, newval);
4759  }
4760 
4761  for (i = 0; i < npairs; i++)
4762  {
4763  JsonbIteratorToken r = JsonbIteratorNext(it, &k, true);
4764 
4765  Assert(r == WJB_KEY);
4766 
4767  if (!done &&
4768  k.val.string.len == VARSIZE_ANY_EXHDR(path_elems[level]) &&
4769  memcmp(k.val.string.val, VARDATA_ANY(path_elems[level]),
4770  k.val.string.len) == 0)
4771  {
4772  if (level == path_len - 1)
4773  {
4774  /*
4775  * called from jsonb_insert(), it forbids redefining an
4776  * existing value
4777  */
4778  if (op_type & (JB_PATH_INSERT_BEFORE | JB_PATH_INSERT_AFTER))
4779  ereport(ERROR,
4780  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4781  errmsg("cannot replace existing key"),
4782  errhint("Try using the function jsonb_set "
4783  "to replace key value.")));
4784 
4785  r = JsonbIteratorNext(it, &v, true); /* skip value */
4786  if (!(op_type & JB_PATH_DELETE))
4787  {
4788  (void) pushJsonbValue(st, WJB_KEY, &k);
4789  addJsonbToParseState(st, newval);
4790  }
4791  done = true;
4792  }
4793  else
4794  {
4795  (void) pushJsonbValue(st, r, &k);
4796  setPath(it, path_elems, path_nulls, path_len,
4797  st, level + 1, newval, op_type);
4798  }
4799  }
4800  else
4801  {
4802  if ((op_type & JB_PATH_CREATE_OR_INSERT) && !done &&
4803  level == path_len - 1 && i == npairs - 1)
4804  {
4805  JsonbValue newkey;
4806 
4807  newkey.type = jbvString;
4808  newkey.val.string.len = VARSIZE_ANY_EXHDR(path_elems[level]);
4809  newkey.val.string.val = VARDATA_ANY(path_elems[level]);
4810 
4811  (void) pushJsonbValue(st, WJB_KEY, &newkey);
4812  addJsonbToParseState(st, newval);
4813  }
4814 
4815  (void) pushJsonbValue(st, r, &k);
4816  r = JsonbIteratorNext(it, &v, false);
4817  (void) pushJsonbValue(st, r, r < WJB_BEGIN_ARRAY ? &v : NULL);
4818  if (r == WJB_BEGIN_ARRAY || r == WJB_BEGIN_OBJECT)
4819  {
4820  int walking_level = 1;
4821 
4822  while (walking_level != 0)
4823  {
4824  r = JsonbIteratorNext(it, &v, false);
4825 
4826  if (r == WJB_BEGIN_ARRAY || r == WJB_BEGIN_OBJECT)
4827  ++walking_level;
4828  if (r == WJB_END_ARRAY || r == WJB_END_OBJECT)
4829  --walking_level;
4830 
4831  (void) pushJsonbValue(st, r, r < WJB_BEGIN_ARRAY ? &v : NULL);
4832  }
4833  }
4834  }
4835  }
4836 }
4837 
4838 /*
4839  * Array walker for setPath
4840  */
4841 static void
4842 setPathArray(JsonbIterator **it, Datum *path_elems, bool *path_nulls,
4843  int path_len, JsonbParseState **st, int level,
4844  Jsonb *newval, uint32 nelems, int op_type)
4845 {
4846  JsonbValue v;
4847  int idx,
4848  i;
4849  bool done = false;
4850 
4851  /* pick correct index */
4852  if (level < path_len && !path_nulls[level])
4853  {
4854  char *c = TextDatumGetCString(path_elems[level]);
4855  long lindex;
4856  char *badp;
4857 
4858  errno = 0;
4859  lindex = strtol(c, &badp, 10);
4860  if (errno != 0 || badp == c || *badp != '\0' || lindex > INT_MAX ||
4861  lindex < INT_MIN)
4862  ereport(ERROR,
4863  (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
4864  errmsg("path element at position %d is not an integer: \"%s\"",
4865  level + 1, c)));
4866  idx = lindex;
