granicus.if.org Git - postgresql/blob - src/backend/utils/adt/json.c

   1 /*-------------------------------------------------------------------------
   2  *
   3  * json.c
   4  *              JSON data type support.
   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/json.c
  11  *
  12  *-------------------------------------------------------------------------
  13  */
  14 #include "postgres.h"
  15
  16 #include "access/htup_details.h"
  17 #include "access/transam.h"
  18 #include "catalog/pg_type.h"
  19 #include "executor/spi.h"
  20 #include "funcapi.h"
  21 #include "lib/stringinfo.h"
  22 #include "libpq/pqformat.h"
  23 #include "mb/pg_wchar.h"
  24 #include "miscadmin.h"
  25 #include "parser/parse_coerce.h"
  26 #include "utils/array.h"
  27 #include "utils/builtins.h"
  28 #include "utils/date.h"
  29 #include "utils/datetime.h"
  30 #include "utils/lsyscache.h"
  31 #include "utils/json.h"
  32 #include "utils/jsonapi.h"
  33 #include "utils/typcache.h"
  34 #include "utils/syscache.h"
  35
  36 /*
  37  * The context of the parser is maintained by the recursive descent
  38  * mechanism, but is passed explicitly to the error reporting routine
  39  * for better diagnostics.
  40  */
  41 typedef enum                                    /* contexts of JSON parser */
  42 {
  43         JSON_PARSE_VALUE,                       /* expecting a value */
  44         JSON_PARSE_STRING,                      /* expecting a string (for a field name) */
  45         JSON_PARSE_ARRAY_START,         /* saw '[', expecting value or ']' */
  46         JSON_PARSE_ARRAY_NEXT,          /* saw array element, expecting ',' or ']' */
  47         JSON_PARSE_OBJECT_START,        /* saw '{', expecting label or '}' */
  48         JSON_PARSE_OBJECT_LABEL,        /* saw object label, expecting ':' */
  49         JSON_PARSE_OBJECT_NEXT,         /* saw object value, expecting ',' or '}' */
  50         JSON_PARSE_OBJECT_COMMA,        /* saw object ',', expecting next label */
  51         JSON_PARSE_END                          /* saw the end of a document, expect nothing */
  52 } JsonParseContext;
  53
  54 typedef enum                                    /* type categories for datum_to_json */
  55 {
  56         JSONTYPE_NULL,                          /* null, so we didn't bother to identify */
  57         JSONTYPE_BOOL,                          /* boolean (built-in types only) */
  58         JSONTYPE_NUMERIC,                       /* numeric (ditto) */
  59         JSONTYPE_DATE,                          /* we use special formatting for datetimes */
  60         JSONTYPE_TIMESTAMP,
  61         JSONTYPE_TIMESTAMPTZ,
  62         JSONTYPE_JSON,                          /* JSON itself (and JSONB) */
  63         JSONTYPE_ARRAY,                         /* array */
  64         JSONTYPE_COMPOSITE,                     /* composite */
  65         JSONTYPE_CAST,                          /* something with an explicit cast to JSON */
  66         JSONTYPE_OTHER                          /* all else */
  67 } JsonTypeCategory;
  68
  69 typedef struct JsonAggState
  70 {
  71         StringInfo      str;
  72         JsonTypeCategory key_category;
  73         Oid                     key_output_func;
  74         JsonTypeCategory val_category;
  75         Oid                     val_output_func;
  76 } JsonAggState;
  77
  78 static inline void json_lex(JsonLexContext *lex);
  79 static inline void json_lex_string(JsonLexContext *lex);
  80 static inline void json_lex_number(JsonLexContext *lex, char *s,
  81                                 bool *num_err, int *total_len);
  82 static inline void parse_scalar(JsonLexContext *lex, JsonSemAction *sem);
  83 static void parse_object_field(JsonLexContext *lex, JsonSemAction *sem);
  84 static void parse_object(JsonLexContext *lex, JsonSemAction *sem);
  85 static void parse_array_element(JsonLexContext *lex, JsonSemAction *sem);
  86 static void parse_array(JsonLexContext *lex, JsonSemAction *sem);
  87 static void report_parse_error(JsonParseContext ctx, JsonLexContext *lex) pg_attribute_noreturn();
  88 static void report_invalid_token(JsonLexContext *lex) pg_attribute_noreturn();
  89 static int      report_json_context(JsonLexContext *lex);
  90 static char *extract_mb_char(char *s);
  91 static void composite_to_json(Datum composite, StringInfo result,
  92                                   bool use_line_feeds);
  93 static void array_dim_to_json(StringInfo result, int dim, int ndims, int *dims,
  94                                   Datum *vals, bool *nulls, int *valcount,
  95                                   JsonTypeCategory tcategory, Oid outfuncoid,
  96                                   bool use_line_feeds);
  97 static void array_to_json_internal(Datum array, StringInfo result,
  98                                            bool use_line_feeds);
  99 static void json_categorize_type(Oid typoid,
 100                                          JsonTypeCategory *tcategory,
 101                                          Oid *outfuncoid);
 102 static void datum_to_json(Datum val, bool is_null, StringInfo result,
 103                           JsonTypeCategory tcategory, Oid outfuncoid,
 104                           bool key_scalar);
 105 static void add_json(Datum val, bool is_null, StringInfo result,
 106                  Oid val_type, bool key_scalar);
 107 static text *catenate_stringinfo_string(StringInfo buffer, const char *addon);
 108
 109 /* the null action object used for pure validation */
 110 static JsonSemAction nullSemAction =
 111 {
 112         NULL, NULL, NULL, NULL, NULL,
 113         NULL, NULL, NULL, NULL, NULL
 114 };
 115
 116 /* Recursive Descent parser support routines */
 117
 118 /*
 119  * lex_peek
 120  *
 121  * what is the current look_ahead token?
 122 */
 123 static inline JsonTokenType
 124 lex_peek(JsonLexContext *lex)
 125 {
 126         return lex->token_type;
 127 }
 128
 129 /*
 130  * lex_accept
 131  *
 132  * accept the look_ahead token and move the lexer to the next token if the
 133  * look_ahead token matches the token parameter. In that case, and if required,
 134  * also hand back the de-escaped lexeme.
 135  *
 136  * returns true if the token matched, false otherwise.
 137  */
 138 static inline bool
 139 lex_accept(JsonLexContext *lex, JsonTokenType token, char **lexeme)
 140 {
 141         if (lex->token_type == token)
 142         {
 143                 if (lexeme != NULL)
 144                 {
 145                         if (lex->token_type == JSON_TOKEN_STRING)
 146                         {
 147                                 if (lex->strval != NULL)
 148                                         *lexeme = pstrdup(lex->strval->data);
 149                         }
 150                         else
 151                         {
 152                                 int                     len = (lex->token_terminator - lex->token_start);
 153                                 char       *tokstr = palloc(len + 1);
 154
 155                                 memcpy(tokstr, lex->token_start, len);
 156                                 tokstr[len] = '\0';
 157                                 *lexeme = tokstr;
 158                         }
 159                 }
 160                 json_lex(lex);
 161                 return true;
 162         }
 163         return false;
 164 }
 165
 166 /*
 167  * lex_accept
 168  *
 169  * move the lexer to the next token if the current look_ahead token matches
 170  * the parameter token. Otherwise, report an error.
 171  */
 172 static inline void
 173 lex_expect(JsonParseContext ctx, JsonLexContext *lex, JsonTokenType token)
 174 {
 175         if (!lex_accept(lex, token, NULL))
 176                 report_parse_error(ctx, lex);
 177 }
 178
 179 /* chars to consider as part of an alphanumeric token */
 180 #define JSON_ALPHANUMERIC_CHAR(c)  \
 181         (((c) >= 'a' && (c) <= 'z') || \
 182          ((c) >= 'A' && (c) <= 'Z') || \
 183          ((c) >= '0' && (c) <= '9') || \
 184          (c) == '_' || \
 185          IS_HIGHBIT_SET(c))
 186
 187 /*
 188  * Utility function to check if a string is a valid JSON number.
 189  *
 190  * str is of length len, and need not be null-terminated.
 191  */
 192 bool
 193 IsValidJsonNumber(const char *str, int len)
 194 {
 195         bool            numeric_error;
 196         int                     total_len;
 197         JsonLexContext dummy_lex;
 198
 199         if (len <= 0)
 200                 return false;
 201
 202         /*
 203          * json_lex_number expects a leading  '-' to have been eaten already.
 204          *
 205          * having to cast away the constness of str is ugly, but there's not much
 206          * easy alternative.
 207          */
 208         if (*str == '-')
 209         {
 210                 dummy_lex.input = unconstify(char *, str) + 1;
 211                 dummy_lex.input_length = len - 1;
 212         }
 213         else
 214         {
 215                 dummy_lex.input = unconstify(char *, str);
 216                 dummy_lex.input_length = len;
 217         }
 218
 219         json_lex_number(&dummy_lex, dummy_lex.input, &numeric_error, &total_len);
 220
 221         return (!numeric_error) && (total_len == dummy_lex.input_length);
 222 }
 223
 224 /*
 225  * Input.
 226  */
 227 Datum
 228 json_in(PG_FUNCTION_ARGS)
 229 {
 230         char       *json = PG_GETARG_CSTRING(0);
 231         text       *result = cstring_to_text(json);
 232         JsonLexContext *lex;
 233
 234         /* validate it */
 235         lex = makeJsonLexContext(result, false);
 236         pg_parse_json(lex, &nullSemAction);
 237
 238         /* Internal representation is the same as text, for now */
 239         PG_RETURN_TEXT_P(result);
 240 }
 241
 242 /*
 243  * Output.