4867  }
4868  else
4869  idx = nelems;
4870 
4871  if (idx < 0)
4872  {
4873  if (-idx > nelems)
4874  idx = INT_MIN;
4875  else
4876  idx = nelems + idx;
4877  }
4878 
4879  if (idx > 0 && idx > nelems)
4880  idx = nelems;
4881 
4882  /*
4883  * if we're creating, and idx == INT_MIN, we prepend the new value to the
4884  * array also if the array is empty - in which case we don't really care
4885  * what the idx value is
4886  */
4887 
4888  if ((idx == INT_MIN || nelems == 0) && (level == path_len - 1) &&
4889  (op_type & JB_PATH_CREATE_OR_INSERT))
4890  {
4891  Assert(newval != NULL);
4892  addJsonbToParseState(st, newval);
4893  done = true;
4894  }
4895 
4896  /* iterate over the array elements */
4897  for (i = 0; i < nelems; i++)
4898  {
4899  JsonbIteratorToken r;
4900 
4901  if (i == idx && level < path_len)
4902  {
4903  if (level == path_len - 1)
4904  {
4905  r = JsonbIteratorNext(it, &v, true); /* skip */
4906 
4907  if (op_type & (JB_PATH_INSERT_BEFORE | JB_PATH_CREATE))
4908  addJsonbToParseState(st, newval);
4909 
4910  /*
4911  * We should keep current value only in case of
4912  * JB_PATH_INSERT_BEFORE or JB_PATH_INSERT_AFTER because
4913  * otherwise it should be deleted or replaced
4914  */
4915  if (op_type & (JB_PATH_INSERT_AFTER | JB_PATH_INSERT_BEFORE))
4916  (void) pushJsonbValue(st, r, &v);
4917 
4918  if (op_type & (JB_PATH_INSERT_AFTER | JB_PATH_REPLACE))
4919  addJsonbToParseState(st, newval);
4920 
4921  done = true;
4922  }
4923  else
4924  (void) setPath(it, path_elems, path_nulls, path_len,
4925  st, level + 1, newval, op_type);
4926  }
4927  else
4928  {
4929  r = JsonbIteratorNext(it, &v, false);
4930 
4931  (void) pushJsonbValue(st, r, r < WJB_BEGIN_ARRAY ? &v : NULL);
4932 
4933  if (r == WJB_BEGIN_ARRAY || r == WJB_BEGIN_OBJECT)
4934  {
4935  int walking_level = 1;
4936 
4937  while (walking_level != 0)
4938  {
4939  r = JsonbIteratorNext(it, &v, false);
4940 
4941  if (r == WJB_BEGIN_ARRAY || r == WJB_BEGIN_OBJECT)
4942  ++walking_level;
4943  if (r == WJB_END_ARRAY || r == WJB_END_OBJECT)
4944  --walking_level;
4945 
4946  (void) pushJsonbValue(st, r, r < WJB_BEGIN_ARRAY ? &v : NULL);
4947  }
4948  }
4949 
4950  if ((op_type & JB_PATH_CREATE_OR_INSERT) && !done &&
4951  level == path_len - 1 && i == nelems - 1)
4952  {
4953  addJsonbToParseState(st, newval);
4954  }
4955  }
4956  }
4957 }
4958 
4959 /*
4960  * Parse information about what elements of a jsonb document we want to iterate
4961  * in functions iterate_json(b)_values. This information is presented in jsonb
4962  * format, so that it can be easily extended in the future.
4963  */
4964 uint32
4965 parse_jsonb_index_flags(Jsonb *jb)
4966 {
4967  JsonbIterator *it;
4968  JsonbValue v;
4969  JsonbIteratorToken type;
4970  uint32 flags = 0;
4971 
4972  it = JsonbIteratorInit(&jb->root);
4973 
4974  type = JsonbIteratorNext(&it, &v, false);
4975 
4976  /*
4977  * We iterate over array (scalar internally is represented as array, so,
4978  * we will accept it too) to check all its elements. Flag names are
4979  * chosen the same as jsonb_typeof uses.
4980  */
4981  if (type != WJB_BEGIN_ARRAY)
4982  ereport(ERROR, (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4983  errmsg("wrong flag type, only arrays and scalars are allowed")));
4984 
4985  while ((type = JsonbIteratorNext(&it, &v, false)) == WJB_ELEM)
4986  {
4987  if (v.type != jbvString)
4988  ereport(ERROR,
4989  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4990  errmsg("flag array element is not a string"),
4991  errhint("Possible values are: \"string\", \"numeric\", \"boolean\", \"key\", and \"all\".")));
4992 
4993  if (v.val.string.len == 3 &&
4994  pg_strncasecmp(v.val.string.val, "all", 3) == 0)
4995  flags |= jtiAll;
4996  else if (v.val.string.len == 3 &&
4997  pg_strncasecmp(v.val.string.val, "key", 3) == 0)
4998  flags |= jtiKey;
4999  else if (v.val.string.len == 6 &&
5000  pg_strncasecmp(v.val.string.val, "string", 5) == 0)
5001  flags |= jtiString;
5002  else if (v.val.string.len == 7 &&
5003  pg_strncasecmp(v.val.string.val, "numeric", 7) == 0)
5004  flags |= jtiNumeric;
5005  else if (v.val.string.len == 7 &&
5006  pg_strncasecmp(v.val.string.val, "boolean", 7) == 0)
5007  flags |= jtiBool;
5008  else
5009  ereport(ERROR,
5010  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5011  errmsg("wrong flag in flag array: \"%s\"",
5012  pnstrdup(v.val.string.val, v.val.string.len)),
5013  errhint("Possible values are: \"string\", \"numeric\", \"boolean\", \"key\", and \"all\".")));
5014  }
5015 
5016  /* expect end of array now */
5017  if (type != WJB_END_ARRAY)
5018  elog(ERROR, "unexpected end of flag array");
5019 
5020  /* get final WJB_DONE and free iterator */
5021  type = JsonbIteratorNext(&it, &v, false);
5022  if (type != WJB_DONE)
5023  elog(ERROR, "unexpected end of flag array");
5024 
5025  return flags;
5026 }
5027 
5028 /*
5029  * Iterate over jsonb values or elements, specified by flags, and pass them
5030  * together with an iteration state to a specified JsonIterateStringValuesAction.
5031  */
5032 void
5033 iterate_jsonb_values(Jsonb *jb, uint32 flags, void *state,
5034  JsonIterateStringValuesAction action)
5035 {
5036  JsonbIterator *it;
5037  JsonbValue v;
5038  JsonbIteratorToken type;
5039 
5040  it = JsonbIteratorInit(&jb->root);
5041 
5042  /*
5043  * Just recursively iterating over jsonb and call callback on all
5044  * corresponding elements
5045  */
5046  while ((type = JsonbIteratorNext(&it, &v, false)) != WJB_DONE)
5047  {
5048  if (type == WJB_KEY)
5049  {
5050  if (flags & jtiKey)
5051  action(state, v.val.string.val, v.val.string.len);
5052 
5053  continue;
5054  }
5055  else if (!(type == WJB_VALUE || type == WJB_ELEM))
5056  {
5057  /* do not call callback for composite JsonbValue */
5058  continue;
5059  }
5060 
5061  /* JsonbValue is a value of object or element of array */
5062  switch (v.type)
5063  {
5064  case jbvString:
5065  if (flags & jtiString)
5066  action(state, v.val.string.val, v.val.string.len);
5067  break;
5068  case jbvNumeric:
5069  if (flags & jtiNumeric)
5070  {
5071  char *val;
5072 
5073  val = DatumGetCString(DirectFunctionCall1(numeric_out,
5074  NumericGetDatum(v.val.numeric)));
5075 
5076  action(state, val, strlen(val));
5077  pfree(val);
5078  }
5079  break;
5080  case jbvBool:
5081  if (flags & jtiBool)
5082  {
5083  if (v.val.boolean)
5084  action(state, "true", 4);
5085  else
5086  action(state, "false", 5);
5087  }
5088  break;
5089  default:
5090  /* do not call callback for composite JsonbValue */
5091  break;
5092  }
5093  }
5094 }
5095 
5096 /*
5097  * Iterate over json values and elements, specified by flags, and pass them
5098  * together with an iteration state to a specified JsonIterateStringValuesAction.