 244  */
 245 Datum
 246 json_out(PG_FUNCTION_ARGS)
 247 {
 248         /* we needn't detoast because text_to_cstring will handle that */
 249         Datum           txt = PG_GETARG_DATUM(0);
 250
 251         PG_RETURN_CSTRING(TextDatumGetCString(txt));
 252 }
 253
 254 /*
 255  * Binary send.
 256  */
 257 Datum
 258 json_send(PG_FUNCTION_ARGS)
 259 {
 260         text       *t = PG_GETARG_TEXT_PP(0);
 261         StringInfoData buf;
 262
 263         pq_begintypsend(&buf);
 264         pq_sendtext(&buf, VARDATA_ANY(t), VARSIZE_ANY_EXHDR(t));
 265         PG_RETURN_BYTEA_P(pq_endtypsend(&buf));
 266 }
 267
 268 /*
 269  * Binary receive.
 270  */
 271 Datum
 272 json_recv(PG_FUNCTION_ARGS)
 273 {
 274         StringInfo      buf = (StringInfo) PG_GETARG_POINTER(0);
 275         char       *str;
 276         int                     nbytes;
 277         JsonLexContext *lex;
 278
 279         str = pq_getmsgtext(buf, buf->len - buf->cursor, &nbytes);
 280
 281         /* Validate it. */
 282         lex = makeJsonLexContextCstringLen(str, nbytes, false);
 283         pg_parse_json(lex, &nullSemAction);
 284
 285         PG_RETURN_TEXT_P(cstring_to_text_with_len(str, nbytes));
 286 }
 287
 288 /*
 289  * makeJsonLexContext
 290  *
 291  * lex constructor, with or without StringInfo object
 292  * for de-escaped lexemes.
 293  *
 294  * Without is better as it makes the processing faster, so only make one
 295  * if really required.
 296  *
 297  * If you already have the json as a text* value, use the first of these
 298  * functions, otherwise use  makeJsonLexContextCstringLen().
 299  */
 300 JsonLexContext *
 301 makeJsonLexContext(text *json, bool need_escapes)
 302 {
 303         return makeJsonLexContextCstringLen(VARDATA_ANY(json),
 304                                                                                 VARSIZE_ANY_EXHDR(json),
 305                                                                                 need_escapes);
 306 }
 307
 308 JsonLexContext *
 309 makeJsonLexContextCstringLen(char *json, int len, bool need_escapes)
 310 {
 311         JsonLexContext *lex = palloc0(sizeof(JsonLexContext));
 312
 313         lex->input = lex->token_terminator = lex->line_start = json;
 314         lex->line_number = 1;
 315         lex->input_length = len;
 316         if (need_escapes)
 317                 lex->strval = makeStringInfo();
 318         return lex;
 319 }
 320
 321 /*
 322  * pg_parse_json
 323  *
 324  * Publicly visible entry point for the JSON parser.
 325  *
 326  * lex is a lexing context, set up for the json to be processed by calling
 327  * makeJsonLexContext(). sem is a structure of function pointers to semantic
 328  * action routines to be called at appropriate spots during parsing, and a
 329  * pointer to a state object to be passed to those routines.
 330  */
 331 void
 332 pg_parse_json(JsonLexContext *lex, JsonSemAction *sem)
 333 {
 334         JsonTokenType tok;
 335
 336         /* get the initial token */
 337         json_lex(lex);
 338
 339         tok = lex_peek(lex);
 340
 341         /* parse by recursive descent */
 342         switch (tok)
 343         {
 344                 case JSON_TOKEN_OBJECT_START:
 345                         parse_object(lex, sem);
 346                         break;
 347                 case JSON_TOKEN_ARRAY_START:
 348                         parse_array(lex, sem);
 349                         break;
 350                 default:
 351                         parse_scalar(lex, sem); /* json can be a bare scalar */
 352         }
 353
 354         lex_expect(JSON_PARSE_END, lex, JSON_TOKEN_END);
 355
 356 }
 357
 358 /*
 359  * json_count_array_elements
 360  *
 361  * Returns number of array elements in lex context at start of array token
 362  * until end of array token at same nesting level.
 363  *
 364  * Designed to be called from array_start routines.
 365  */
 366 int
 367 json_count_array_elements(JsonLexContext *lex)
 368 {
 369         JsonLexContext copylex;
 370         int                     count;
 371
 372         /*
 373          * It's safe to do this with a shallow copy because the lexical routines
 374          * don't scribble on the input. They do scribble on the other pointers
 375          * etc, so doing this with a copy makes that safe.
 376          */
 377         memcpy(&copylex, lex, sizeof(JsonLexContext));
 378         copylex.strval = NULL;          /* not interested in values here */
 379         copylex.lex_level++;
 380
 381         count = 0;
 382         lex_expect(JSON_PARSE_ARRAY_START, &copylex, JSON_TOKEN_ARRAY_START);
 383         if (lex_peek(&copylex) != JSON_TOKEN_ARRAY_END)
 384         {
 385                 do
 386                 {
 387                         count++;
 388                         parse_array_element(&copylex, &nullSemAction);
 389                 }
 390                 while (lex_accept(&copylex, JSON_TOKEN_COMMA, NULL));
 391         }
 392         lex_expect(JSON_PARSE_ARRAY_NEXT, &copylex, JSON_TOKEN_ARRAY_END);
 393
 394         return count;
 395 }
 396
 397 /*
 398  *      Recursive Descent parse routines. There is one for each structural
 399  *      element in a json document:
 400  *        - scalar (string, number, true, false, null)
 401  *        - array  ( [ ] )
 402  *        - array element
 403  *        - object ( { } )
 404  *        - object field
 405  */
 406 static inline void
 407 parse_scalar(JsonLexContext *lex, JsonSemAction *sem)
 408 {
 409         char       *val = NULL;
 410         json_scalar_action sfunc = sem->scalar;
 411         char      **valaddr;
 412         JsonTokenType tok = lex_peek(lex);
 413
 414         valaddr = sfunc == NULL ? NULL : &val;
 415
 416         /* a scalar must be a string, a number, true, false, or null */
 417         switch (tok)
 418         {
 419                 case JSON_TOKEN_TRUE:
 420                         lex_accept(lex, JSON_TOKEN_TRUE, valaddr);
 421                         break;
 422                 case JSON_TOKEN_FALSE:
 423                         lex_accept(lex, JSON_TOKEN_FALSE, valaddr);
 424                         break;
 425                 case JSON_TOKEN_NULL:
 426                         lex_accept(lex, JSON_TOKEN_NULL, valaddr);
 427                         break;
 428                 case JSON_TOKEN_NUMBER:
 429                         lex_accept(lex, JSON_TOKEN_NUMBER, valaddr);
 430                         break;
 431                 case JSON_TOKEN_STRING:
 432                         lex_accept(lex, JSON_TOKEN_STRING, valaddr);
 433                         break;
 434                 default:
 435                         report_parse_error(JSON_PARSE_VALUE, lex);
 436         }
 437
 438         if (sfunc != NULL)
 439                 (*sfunc) (sem->semstate, val, tok);
 440 }
 441
 442 static void
 443 parse_object_field(JsonLexContext *lex, JsonSemAction *sem)
 444 {
 445         /*
 446          * An object field is "fieldname" : value where value can be a scalar,
 447          * object or array.  Note: in user-facing docs and error messages, we
 448          * generally call a field name a "key".
 449          */
 450
 451         char       *fname = NULL;       /* keep compiler quiet */
 452         json_ofield_action ostart = sem->object_field_start;
 453         json_ofield_action oend = sem->object_field_end;
 454         bool            isnull;
 455         char      **fnameaddr = NULL;
 456         JsonTokenType tok;
 457
 458         if (ostart != NULL || oend != NULL)
 459                 fnameaddr = &fname;
 460
 461         if (!lex_accept(lex, JSON_TOKEN_STRING, fnameaddr))
 462                 report_parse_error(JSON_PARSE_STRING, lex);
 463
 464         lex_expect(JSON_PARSE_OBJECT_LABEL, lex, JSON_TOKEN_COLON);
 465
 466         tok = lex_peek(lex);
 467         isnull = tok == JSON_TOKEN_NULL;
 468
 469         if (ostart != NULL)
 470                 (*ostart) (sem->semstate, fname, isnull);
 471
 472         switch (tok)
 473         {
 474                 case JSON_TOKEN_OBJECT_START:
 475                         parse_object(lex, sem);
 476                         break;
 477                 case JSON_TOKEN_ARRAY_START:
 478                         parse_array(lex, sem);
 479                         break;
 480                 default:
 481                         parse_scalar(lex, sem);
 482         }
 483
 484         if (oend != NULL)
 485                 (*oend) (sem->semstate, fname, isnull);
 486 }
 487
 488 static void
 489 parse_object(JsonLexContext *lex, JsonSemAction *sem)
 490 {
 491         /*
 492          * an object is a possibly empty sequence of object fields, separated by
 493          * commas and surrounded by curly braces.
 494          */
 495         json_struct_action ostart = sem->object_start;
 496         json_struct_action oend = sem->object_end;
 497         JsonTokenType tok;
 498
 499         check_stack_depth();
 500
 501         if (ostart != NULL)
 502                 (*ostart) (sem->semstate);
 503
 504         /*
 505          * Data inside an object is at a higher nesting level than the object
 506          * itself. Note that we increment this after we call the semantic routine
 507          * for the object start and restore it before we call the routine for the
 508          * object end.
 509          */
 510         lex->lex_level++;
 511
 512         /* we know this will succeed, just clearing the token */
 513         lex_expect(JSON_PARSE_OBJECT_START, lex, JSON_TOKEN_OBJECT_START);
 514
 515         tok = lex_peek(lex);
 516         switch (tok)
 517         {
 518                 case JSON_TOKEN_STRING:
 519                         parse_object_field(lex, sem);
 520                         while (lex_accept(lex, JSON_TOKEN_COMMA, NULL))
 521                                 parse_object_field(lex, sem);
 522                         break;
 523                 case JSON_TOKEN_OBJECT_END:
 524                         break;
 525                 default:
 526                         /* case of an invalid initial token inside the object */
 527                         report_parse_error(JSON_PARSE_OBJECT_START, lex);
 528         }
 529
 530         lex_expect(JSON_PARSE_OBJECT_NEXT, lex, JSON_TOKEN_OBJECT_END);
 531
 532         lex->lex_level--;
 533
 534         if (oend != NULL)
 535                 (*oend) (sem->semstate);
 536 }
 537
 538 static void
 539 parse_array_element(JsonLexContext *lex, JsonSemAction *sem)
 540 {
 541         json_aelem_action astart = sem->array_element_start;
 542         json_aelem_action aend = sem->array_element_end;
 543         JsonTokenType tok = lex_peek(lex);
 544
 545         bool            isnull;
 546
 547         isnull = tok == JSON_TOKEN_NULL;
 548
 549         if (astart != NULL)
 550                 (*astart) (sem->semstate, isnull);
 551
 552         /* an array element is any object, array or scalar */
 553         switch (tok)
 554         {
 555                 case JSON_TOKEN_OBJECT_START:
 556                         parse_object(lex, sem);
 557                         break;
 558                 case JSON_TOKEN_ARRAY_START:
 559                         parse_array(lex, sem);
 560                         break;
 561                 default:
 562                         parse_scalar(lex, sem);
 563         }
 564
 565         if (aend != NULL)
 566                 (*aend) (sem->semstate, isnull);
 567 }
 568
 569 static void
 570 parse_array(JsonLexContext *lex, JsonSemAction *sem)
 571 {
 572         /*
 573          * an array is a possibly empty sequence of array elements, separated by
 574          * commas and surrounded by square brackets.
 575          */
 576         json_struct_action astart = sem->array_start;
 577         json_struct_action aend = sem->array_end;
 578
 579         check_stack_depth();
 580
 581         if (astart != NULL)
 582                 (*astart) (sem->semstate);
 583
 584         /*
 585          * Data inside an array is at a higher nesting level than the array
 586          * itself. Note that we increment this after we call the semantic routine
 587          * for the array start and restore it before we call the routine for the
 588          * array end.
 589          */
 590         lex->lex_level++;
 591
 592         lex_expect(JSON_PARSE_ARRAY_START, lex, JSON_TOKEN_ARRAY_START);
 593         if (lex_peek(lex) != JSON_TOKEN_ARRAY_END)
 594         {
 595
 596                 parse_array_element(lex, sem);
 597
 598                 while (lex_accept(lex, JSON_TOKEN_COMMA, NULL))
 599                         parse_array_element(lex, sem);
 600         }
 601
 602         lex_expect(JSON_PARSE_ARRAY_NEXT, lex, JSON_TOKEN_ARRAY_END);
 603
 604         lex->lex_level--;
 605
 606         if (aend != NULL)
 607                 (*aend) (sem->semstate);
 608 }
 609
 610 /*
 611  * Lex one token from the input stream.
 612  */
 613 static inline void
 614 json_lex(JsonLexContext *lex)
 615 {
 616         char       *s;
 617         int                     len;
 618
 619         /* Skip leading whitespace. */
 620         s = lex->token_terminator;
 621         len = s - lex->input;
 622         while (len < lex->input_length &&
 623                    (*s == ' ' || *s == '\t' || *s == '\n' || *s == '\r'))
 624         {
 625                 if (*s == '\n')
 626                         ++lex->line_number;
 627                 ++s;
 628                 ++len;
 629         }
 630         lex->token_start = s;
 631
 632         /* Determine token type. */
 633         if (len >= lex->input_length)
 634         {
 635                 lex->token_start = NULL;
 636                 lex->prev_token_terminator = lex->token_terminator;
 637                 lex->token_terminator = s;
 638                 lex->token_type = JSON_TOKEN_END;
 639         }
 640         else
 641                 switch (*s)
 642                 {
 643                                 /* Single-character token, some kind of punctuation mark. */
 644                         case '{':
 645                                 lex->prev_token_terminator = lex->token_terminator;
 646                                 lex->token_terminator = s + 1;
 647                                 lex->token_type = JSON_TOKEN_OBJECT_START;
 648                                 break;
 649                         case '}':
 650                                 lex->prev_token_terminator = lex->token_terminator;
 651                                 lex->token_terminator = s + 1;
 652                                 lex->token_type = JSON_TOKEN_OBJECT_END;
 653                                 break;
 654                         case '[':
 655                                 lex->prev_token_terminator = lex->token_terminator;
 656                                 lex->token_terminator = s + 1;
 657                                 lex->token_type = JSON_TOKEN_ARRAY_START;
 658                                 break;
 659                         case ']':
 660                                 lex->prev_token_terminator = lex->token_terminator;
 661                                 lex->token_terminator = s + 1;
 662                                 lex->token_type = JSON_TOKEN_ARRAY_END;
 663                                 break;
 664                         case ',':
 665                                 lex->prev_token_terminator = lex->token_terminator;
 666                                 lex->token_terminator = s + 1;
 667                                 lex->token_type = JSON_TOKEN_COMMA;
 668                                 break;
 669                         case ':':
 670                                 lex->prev_token_terminator = lex->token_terminator;
 671                                 lex->token_terminator = s + 1;
 672                                 lex->token_type = JSON_TOKEN_COLON;
 673                                 break;
 674                         case '"':
 675                                 /* string */
 676                                 json_lex_string(lex);
 677                                 lex->token_type = JSON_TOKEN_STRING;
 678                                 break;
 679                         case '-':
 680                                 /* Negative number. */
 681                                 json_lex_number(lex, s + 1, NULL, NULL);
 682                                 lex->token_type = JSON_TOKEN_NUMBER;
 683                                 break;
 684                         case '0':
 685                         case '1':
 686                         case '2':
 687                         case '3':
 688                         case '4':
 689                         case '5':
 690                         case '6':
 691                         case '7':
 692                         case '8':
 693                         case '9':
 694                                 /* Positive number. */
 695                                 json_lex_number(lex, s, NULL, NULL);
 696                                 lex->token_type = JSON_TOKEN_NUMBER;
 697                                 break;
 698                         default:
 699                                 {
 700                                         char       *p;
 701
 702                                         /*
 703                                          * We're not dealing with a string, number, legal
 704                                          * punctuation mark, or end of string.  The only legal
 705                                          * tokens we might find here are true, false, and null,
 706                                          * but for error reporting purposes we scan until we see a
 707                                          * non-alphanumeric character.  That way, we can report
 708                                          * the whole word as an unexpected token, rather than just
 709                                          * some unintuitive prefix thereof.
 710                                          */
 711                                         for (p = s; p - s < lex->input_length - len && JSON_ALPHANUMERIC_CHAR(*p); p++)
 712                                                  /* skip */ ;
 713
 714                                         /*
 715                                          * We got some sort of unexpected punctuation or an
 716                                          * otherwise unexpected character, so just complain about
 717                                          * that one character.
 718                                          */
 719                                         if (p == s)
 720                                         {
 721                                                 lex->prev_token_terminator = lex->token_terminator;
 722                                                 lex->token_terminator = s + 1;
 723                                                 report_invalid_token(lex);
 724                                         }
 725
 726                                         /*
 727                                          * We've got a real alphanumeric token here.  If it
 728                                          * happens to be true, false, or null, all is well.  If
 729                                          * not, error out.
 730                                          */
 731                                         lex->prev_token_terminator = lex->token_terminator;
 732                                         lex->token_terminator = p;
 733                                         if (p - s == 4)
 734                                         {
 735                                                 if (memcmp(s, "true", 4) == 0)
 736                                                         lex->token_type = JSON_TOKEN_TRUE;
 737                                                 else if (memcmp(s, "null", 4) == 0)
 738                                                         lex->token_type = JSON_TOKEN_NULL;
 739                                                 else
 740                                                         report_invalid_token(lex);
 741                                         }
 742                                         else if (p - s == 5 && memcmp(s, "false", 5) == 0)
 743                                                 lex->token_type = JSON_TOKEN_FALSE;
 744                                         else
 745                                                 report_invalid_token(lex);
 746
 747                                 }
 748                 }                                               /* end of switch */
 749 }
 750
 751 /*
 752  * The next token in the input stream is known to be a string; lex it.