5099  */
5100 void
5101 iterate_json_values(text *json, uint32 flags, void *action_state,
5102  JsonIterateStringValuesAction action)
5103 {
5104  JsonLexContext *lex = makeJsonLexContext(json, true);
5105  JsonSemAction *sem = palloc0(sizeof(JsonSemAction));
5106  IterateJsonStringValuesState *state = palloc0(sizeof(IterateJsonStringValuesState));
5107 
5108  state->lex = lex;
5109  state->action = action;
5110  state->action_state = action_state;
5111  state->flags = flags;
5112 
5113  sem->semstate = (void *) state;
5114  sem->scalar = iterate_values_scalar;
5115  sem->object_field_start = iterate_values_object_field_start;
5116 
5117  pg_parse_json(lex, sem);
5118 }
5119 
5120 /*
5121  * An auxiliary function for iterate_json_values to invoke a specified
5122  * JsonIterateStringValuesAction for specified values.
5123  */
5124 static void
5125 iterate_values_scalar(void *state, char *token, JsonTokenType tokentype)
5126 {
5127  IterateJsonStringValuesState *_state = (IterateJsonStringValuesState *) state;
5128 
5129  switch (tokentype)
5130  {
5131  case JSON_TOKEN_STRING:
5132  if (_state->flags & jtiString)
5133  _state->action(_state->action_state, token, strlen(token));
5134  break;
5135  case JSON_TOKEN_NUMBER:
5136  if (_state->flags & jtiNumeric)
5137  _state->action(_state->action_state, token, strlen(token));
5138  break;
5139  case JSON_TOKEN_TRUE:
5140  case JSON_TOKEN_FALSE:
5141  if (_state->flags & jtiBool)
5142  _state->action(_state->action_state, token, strlen(token));
5143  break;
5144  default:
5145  /* do not call callback for any other token */
5146  break;
5147  }
5148 }
5149 
5150 static void
5151 iterate_values_object_field_start(void *state, char *fname, bool isnull)
5152 {
5153  IterateJsonStringValuesState *_state = (IterateJsonStringValuesState *) state;
5154 
5155  if (_state->flags & jtiKey)
5156  {
5157  char *val = pstrdup(fname);
5158 
5159  _state->action(_state->action_state, val, strlen(val));
5160  }
5161 }
5162 
5163 /*
5164  * Iterate over a jsonb, and apply a specified JsonTransformStringValuesAction
5165  * to every string value or element. Any necessary context for a
5166  * JsonTransformStringValuesAction can be passed in the action_state variable.
5167  * Function returns a copy of an original jsonb object with transformed values.
5168  */
5169 Jsonb *
5170 transform_jsonb_string_values(Jsonb *jsonb, void *action_state,
5171  JsonTransformStringValuesAction transform_action)
5172 {
5173  JsonbIterator *it;
5174  JsonbValue v,
5175  *res = NULL;
5176  JsonbIteratorToken type;
5177  JsonbParseState *st = NULL;
5178  text *out;
5179  bool is_scalar = false;
5180 
5181  it = JsonbIteratorInit(&jsonb->root);
5182  is_scalar = it->isScalar;
5183 
5184  while ((type = JsonbIteratorNext(&it, &v, false)) != WJB_DONE)
5185  {
5186  if ((type == WJB_VALUE || type == WJB_ELEM) && v.type == jbvString)
5187  {
5188  out = transform_action(action_state, v.val.string.val, v.val.string.len);
5189  v.val.string.val = VARDATA_ANY(out);
5190  v.val.string.len = VARSIZE_ANY_EXHDR(out);
5191  res = pushJsonbValue(&st, type, type < WJB_BEGIN_ARRAY ? &v : NULL);
5192  }
5193  else
5194  {
5195  res = pushJsonbValue(&st, type, (type == WJB_KEY ||
5196  type == WJB_VALUE ||
5197  type == WJB_ELEM) ? &v : NULL);
5198  }
5199  }
5200 
5201  if (res->type == jbvArray)
5202  res->val.array.rawScalar = is_scalar;
5203 
5204  return JsonbValueToJsonb(res);
5205 }
5206 
5207 /*
5208  * Iterate over a json, and apply a specified JsonTransformStringValuesAction
5209  * to every string value or element. Any necessary context for a
5210  * JsonTransformStringValuesAction can be passed in the action_state variable.