 753  */
 754 static inline void
 755 json_lex_string(JsonLexContext *lex)
 756 {
 757         char       *s;
 758         int                     len;
 759         int                     hi_surrogate = -1;
 760
 761         if (lex->strval != NULL)
 762                 resetStringInfo(lex->strval);
 763
 764         Assert(lex->input_length > 0);
 765         s = lex->token_start;
 766         len = lex->token_start - lex->input;
 767         for (;;)
 768         {
 769                 s++;
 770                 len++;
 771                 /* Premature end of the string. */
 772                 if (len >= lex->input_length)
 773                 {
 774                         lex->token_terminator = s;
 775                         report_invalid_token(lex);
 776                 }
 777                 else if (*s == '"')
 778                         break;
 779                 else if ((unsigned char) *s < 32)
 780                 {
 781                         /* Per RFC4627, these characters MUST be escaped. */
 782                         /* Since *s isn't printable, exclude it from the context string */
 783                         lex->token_terminator = s;
 784                         ereport(ERROR,
 785                                         (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
 786                                          errmsg("invalid input syntax for type %s", "json"),
 787                                          errdetail("Character with value 0x%02x must be escaped.",
 788                                                            (unsigned char) *s),
 789                                          report_json_context(lex)));
 790                 }
 791                 else if (*s == '\\')
 792                 {
 793                         /* OK, we have an escape character. */
 794                         s++;
 795                         len++;
 796                         if (len >= lex->input_length)
 797                         {
 798                                 lex->token_terminator = s;
 799                                 report_invalid_token(lex);
 800                         }
 801                         else if (*s == 'u')
 802                         {
 803                                 int                     i;
 804                                 int                     ch = 0;
 805
 806                                 for (i = 1; i <= 4; i++)
 807                                 {
 808                                         s++;
 809                                         len++;
 810                                         if (len >= lex->input_length)
 811                                         {
 812                                                 lex->token_terminator = s;
 813                                                 report_invalid_token(lex);
 814                                         }
 815                                         else if (*s >= '0' && *s <= '9')
 816                                                 ch = (ch * 16) + (*s - '0');
 817                                         else if (*s >= 'a' && *s <= 'f')
 818                                                 ch = (ch * 16) + (*s - 'a') + 10;
 819                                         else if (*s >= 'A' && *s <= 'F')
 820                                                 ch = (ch * 16) + (*s - 'A') + 10;
 821                                         else
 822                                         {
 823                                                 lex->token_terminator = s + pg_mblen(s);
 824                                                 ereport(ERROR,
 825                                                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
 826                                                                  errmsg("invalid input syntax for type %s",
 827                                                                                 "json"),
 828                                                                  errdetail("\"\\u\" must be followed by four hexadecimal digits."),
 829                                                                  report_json_context(lex)));
 830                                         }
 831                                 }
 832                                 if (lex->strval != NULL)
 833                                 {
 834                                         char            utf8str[5];
 835                                         int                     utf8len;
 836
 837                                         if (ch >= 0xd800 && ch <= 0xdbff)
 838                                         {
 839                                                 if (hi_surrogate != -1)
 840                                                         ereport(ERROR,
 841                                                                         (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
 842                                                                          errmsg("invalid input syntax for type %s",
 843                                                                                         "json"),
 844                                                                          errdetail("Unicode high surrogate must not follow a high surrogate."),
 845                                                                          report_json_context(lex)));
 846                                                 hi_surrogate = (ch & 0x3ff) << 10;
 847                                                 continue;
 848                                         }
 849                                         else if (ch >= 0xdc00 && ch <= 0xdfff)
 850                                         {
 851                                                 if (hi_surrogate == -1)
 852                                                         ereport(ERROR,
 853                                                                         (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
 854                                                                          errmsg("invalid input syntax for type %s", "json"),
 855                                                                          errdetail("Unicode low surrogate must follow a high surrogate."),
 856                                                                          report_json_context(lex)));
 857                                                 ch = 0x10000 + hi_surrogate + (ch & 0x3ff);
 858                                                 hi_surrogate = -1;
 859                                         }
 860
 861                                         if (hi_surrogate != -1)
 862                                                 ereport(ERROR,
 863                                                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
 864                                                                  errmsg("invalid input syntax for type %s", "json"),
 865                                                                  errdetail("Unicode low surrogate must follow a high surrogate."),
 866                                                                  report_json_context(lex)));
 867
 868                                         /*
 869                                          * For UTF8, replace the escape sequence by the actual
 870                                          * utf8 character in lex->strval. Do this also for other
 871                                          * encodings if the escape designates an ASCII character,
 872                                          * otherwise raise an error.
 873                                          */
 874
 875                                         if (ch == 0)
 876                                         {
 877                                                 /* We can't allow this, since our TEXT type doesn't */
 878                                                 ereport(ERROR,
 879                                                                 (errcode(ERRCODE_UNTRANSLATABLE_CHARACTER),
 880                                                                  errmsg("unsupported Unicode escape sequence"),
 881                                                                  errdetail("\\u0000 cannot be converted to text."),
 882                                                                  report_json_context(lex)));
 883                                         }
 884                                         else if (GetDatabaseEncoding() == PG_UTF8)
 885                                         {
 886                                                 unicode_to_utf8(ch, (unsigned char *) utf8str);
 887                                                 utf8len = pg_utf_mblen((unsigned char *) utf8str);
 888                                                 appendBinaryStringInfo(lex->strval, utf8str, utf8len);
 889                                         }
 890                                         else if (ch <= 0x007f)
 891                                         {
 892                                                 /*
 893                                                  * This is the only way to designate things like a
 894                                                  * form feed character in JSON, so it's useful in all
 895                                                  * encodings.
 896                                                  */
 897                                                 appendStringInfoChar(lex->strval, (char) ch);
 898                                         }
 899                                         else
 900                                         {
 901                                                 ereport(ERROR,
 902                                                                 (errcode(ERRCODE_UNTRANSLATABLE_CHARACTER),
 903                                                                  errmsg("unsupported Unicode escape sequence"),
 904                                                                  errdetail("Unicode escape values cannot be used for code point values above 007F when the server encoding is not UTF8."),
 905                                                                  report_json_context(lex)));
 906                                         }
 907
 908                                 }
 909                         }
 910                         else if (lex->strval != NULL)
 911                         {
 912                                 if (hi_surrogate != -1)
 913                                         ereport(ERROR,
 914                                                         (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
 915                                                          errmsg("invalid input syntax for type %s",
 916                                                                         "json"),
 917                                                          errdetail("Unicode low surrogate must follow a high surrogate."),
 918                                                          report_json_context(lex)));
 919
 920                                 switch (*s)
 921                                 {
 922                                         case '"':
 923                                         case '\\':
 924                                         case '/':
 925                                                 appendStringInfoChar(lex->strval, *s);
 926                                                 break;
 927                                         case 'b':
 928                                                 appendStringInfoChar(lex->strval, '\b');
 929                                                 break;
 930                                         case 'f':
 931                                                 appendStringInfoChar(lex->strval, '\f');
 932                                                 break;
 933                                         case 'n':
 934                                                 appendStringInfoChar(lex->strval, '\n');
 935                                                 break;
 936                                         case 'r':
 937                                                 appendStringInfoChar(lex->strval, '\r');
 938                                                 break;
 939                                         case 't':
 940                                                 appendStringInfoChar(lex->strval, '\t');
 941                                                 break;
 942                                         default:
 943                                                 /* Not a valid string escape, so error out. */
 944                                                 lex->token_terminator = s + pg_mblen(s);
 945                                                 ereport(ERROR,
 946                                                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
 947                                                                  errmsg("invalid input syntax for type %s",
 948                                                                                 "json"),
 949                                                                  errdetail("Escape sequence \"\\%s\" is invalid.",
 950                                                                                    extract_mb_char(s)),
 951                                                                  report_json_context(lex)));
 952                                 }
 953                         }
 954                         else if (strchr("\"\\/bfnrt", *s) == NULL)
 955                         {
 956                                 /*
 957                                  * Simpler processing if we're not bothered about de-escaping
 958                                  *
 959                                  * It's very tempting to remove the strchr() call here and
 960                                  * replace it with a switch statement, but testing so far has
 961                                  * shown it's not a performance win.
 962                                  */
 963                                 lex->token_terminator = s + pg_mblen(s);
 964                                 ereport(ERROR,
 965                                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
 966                                                  errmsg("invalid input syntax for type %s", "json"),
 967                                                  errdetail("Escape sequence \"\\%s\" is invalid.",
 968                                                                    extract_mb_char(s)),
 969                                                  report_json_context(lex)));
 970                         }
 971
 972                 }
 973                 else if (lex->strval != NULL)
 974                 {
 975                         if (hi_surrogate != -1)
 976                                 ereport(ERROR,
 977                                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
 978                                                  errmsg("invalid input syntax for type %s", "json"),
 979                                                  errdetail("Unicode low surrogate must follow a high surrogate."),
 980                                                  report_json_context(lex)));
 981
 982                         appendStringInfoChar(lex->strval, *s);
 983                 }
 984
 985         }
 986
 987         if (hi_surrogate != -1)
 988                 ereport(ERROR,
 989                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
 990                                  errmsg("invalid input syntax for type %s", "json"),
 991                                  errdetail("Unicode low surrogate must follow a high surrogate."),
 992                                  report_json_context(lex)));
 993
 994         /* Hooray, we found the end of the string! */
 995         lex->prev_token_terminator = lex->token_terminator;
 996         lex->token_terminator = s + 1;
 997 }
 998
 999 /*
1000  * The next token in the input stream is known to be a number; lex it.
1001  *
1002  * In JSON, a number consists of four parts:
1003  *
1004  * (1) An optional minus sign ('-').
1005  *
1006  * (2) Either a single '0', or a string of one or more digits that does not
1007  *         begin with a '0'.
1008  *
1009  * (3) An optional decimal part, consisting of a period ('.') followed by
1010  *         one or more digits.  (Note: While this part can be omitted
1011  *         completely, it's not OK to have only the decimal point without
1012  *         any digits afterwards.)
1013  *
1014  * (4) An optional exponent part, consisting of 'e' or 'E', optionally
1015  *         followed by '+' or '-', followed by one or more digits.  (Note:
1016  *         As with the decimal part, if 'e' or 'E' is present, it must be
1017  *         followed by at least one digit.)
1018  *
1019  * The 's' argument to this function points to the ostensible beginning
1020  * of part 2 - i.e. the character after any optional minus sign, or the
1021  * first character of the string if there is none.
1022  *
1023  * If num_err is not NULL, we return an error flag to *num_err rather than
1024  * raising an error for a badly-formed number.  Also, if total_len is not NULL
1025  * the distance from lex->input to the token end+1 is returned to *total_len.
1026  */
1027 static inline void
1028 json_lex_number(JsonLexContext *lex, char *s,
1029                                 bool *num_err, int *total_len)
1030 {
1031         bool            error = false;
1032         int                     len = s - lex->input;
1033
1034         /* Part (1): leading sign indicator. */
1035         /* Caller already did this for us; so do nothing. */
1036
1037         /* Part (2): parse main digit string. */
1038         if (len < lex->input_length && *s == '0')
1039         {
1040                 s++;
1041                 len++;
1042         }
1043         else if (len < lex->input_length && *s >= '1' && *s <= '9')
1044         {
1045                 do
1046                 {
1047                         s++;
1048                         len++;
1049                 } while (len < lex->input_length && *s >= '0' && *s <= '9');
1050         }
1051         else
1052                 error = true;
1053
1054         /* Part (3): parse optional decimal portion. */
1055         if (len < lex->input_length && *s == '.')
1056         {
1057                 s++;
1058                 len++;
1059                 if (len == lex->input_length || *s < '0' || *s > '9')
1060                         error = true;
1061                 else
1062                 {
1063                         do
1064                         {
1065                                 s++;
1066                                 len++;
1067                         } while (len < lex->input_length && *s >= '0' && *s <= '9');
1068                 }
1069         }
1070
1071         /* Part (4): parse optional exponent. */
1072         if (len < lex->input_length && (*s == 'e' || *s == 'E'))
1073         {
1074                 s++;
1075                 len++;
1076                 if (len < lex->input_length && (*s == '+' || *s == '-'))
1077                 {
1078                         s++;
1079                         len++;
1080                 }
1081                 if (len == lex->input_length || *s < '0' || *s > '9')
1082                         error = true;
1083                 else
1084                 {
1085                         do
1086                         {
1087                                 s++;
1088                                 len++;
1089                         } while (len < lex->input_length && *s >= '0' && *s <= '9');
1090                 }
1091         }
1092
1093         /*
1094          * Check for trailing garbage.  As in json_lex(), any alphanumeric stuff
1095          * here should be considered part of the token for error-reporting
1096          * purposes.
1097          */
1098         for (; len < lex->input_length && JSON_ALPHANUMERIC_CHAR(*s); s++, len++)
1099                 error = true;
1100
1101         if (total_len != NULL)
1102                 *total_len = len;
1103
1104         if (num_err != NULL)
1105         {
1106                 /* let the caller handle any error */
1107                 *num_err = error;
1108         }
1109         else
1110         {
1111                 /* return token endpoint */
1112                 lex->prev_token_terminator = lex->token_terminator;
1113                 lex->token_terminator = s;
1114                 /* handle error if any */
1115                 if (error)
1116                         report_invalid_token(lex);
1117         }
1118 }
1119
1120 /*
1121  * Report a parse error.
1122  *
1123  * lex->token_start and lex->token_terminator must identify the current token.
1124  */
1125 static void
1126 report_parse_error(JsonParseContext ctx, JsonLexContext *lex)
1127 {
1128         char       *token;
1129         int                     toklen;
1130
1131         /* Handle case where the input ended prematurely. */
1132         if (lex->token_start == NULL || lex->token_type == JSON_TOKEN_END)
1133                 ereport(ERROR,
1134                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
1135                                  errmsg("invalid input syntax for type %s", "json"),
1136                                  errdetail("The input string ended unexpectedly."),
1137                                  report_json_context(lex)));
1138
1139         /* Separate out the current token. */
1140         toklen = lex->token_terminator - lex->token_start;
1141         token = palloc(toklen + 1);
1142         memcpy(token, lex->token_start, toklen);
1143         token[toklen] = '\0';
1144
1145         /* Complain, with the appropriate detail message. */
1146         if (ctx == JSON_PARSE_END)
1147                 ereport(ERROR,
1148                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
1149                                  errmsg("invalid input syntax for type %s", "json"),
1150                                  errdetail("Expected end of input, but found \"%s\".",
1151                                                    token),
1152                                  report_json_context(lex)));
1153         else
1154         {
1155                 switch (ctx)
1156                 {
1157                         case JSON_PARSE_VALUE:
1158                                 ereport(ERROR,
1159                                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
1160                                                  errmsg("invalid input syntax for type %s", "json"),
1161                                                  errdetail("Expected JSON value, but found \"%s\".",
1162                                                                    token),
1163                                                  report_json_context(lex)));
1164                                 break;
1165                         case JSON_PARSE_STRING:
1166                                 ereport(ERROR,
1167                                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
1168                                                  errmsg("invalid input syntax for type %s", "json"),
1169                                                  errdetail("Expected string, but found \"%s\".",
1170                                                                    token),
1171                                                  report_json_context(lex)));
1172                                 break;
1173                         case JSON_PARSE_ARRAY_START:
1174                                 ereport(ERROR,
1175                                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
1176                                                  errmsg("invalid input syntax for type %s", "json"),
1177                                                  errdetail("Expected array element or \"]\", but found \"%s\".",
1178                                                                    token),
1179                                                  report_json_context(lex)));
1180                                 break;
1181                         case JSON_PARSE_ARRAY_NEXT:
1182                                 ereport(ERROR,
1183                                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
1184                                                  errmsg("invalid input syntax for type %s", "json"),
1185                                                  errdetail("Expected \",\" or \"]\", but found \"%s\".",
1186                                                                    token),
1187                                                  report_json_context(lex)));
1188                                 break;
1189                         case JSON_PARSE_OBJECT_START:
1190                                 ereport(ERROR,
1191                                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
1192                                                  errmsg("invalid input syntax for type %s", "json"),
1193                                                  errdetail("Expected string or \"}\", but found \"%s\".",
1194                                                                    token),
1195                                                  report_json_context(lex)));
1196                                 break;
1197                         case JSON_PARSE_OBJECT_LABEL:
1198                                 ereport(ERROR,
1199                                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
1200                                                  errmsg("invalid input syntax for type %s", "json"),
1201                                                  errdetail("Expected \":\", but found \"%s\".",
1202                                                                    token),
1203                                                  report_json_context(lex)));
1204                                 break;
1205                         case JSON_PARSE_OBJECT_NEXT:
1206                                 ereport(ERROR,
1207                                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
1208                                                  errmsg("invalid input syntax for type %s", "json"),
1209                                                  errdetail("Expected \",\" or \"}\", but found \"%s\".",
1210                                                                    token),
1211                                                  report_json_context(lex)));
1212                                 break;
1213                         case JSON_PARSE_OBJECT_COMMA:
1214                                 ereport(ERROR,
1215                                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
1216                                                  errmsg("invalid input syntax for type %s", "json"),
1217                                                  errdetail("Expected string, but found \"%s\".",
1218                                                                    token),
1219                                                  report_json_context(lex)));
1220                                 break;
1221                         default:
1222                                 elog(ERROR, "unexpected json parse state: %d", ctx);
1223                 }
1224         }
1225 }
1226
1227 /*
1228  * Report an invalid input token.
1229  *
1230  * lex->token_start and lex->token_terminator must identify the token.
1231  */
1232 static void
1233 report_invalid_token(JsonLexContext *lex)
1234 {
1235         char       *token;
1236         int                     toklen;
1237
1238         /* Separate out the offending token. */
1239         toklen = lex->token_terminator - lex->token_start;
1240         token = palloc(toklen + 1);
1241         memcpy(token, lex->token_start, toklen);
1242         token[toklen] = '\0';
1243
1244         ereport(ERROR,
1245                         (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
1246                          errmsg("invalid input syntax for type %s", "json"),
1247                          errdetail("Token \"%s\" is invalid.", token),
1248                          report_json_context(lex)));
1249 }
1250
1251 /*
1252  * Report a CONTEXT line for bogus JSON input.
1253  *
1254  * lex->token_terminator must be set to identify the spot where we detected
1255  * the error.  Note that lex->token_start might be NULL, in case we recognized
1256  * error at EOF.
1257  *
1258  * The return value isn't meaningful, but we make it non-void so that this
1259  * can be invoked inside ereport().
1260  */
1261 static int
1262 report_json_context(JsonLexContext *lex)
1263 {
1264         const char *context_start;
1265         const char *context_end;
1266         const char *line_start;
1267         int                     line_number;
1268         char       *ctxt;
1269         int                     ctxtlen;
1270         const char *prefix;
1271         const char *suffix;
1272
1273         /* Choose boundaries for the part of the input we will display */
1274         context_start = lex->input;
1275         context_end = lex->token_terminator;
1276         line_start = context_start;
1277         line_number = 1;
1278         for (;;)
1279         {
1280                 /* Always advance over newlines */
1281                 if (context_start < context_end && *context_start == '\n')
1282                 {
1283                         context_start++;
1284                         line_start = context_start;
1285                         line_number++;
1286                         continue;
1287                 }
1288                 /* Otherwise, done as soon as we are close enough to context_end */
1289                 if (context_end - context_start < 50)
1290                         break;
1291                 /* Advance to next multibyte character */
1292                 if (IS_HIGHBIT_SET(*context_start))
1293                         context_start += pg_mblen(context_start);
1294                 else
1295                         context_start++;
1296         }
1297
1298         /*
1299          * We add "..." to indicate that the excerpt doesn't start at the
1300          * beginning of the line ... but if we're within 3 characters of the
1301          * beginning of the line, we might as well just show the whole line.
1302          */
1303         if (context_start - line_start <= 3)
1304                 context_start = line_start;
1305
1306         /* Get a null-terminated copy of the data to present */
1307         ctxtlen = context_end - context_start;
1308         ctxt = palloc(ctxtlen + 1);
1309         memcpy(ctxt, context_start, ctxtlen);
1310         ctxt[ctxtlen] = '\0';
1311
1312         /*
1313          * Show the context, prefixing "..." if not starting at start of line, and
1314          * suffixing "..." if not ending at end of line.
1315          */
1316         prefix = (context_start > line_start) ? "..." : "";
1317         suffix = (lex->token_type != JSON_TOKEN_END && context_end - lex->input < lex->input_length && *context_end != '\n' && *context_end != '\r') ? "..." : "";
1318
1319         return errcontext("JSON data, line %d: %s%s%s",
1320                                           line_number, prefix, ctxt, suffix);
1321 }
1322
1323 /*
1324  * Extract a single, possibly multi-byte char from the input string.