5211  * Function returns a StringInfo, which is a copy of an original json with
5212  * transformed values.
5213  */
5214 text *
5215 transform_json_string_values(text *json, void *action_state,
5216  JsonTransformStringValuesAction transform_action)
5217 {
5218  JsonLexContext *lex = makeJsonLexContext(json, true);
5219  JsonSemAction *sem = palloc0(sizeof(JsonSemAction));
5220  TransformJsonStringValuesState *state = palloc0(sizeof(TransformJsonStringValuesState));
5221 
5222  state->lex = lex;
5223  state->strval = makeStringInfo();
5224  state->action = transform_action;
5225  state->action_state = action_state;
5226 
5227  sem->semstate = (void *) state;
5228  sem->scalar = transform_string_values_scalar;
5229  sem->object_start = transform_string_values_object_start;
5230  sem->object_end = transform_string_values_object_end;
5231  sem->array_start = transform_string_values_array_start;
5232  sem->array_end = transform_string_values_array_end;
5233  sem->scalar = transform_string_values_scalar;
5234  sem->array_element_start = transform_string_values_array_element_start;
5235  sem->object_field_start = transform_string_values_object_field_start;
5236 
5237  pg_parse_json(lex, sem);
5238 
5239  return cstring_to_text_with_len(state->strval->data, state->strval->len);
5240 }
5241 
5242 /*
5243  * Set of auxiliary functions for transform_json_string_values to invoke a
5244  * specified JsonTransformStringValuesAction for all values and left everything
5245  * else untouched.
5246  */
5247 static void
5248 transform_string_values_object_start(void *state)
5249 {
5250  TransformJsonStringValuesState *_state = (TransformJsonStringValuesState *) state;
5251 
5252  appendStringInfoCharMacro(_state->strval, '{');
5253 }
5254 
5255 static void
5256 transform_string_values_object_end(void *state)
5257 {
5258  TransformJsonStringValuesState *_state = (TransformJsonStringValuesState *) state;
5259 
5260  appendStringInfoCharMacro(_state->strval, '}');
5261 }
5262 
5263 static void
5264 transform_string_values_array_start(void *state)
5265 {
5266  TransformJsonStringValuesState *_state = (TransformJsonStringValuesState *) state;
5267 
5268  appendStringInfoCharMacro(_state->strval, '[');
5269 }
5270 
5271 static void
5272 transform_string_values_array_end(void *state)
5273 {
5274  TransformJsonStringValuesState *_state = (TransformJsonStringValuesState *) state;
5275 
5276  appendStringInfoCharMacro(_state->strval, ']');
5277 }
5278 
5279 static void
5280 transform_string_values_object_field_start(void *state, char *fname, bool isnull)
5281 {
5282  TransformJsonStringValuesState *_state = (TransformJsonStringValuesState *) state;
5283 
5284  if (_state->strval->data[_state->strval->len - 1] != '{')
5285  appendStringInfoCharMacro(_state->strval, ',');
5286 
5287  /*
5288  * Unfortunately we don't have the quoted and escaped string any more, so
5289  * we have to re-escape it.