1325  */
1326 static char *
1327 extract_mb_char(char *s)
1328 {
1329         char       *res;
1330         int                     len;
1331
1332         len = pg_mblen(s);
1333         res = palloc(len + 1);
1334         memcpy(res, s, len);
1335         res[len] = '\0';
1336
1337         return res;
1338 }
1339
1340 /*
1341  * Determine how we want to print values of a given type in datum_to_json.
1342  *
1343  * Given the datatype OID, return its JsonTypeCategory, as well as the type's
1344  * output function OID.  If the returned category is JSONTYPE_CAST, we
1345  * return the OID of the type->JSON cast function instead.
1346  */
1347 static void
1348 json_categorize_type(Oid typoid,
1349                                          JsonTypeCategory *tcategory,
1350                                          Oid *outfuncoid)
1351 {
1352         bool            typisvarlena;
1353
1354         /* Look through any domain */
1355         typoid = getBaseType(typoid);
1356
1357         *outfuncoid = InvalidOid;
1358
1359         /*
1360          * We need to get the output function for everything except date and
1361          * timestamp types, array and composite types, booleans, and non-builtin
1362          * types where there's a cast to json.
1363          */
1364
1365         switch (typoid)
1366         {
1367                 case BOOLOID:
1368                         *tcategory = JSONTYPE_BOOL;
1369                         break;
1370
1371                 case INT2OID:
1372                 case INT4OID:
1373                 case INT8OID:
1374                 case FLOAT4OID:
1375                 case FLOAT8OID:
1376                 case NUMERICOID:
1377                         getTypeOutputInfo(typoid, outfuncoid, &typisvarlena);
1378                         *tcategory = JSONTYPE_NUMERIC;
1379                         break;
1380
1381                 case DATEOID:
1382                         *tcategory = JSONTYPE_DATE;
1383                         break;
1384
1385                 case TIMESTAMPOID:
1386                         *tcategory = JSONTYPE_TIMESTAMP;
1387                         break;
1388
1389                 case TIMESTAMPTZOID:
1390                         *tcategory = JSONTYPE_TIMESTAMPTZ;
1391                         break;
1392
1393                 case JSONOID:
1394                 case JSONBOID:
1395                         getTypeOutputInfo(typoid, outfuncoid, &typisvarlena);
1396                         *tcategory = JSONTYPE_JSON;
1397                         break;
1398
1399                 default:
1400                         /* Check for arrays and composites */
1401                         if (OidIsValid(get_element_type(typoid)) || typoid == ANYARRAYOID
1402                                 || typoid == RECORDARRAYOID)
1403                                 *tcategory = JSONTYPE_ARRAY;
1404                         else if (type_is_rowtype(typoid))       /* includes RECORDOID */
1405                                 *tcategory = JSONTYPE_COMPOSITE;
1406                         else
1407                         {
1408                                 /* It's probably the general case ... */
1409                                 *tcategory = JSONTYPE_OTHER;
1410                                 /* but let's look for a cast to json, if it's not built-in */
1411                                 if (typoid >= FirstNormalObjectId)
1412                                 {
1413                                         Oid                     castfunc;
1414                                         CoercionPathType ctype;
1415
1416                                         ctype = find_coercion_pathway(JSONOID, typoid,
1417                                                                                                   COERCION_EXPLICIT,
1418                                                                                                   &castfunc);
1419                                         if (ctype == COERCION_PATH_FUNC && OidIsValid(castfunc))
1420                                         {
1421                                                 *tcategory = JSONTYPE_CAST;
1422                                                 *outfuncoid = castfunc;
1423                                         }
1424                                         else
1425                                         {
1426                                                 /* non builtin type with no cast */
1427                                                 getTypeOutputInfo(typoid, outfuncoid, &typisvarlena);
1428                                         }
1429                                 }
1430                                 else
1431                                 {
1432                                         /* any other builtin type */
1433                                         getTypeOutputInfo(typoid, outfuncoid, &typisvarlena);
1434                                 }
1435                         }
1436                         break;
1437         }
1438 }
1439
1440 /*
1441  * Turn a Datum into JSON text, appending the string to "result".
1442  *
1443  * tcategory and outfuncoid are from a previous call to json_categorize_type,
1444  * except that if is_null is true then they can be invalid.
1445  *
1446  * If key_scalar is true, the value is being printed as a key, so insist
1447  * it's of an acceptable type, and force it to be quoted.
1448  */
1449 static void
1450 datum_to_json(Datum val, bool is_null, StringInfo result,
1451                           JsonTypeCategory tcategory, Oid outfuncoid,
1452                           bool key_scalar)
1453 {
1454         char       *outputstr;
1455         text       *jsontext;
1456
1457         check_stack_depth();
1458
1459         /* callers are expected to ensure that null keys are not passed in */
1460         Assert(!(key_scalar && is_null));
1461
1462         if (is_null)
1463         {
1464                 appendStringInfoString(result, "null");
1465                 return;
1466         }
1467
1468         if (key_scalar &&
1469                 (tcategory == JSONTYPE_ARRAY ||
1470                  tcategory == JSONTYPE_COMPOSITE ||
1471                  tcategory == JSONTYPE_JSON ||
1472                  tcategory == JSONTYPE_CAST))
1473                 ereport(ERROR,
1474                                 (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1475                                  errmsg("key value must be scalar, not array, composite, or json")));
1476
1477         switch (tcategory)
1478         {
1479                 case JSONTYPE_ARRAY:
1480                         array_to_json_internal(val, result, false);
1481                         break;
1482                 case JSONTYPE_COMPOSITE:
1483                         composite_to_json(val, result, false);
1484                         break;
1485                 case JSONTYPE_BOOL:
1486                         outputstr = DatumGetBool(val) ? "true" : "false";
1487                         if (key_scalar)
1488                                 escape_json(result, outputstr);
1489                         else
1490                                 appendStringInfoString(result, outputstr);
1491                         break;
1492                 case JSONTYPE_NUMERIC:
1493                         outputstr = OidOutputFunctionCall(outfuncoid, val);
1494
1495                         /*
1496                          * Don't call escape_json for a non-key if it's a valid JSON
1497                          * number.
1498                          */
1499                         if (!key_scalar && IsValidJsonNumber(outputstr, strlen(outputstr)))
1500                                 appendStringInfoString(result, outputstr);
1501                         else
1502                                 escape_json(result, outputstr);
1503                         pfree(outputstr);
1504                         break;
1505                 case JSONTYPE_DATE:
1506                         {
1507                                 char            buf[MAXDATELEN + 1];
1508
1509                                 JsonEncodeDateTime(buf, val, DATEOID);
1510                                 appendStringInfo(result, "\"%s\"", buf);
1511                         }
1512                         break;
1513                 case JSONTYPE_TIMESTAMP:
1514                         {
1515                                 char            buf[MAXDATELEN + 1];
1516
1517                                 JsonEncodeDateTime(buf, val, TIMESTAMPOID);
1518                                 appendStringInfo(result, "\"%s\"", buf);
1519                         }
1520                         break;
1521                 case JSONTYPE_TIMESTAMPTZ:
1522                         {
1523                                 char            buf[MAXDATELEN + 1];
1524
1525                                 JsonEncodeDateTime(buf, val, TIMESTAMPTZOID);
1526                                 appendStringInfo(result, "\"%s\"", buf);
1527                         }
1528                         break;
1529                 case JSONTYPE_JSON:
1530                         /* JSON and JSONB output will already be escaped */
1531                         outputstr = OidOutputFunctionCall(outfuncoid, val);
1532                         appendStringInfoString(result, outputstr);
1533                         pfree(outputstr);
1534                         break;
1535                 case JSONTYPE_CAST:
1536                         /* outfuncoid refers to a cast function, not an output function */
1537                         jsontext = DatumGetTextPP(OidFunctionCall1(outfuncoid, val));
1538                         outputstr = text_to_cstring(jsontext);
1539                         appendStringInfoString(result, outputstr);
1540                         pfree(outputstr);
1541                         pfree(jsontext);
1542                         break;
1543                 default:
1544                         outputstr = OidOutputFunctionCall(outfuncoid, val);
1545                         escape_json(result, outputstr);
1546                         pfree(outputstr);
1547                         break;
1548         }
1549 }
1550
1551 /*
1552  * Encode 'value' of datetime type 'typid' into JSON string in ISO format using
1553  * optionally preallocated buffer 'buf'.
1554  */
1555 char *
1556 JsonEncodeDateTime(char *buf, Datum value, Oid typid)
1557 {
1558         if (!buf)
1559                 buf = palloc(MAXDATELEN + 1);
1560
1561         switch (typid)
1562         {
1563                 case DATEOID:
1564                         {
1565                                 DateADT         date;
1566                                 struct pg_tm tm;
1567
1568                                 date = DatumGetDateADT(value);
1569
1570                                 /* Same as date_out(), but forcing DateStyle */
1571                                 if (DATE_NOT_FINITE(date))
1572                                         EncodeSpecialDate(date, buf);
1573                                 else
1574                                 {
1575                                         j2date(date + POSTGRES_EPOCH_JDATE,
1576                                                    &(tm.tm_year), &(tm.tm_mon), &(tm.tm_mday));
1577                                         EncodeDateOnly(&tm, USE_XSD_DATES, buf);
1578                                 }
1579                         }
1580                         break;
1581                 case TIMEOID:
1582                         {
1583                                 TimeADT         time = DatumGetTimeADT(value);
1584                                 struct pg_tm tt,
1585                                                    *tm = &tt;
1586                                 fsec_t          fsec;
1587
1588                                 /* Same as time_out(), but forcing DateStyle */
1589                                 time2tm(time, tm, &fsec);
1590                                 EncodeTimeOnly(tm, fsec, false, 0, USE_XSD_DATES, buf);
1591                         }
1592                         break;
1593                 case TIMETZOID:
1594                         {
1595                                 TimeTzADT  *time = DatumGetTimeTzADTP(value);
1596                                 struct pg_tm tt,
1597                                                    *tm = &tt;
1598                                 fsec_t          fsec;
1599                                 int                     tz;
1600
1601                                 /* Same as timetz_out(), but forcing DateStyle */
1602                                 timetz2tm(time, tm, &fsec, &tz);
1603                                 EncodeTimeOnly(tm, fsec, true, tz, USE_XSD_DATES, buf);
1604                         }
1605                         break;
1606                 case TIMESTAMPOID:
1607                         {
1608                                 Timestamp       timestamp;
1609                                 struct pg_tm tm;
1610                                 fsec_t          fsec;
1611
1612                                 timestamp = DatumGetTimestamp(value);
1613                                 /* Same as timestamp_out(), but forcing DateStyle */
1614                                 if (TIMESTAMP_NOT_FINITE(timestamp))
1615                                         EncodeSpecialTimestamp(timestamp, buf);
1616                                 else if (timestamp2tm(timestamp, NULL, &tm, &fsec, NULL, NULL) == 0)
1617                                         EncodeDateTime(&tm, fsec, false, 0, NULL, USE_XSD_DATES, buf);
1618                                 else
1619                                         ereport(ERROR,
1620                                                         (errcode(ERRCODE_DATETIME_VALUE_OUT_OF_RANGE),
1621                                                          errmsg("timestamp out of range")));
1622                         }
1623                         break;
1624                 case TIMESTAMPTZOID:
1625                         {
1626                                 TimestampTz timestamp;
1627                                 struct pg_tm tm;
1628                                 int                     tz;
1629                                 fsec_t          fsec;
1630                                 const char *tzn = NULL;
1631
1632                                 timestamp = DatumGetTimestampTz(value);
1633                                 /* Same as timestamptz_out(), but forcing DateStyle */
1634                                 if (TIMESTAMP_NOT_FINITE(timestamp))
1635                                         EncodeSpecialTimestamp(timestamp, buf);
1636                                 else if (timestamp2tm(timestamp, &tz, &tm, &fsec, &tzn, NULL) == 0)
1637                                         EncodeDateTime(&tm, fsec, true, tz, tzn, USE_XSD_DATES, buf);
1638                                 else
1639                                         ereport(ERROR,
1640                                                         (errcode(ERRCODE_DATETIME_VALUE_OUT_OF_RANGE),
1641                                                          errmsg("timestamp out of range")));
1642                         }
1643                         break;
1644                 default:
1645                         elog(ERROR, "unknown jsonb value datetime type oid %d", typid);
1646                         return NULL;
1647         }
1648
1649         return buf;
1650 }
1651
1652 /*
1653  * Process a single dimension of an array.
1654  * If it's the innermost dimension, output the values, otherwise call
1655  * ourselves recursively to process the next dimension.
1656  */
1657 static void
1658 array_dim_to_json(StringInfo result, int dim, int ndims, int *dims, Datum *vals,
1659                                   bool *nulls, int *valcount, JsonTypeCategory tcategory,
1660                                   Oid outfuncoid, bool use_line_feeds)
1661 {
1662         int                     i;
1663         const char *sep;
1664
1665         Assert(dim < ndims);
1666
1667         sep = use_line_feeds ? ",\n " : ",";
1668
1669         appendStringInfoChar(result, '[');
1670
1671         for (i = 1; i <= dims[dim]; i++)
1672         {
1673                 if (i > 1)
1674                         appendStringInfoString(result, sep);
1675
1676                 if (dim + 1 == ndims)
1677                 {
1678                         datum_to_json(vals[*valcount], nulls[*valcount], result, tcategory,
1679                                                   outfuncoid, false);
1680                         (*valcount)++;
1681                 }
1682                 else
1683                 {
1684                         /*
1685                          * Do we want line feeds on inner dimensions of arrays? For now
1686                          * we'll say no.
1687                          */
1688                         array_dim_to_json(result, dim + 1, ndims, dims, vals, nulls,
1689                                                           valcount, tcategory, outfuncoid, false);
1690                 }
1691         }
1692
1693         appendStringInfoChar(result, ']');
1694 }
1695
1696 /*
1697  * Turn an array into JSON.
1698  */
1699 static void
1700 array_to_json_internal(Datum array, StringInfo result, bool use_line_feeds)
1701 {
1702         ArrayType  *v = DatumGetArrayTypeP(array);
1703         Oid                     element_type = ARR_ELEMTYPE(v);
1704         int                *dim;
1705         int                     ndim;
1706         int                     nitems;
1707         int                     count = 0;
1708         Datum      *elements;
1709         bool       *nulls;
1710         int16           typlen;
1711         bool            typbyval;
1712         char            typalign;
1713         JsonTypeCategory tcategory;
1714         Oid                     outfuncoid;
1715
1716         ndim = ARR_NDIM(v);
1717         dim = ARR_DIMS(v);
1718         nitems = ArrayGetNItems(ndim, dim);
1719
1720         if (nitems <= 0)
1721         {
1722                 appendStringInfoString(result, "[]");
1723                 return;
1724         }
1725
1726         get_typlenbyvalalign(element_type,
1727                                                  &typlen, &typbyval, &typalign);
1728
1729         json_categorize_type(element_type,
1730                                                  &tcategory, &outfuncoid);
1731
1732         deconstruct_array(v, element_type, typlen, typbyval,
1733                                           typalign, &elements, &nulls,
1734                                           &nitems);
1735
1736         array_dim_to_json(result, 0, ndim, dim, elements, nulls, &count, tcategory,
1737                                           outfuncoid, use_line_feeds);
1738
1739         pfree(elements);
1740         pfree(nulls);
1741 }
1742
1743 /*
1744  * Turn a composite / record into JSON.
1745  */
1746 static void
1747 composite_to_json(Datum composite, StringInfo result, bool use_line_feeds)
1748 {
1749         HeapTupleHeader td;
1750         Oid                     tupType;
1751         int32           tupTypmod;
1752         TupleDesc       tupdesc;
1753         HeapTupleData tmptup,
1754                            *tuple;
1755         int                     i;
1756         bool            needsep = false;
1757         const char *sep;
1758
1759         sep = use_line_feeds ? ",\n " : ",";
1760
1761         td = DatumGetHeapTupleHeader(composite);
1762
1763         /* Extract rowtype info and find a tupdesc */
1764         tupType = HeapTupleHeaderGetTypeId(td);
1765         tupTypmod = HeapTupleHeaderGetTypMod(td);
1766         tupdesc = lookup_rowtype_tupdesc(tupType, tupTypmod);
1767
1768         /* Build a temporary HeapTuple control structure */
1769         tmptup.t_len = HeapTupleHeaderGetDatumLength(td);
1770         tmptup.t_data = td;
1771         tuple = &tmptup;
1772
1773         appendStringInfoChar(result, '{');
1774
1775         for (i = 0; i < tupdesc->natts; i++)
1776         {
1777                 Datum           val;
1778                 bool            isnull;
1779                 char       *attname;
1780                 JsonTypeCategory tcategory;
1781                 Oid                     outfuncoid;
1782                 Form_pg_attribute att = TupleDescAttr(tupdesc, i);
1783
1784                 if (att->attisdropped)
1785                         continue;
1786
1787                 if (needsep)
1788                         appendStringInfoString(result, sep);
1789                 needsep = true;
1790
1791                 attname = NameStr(att->attname);
1792                 escape_json(result, attname);
1793                 appendStringInfoChar(result, ':');
1794
1795                 val = heap_getattr(tuple, i + 1, tupdesc, &isnull);
1796
1797                 if (isnull)
1798                 {
1799                         tcategory = JSONTYPE_NULL;
1800                         outfuncoid = InvalidOid;
1801                 }
1802                 else
1803                         json_categorize_type(att->atttypid, &tcategory, &outfuncoid);
1804
1805                 datum_to_json(val, isnull, result, tcategory, outfuncoid, false);
1806         }
1807
1808         appendStringInfoChar(result, '}');
1809         ReleaseTupleDesc(tupdesc);
1810 }
1811
1812 /*
1813  * Append JSON text for "val" to "result".
1814  *
1815  * This is just a thin wrapper around datum_to_json.  If the same type will be
1816  * printed many times, avoid using this; better to do the json_categorize_type
1817  * lookups only once.