5290  */
5291  escape_json(_state->strval, fname);
5292  appendStringInfoCharMacro(_state->strval, ':');
5293 }
5294 
5295 static void
5296 transform_string_values_array_element_start(void *state, bool isnull)
5297 {
5298  TransformJsonStringValuesState *_state = (TransformJsonStringValuesState *) state;
5299 
5300  if (_state->strval->data[_state->strval->len - 1] != '[')
5301  appendStringInfoCharMacro(_state->strval, ',');
5302 }
5303 
5304 static void
5305 transform_string_values_scalar(void *state, char *token, JsonTokenType tokentype)
5306 {
5307  TransformJsonStringValuesState *_state = (TransformJsonStringValuesState *) state;
5308 
5309  if (tokentype == JSON_TOKEN_STRING)
5310  {
5311  text *out = _state->action(_state->action_state, token, strlen(token));
5312 
5313  escape_json(_state->strval, text_to_cstring(out));
5314  }
5315  else
5316  appendStringInfoString(_state->strval, token);
5317 }
JsonSemAction::array_end
json_struct_action array_end
Definition: jsonapi.h:88
get_array_end
static void get_array_end(void *state)
Definition: jsonfuncs.c:1231
elements_worker_jsonb
static Datum elements_worker_jsonb(FunctionCallInfo fcinfo, const char *funcname, bool as_text)
Definition: jsonfuncs.c:1989
PG_RETURN_POINTER
#define PG_RETURN_POINTER(x)
Definition: fmgr.h:351
makeMdArrayResult
Datum makeMdArrayResult(ArrayBuildState *astate, int ndims, int *dims, int *lbs, MemoryContext rcontext, bool release)
Definition: arrayfuncs.c:5149
populate_recordset_object_start
static void populate_recordset_object_start(void *state)
Definition: jsonfuncs.c:3749
jsonb_array_elements
Datum jsonb_array_elements(PG_FUNCTION_ARGS)
Definition: jsonfuncs.c:1977
OkeysState::result_count
int result_count
Definition: jsonfuncs.c:52
ElementsState::normalize_results
bool normalize_results
Definition: jsonfuncs.c:122
PG_GETARG_INT32
#define PG_GETARG_INT32(n)
Definition: fmgr.h:264
FmgrInfo
Definition: fmgr.h:56
jsonb_array_length
Datum jsonb_array_length(PG_FUNCTION_ARGS)
Definition: jsonfuncs.c:1584
StripnullState::skip_next_null
bool skip_next_null
Definition: jsonfuncs.c:286
CompositeIOData::record_io
RecordIOData * record_io
Definition: jsonfuncs.c:172
hash_destroy
void hash_destroy(HTAB *hashp)
Definition: dynahash.c:814
getBaseTypeAndTypmod
Oid getBaseTypeAndTypmod(Oid typid, int32 *typmod)
Definition: lsyscache.c:2316
ArrayIOData::element_info
ColumnIOData * element_info
Definition: jsonfuncs.c:160
CreateTupleDescCopy
TupleDesc CreateTupleDescCopy(TupleDesc tupdesc)
Definition: tupdesc.c:110
IterateJsonStringValuesState::action
JsonIterateStringValuesAction action
Definition: jsonfuncs.c:60
IsA
#define IsA(nodeptr, _type_)
Definition: nodes.h:575
EachState::tuple_store
Tuplestorestate * tuple_store
Definition: jsonfuncs.c:104
MemoryContextDelete
void MemoryContextDelete(MemoryContext context)
Definition: mcxt.c:211
AllocSetContextCreate
#define AllocSetContextCreate
Definition: memutils.h:169
JB_ROOT_IS_OBJECT
#define JB_ROOT_IS_OBJECT(jbp_)
Definition: jsonb.h:229
each_object_field_start
static void each_object_field_start(void *state, char *fname, bool isnull)
Definition: jsonfuncs.c:1875
ElementsState::function_name
const char * function_name
Definition: jsonfuncs.c:117
get_call_result_type
TypeFuncClass get_call_result_type(FunctionCallInfo fcinfo, Oid *resultTypeId, TupleDesc *resultTupleDesc)
Definition: funcapi.c:196
setPathArray
static void setPathArray(JsonbIterator **it, Datum *path_elems, bool *path_nulls, int path_len, JsonbParseState **st, int level, Jsonb *newval, uint32 nelems, int op_type)
Definition: jsonfuncs.c:4842
errhint
int errhint(const char *fmt,...)
Definition: elog.c:974
pnstrdup
char * pnstrdup(const char *in, Size len)
Definition: mcxt.c:1172
GetState::array_cur_index
int * array_cur_index