1818  */
1819 static void
1820 add_json(Datum val, bool is_null, StringInfo result,
1821                  Oid val_type, bool key_scalar)
1822 {
1823         JsonTypeCategory tcategory;
1824         Oid                     outfuncoid;
1825
1826         if (val_type == InvalidOid)
1827                 ereport(ERROR,
1828                                 (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1829                                  errmsg("could not determine input data type")));
1830
1831         if (is_null)
1832         {
1833                 tcategory = JSONTYPE_NULL;
1834                 outfuncoid = InvalidOid;
1835         }
1836         else
1837                 json_categorize_type(val_type,
1838                                                          &tcategory, &outfuncoid);
1839
1840         datum_to_json(val, is_null, result, tcategory, outfuncoid, key_scalar);
1841 }
1842
1843 /*
1844  * SQL function array_to_json(row)
1845  */
1846 Datum
1847 array_to_json(PG_FUNCTION_ARGS)
1848 {
1849         Datum           array = PG_GETARG_DATUM(0);
1850         StringInfo      result;
1851
1852         result = makeStringInfo();
1853
1854         array_to_json_internal(array, result, false);
1855
1856         PG_RETURN_TEXT_P(cstring_to_text_with_len(result->data, result->len));
1857 }
1858
1859 /*
1860  * SQL function array_to_json(row, prettybool)
1861  */
1862 Datum
1863 array_to_json_pretty(PG_FUNCTION_ARGS)
1864 {
1865         Datum           array = PG_GETARG_DATUM(0);
1866         bool            use_line_feeds = PG_GETARG_BOOL(1);
1867         StringInfo      result;
1868
1869         result = makeStringInfo();
1870
1871         array_to_json_internal(array, result, use_line_feeds);
1872
1873         PG_RETURN_TEXT_P(cstring_to_text_with_len(result->data, result->len));
1874 }
1875
1876 /*
1877  * SQL function row_to_json(row)
1878  */
1879 Datum
1880 row_to_json(PG_FUNCTION_ARGS)
1881 {
1882         Datum           array = PG_GETARG_DATUM(0);
1883         StringInfo      result;
1884
1885         result = makeStringInfo();
1886
1887         composite_to_json(array, result, false);
1888
1889         PG_RETURN_TEXT_P(cstring_to_text_with_len(result->data, result->len));
1890 }
1891
1892 /*
1893  * SQL function row_to_json(row, prettybool)
1894  */
1895 Datum
1896 row_to_json_pretty(PG_FUNCTION_ARGS)
1897 {
1898         Datum           array = PG_GETARG_DATUM(0);
1899         bool            use_line_feeds = PG_GETARG_BOOL(1);
1900         StringInfo      result;
1901
1902         result = makeStringInfo();
1903
1904         composite_to_json(array, result, use_line_feeds);
1905
1906         PG_RETURN_TEXT_P(cstring_to_text_with_len(result->data, result->len));
1907 }
1908
1909 /*
1910  * SQL function to_json(anyvalue)
1911  */
1912 Datum
1913 to_json(PG_FUNCTION_ARGS)
1914 {
1915         Datum           val = PG_GETARG_DATUM(0);
1916         Oid                     val_type = get_fn_expr_argtype(fcinfo->flinfo, 0);
1917         StringInfo      result;
1918         JsonTypeCategory tcategory;
1919         Oid                     outfuncoid;
1920
1921         if (val_type == InvalidOid)
1922                 ereport(ERROR,
1923                                 (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1924                                  errmsg("could not determine input data type")));
1925
1926         json_categorize_type(val_type,
1927                                                  &tcategory, &outfuncoid);
1928
1929         result = makeStringInfo();
1930
1931         datum_to_json(val, false, result, tcategory, outfuncoid, false);
1932
1933         PG_RETURN_TEXT_P(cstring_to_text_with_len(result->data, result->len));
1934 }
1935
1936 /*
1937  * json_agg transition function
1938  *
1939  * aggregate input column as a json array value.
1940  */
1941 Datum
1942 json_agg_transfn(PG_FUNCTION_ARGS)
1943 {
1944         MemoryContext aggcontext,
1945                                 oldcontext;
1946         JsonAggState *state;
1947         Datum           val;
1948
1949         if (!AggCheckCallContext(fcinfo, &aggcontext))
1950         {
1951                 /* cannot be called directly because of internal-type argument */
1952                 elog(ERROR, "json_agg_transfn called in non-aggregate context");
1953         }
1954
1955         if (PG_ARGISNULL(0))
1956         {
1957                 Oid                     arg_type = get_fn_expr_argtype(fcinfo->flinfo, 1);
1958
1959                 if (arg_type == InvalidOid)
1960                         ereport(ERROR,
1961                                         (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1962                                          errmsg("could not determine input data type")));
1963
1964                 /*
1965                  * Make this state object in a context where it will persist for the
1966                  * duration of the aggregate call.  MemoryContextSwitchTo is only
1967                  * needed the first time, as the StringInfo routines make sure they
1968                  * use the right context to enlarge the object if necessary.
1969                  */
1970                 oldcontext = MemoryContextSwitchTo(aggcontext);
1971                 state = (JsonAggState *) palloc(sizeof(JsonAggState));
1972                 state->str = makeStringInfo();
1973                 MemoryContextSwitchTo(oldcontext);
1974
1975                 appendStringInfoChar(state->str, '[');
1976                 json_categorize_type(arg_type, &state->val_category,
1977                                                          &state->val_output_func);
1978         }
1979         else
1980         {
1981                 state = (JsonAggState *) PG_GETARG_POINTER(0);
1982                 appendStringInfoString(state->str, ", ");
1983         }
1984
1985         /* fast path for NULLs */
1986         if (PG_ARGISNULL(1))
1987         {
1988                 datum_to_json((Datum) 0, true, state->str, JSONTYPE_NULL,
1989                                           InvalidOid, false);
1990                 PG_RETURN_POINTER(state);
1991         }
1992
1993         val = PG_GETARG_DATUM(1);
1994
1995         /* add some whitespace if structured type and not first item */
1996         if (!PG_ARGISNULL(0) &&
1997                 (state->val_category == JSONTYPE_ARRAY ||
1998                  state->val_category == JSONTYPE_COMPOSITE))
1999         {
2000                 appendStringInfoString(state->str, "\n ");
2001         }
2002
2003         datum_to_json(val, false, state->str, state->val_category,
2004                                   state->val_output_func, false);
2005
2006         /*
2007          * The transition type for json_agg() is declared to be "internal", which
2008          * is a pass-by-value type the same size as a pointer.  So we can safely
2009          * pass the JsonAggState pointer through nodeAgg.c's machinations.
2010          */
2011         PG_RETURN_POINTER(state);
2012 }
2013
2014 /*
2015  * json_agg final function
2016  */
2017 Datum
2018 json_agg_finalfn(PG_FUNCTION_ARGS)
2019 {
2020         JsonAggState *state;
2021
2022         /* cannot be called directly because of internal-type argument */
2023         Assert(AggCheckCallContext(fcinfo, NULL));
2024
2025         state = PG_ARGISNULL(0) ?
2026                 NULL :
2027                 (JsonAggState *) PG_GETARG_POINTER(0);
2028
2029         /* NULL result for no rows in, as is standard with aggregates */
2030         if (state == NULL)
2031                 PG_RETURN_NULL();
2032
2033         /* Else return state with appropriate array terminator added */
2034         PG_RETURN_TEXT_P(catenate_stringinfo_string(state->str, "]"));
2035 }
2036
2037 /*
2038  * json_object_agg transition function.
2039  *
2040  * aggregate two input columns as a single json object value.
2041  */
2042 Datum
2043 json_object_agg_transfn(PG_FUNCTION_ARGS)
2044 {
2045         MemoryContext aggcontext,
2046                                 oldcontext;
2047         JsonAggState *state;
2048         Datum           arg;
2049
2050         if (!AggCheckCallContext(fcinfo, &aggcontext))
2051         {
2052                 /* cannot be called directly because of internal-type argument */
2053                 elog(ERROR, "json_object_agg_transfn called in non-aggregate context");
2054         }
2055
2056         if (PG_ARGISNULL(0))
2057         {
2058                 Oid                     arg_type;
2059
2060                 /*
2061                  * Make the StringInfo in a context where it will persist for the
2062                  * duration of the aggregate call. Switching context is only needed
2063                  * for this initial step, as the StringInfo routines make sure they
2064                  * use the right context to enlarge the object if necessary.
2065                  */
2066                 oldcontext = MemoryContextSwitchTo(aggcontext);
2067                 state = (JsonAggState *) palloc(sizeof(JsonAggState));
2068                 state->str = makeStringInfo();
2069                 MemoryContextSwitchTo(oldcontext);
2070
2071                 arg_type = get_fn_expr_argtype(fcinfo->flinfo, 1);
2072
2073                 if (arg_type == InvalidOid)
2074                         ereport(ERROR,
2075                                         (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
2076                                          errmsg("could not determine data type for argument %d", 1)));
2077
2078                 json_categorize_type(arg_type, &state->key_category,
2079                                                          &state->key_output_func);
2080
2081                 arg_type = get_fn_expr_argtype(fcinfo->flinfo, 2);
2082
2083                 if (arg_type == InvalidOid)
2084                         ereport(ERROR,
2085                                         (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
2086                                          errmsg("could not determine data type for argument %d", 2)));
2087
2088                 json_categorize_type(arg_type, &state->val_category,
2089                                                          &state->val_output_func);
2090
2091                 appendStringInfoString(state->str, "{ ");
2092         }
2093         else
2094         {
2095                 state = (JsonAggState *) PG_GETARG_POINTER(0);
2096                 appendStringInfoString(state->str, ", ");
2097         }
2098
2099         /*
2100          * Note: since json_object_agg() is declared as taking type "any", the
2101          * parser will not do any type conversion on unknown-type literals (that
2102          * is, undecorated strings or NULLs).  Such values will arrive here as
2103          * type UNKNOWN, which fortunately does not matter to us, since
2104          * unknownout() works fine.
2105          */
2106
2107         if (PG_ARGISNULL(1))
2108                 ereport(ERROR,
2109                                 (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
2110                                  errmsg("field name must not be null")));
2111
2112         arg = PG_GETARG_DATUM(1);
2113
2114         datum_to_json(arg, false, state->str, state->key_category,
2115                                   state->key_output_func, true);
2116
2117         appendStringInfoString(state->str, " : ");
2118
2119         if (PG_ARGISNULL(2))
2120                 arg = (Datum) 0;
2121         else
2122                 arg = PG_GETARG_DATUM(2);
2123
2124         datum_to_json(arg, PG_ARGISNULL(2), state->str, state->val_category,
2125                                   state->val_output_func, false);
2126
2127         PG_RETURN_POINTER(state);
2128 }
2129
2130 /*
2131  * json_object_agg final function.
2132  */
2133 Datum
2134 json_object_agg_finalfn(PG_FUNCTION_ARGS)
2135 {
2136         JsonAggState *state;
2137
2138         /* cannot be called directly because of internal-type argument */
2139         Assert(AggCheckCallContext(fcinfo, NULL));
2140
2141         state = PG_ARGISNULL(0) ? NULL : (JsonAggState *) PG_GETARG_POINTER(0);
2142
2143         /* NULL result for no rows in, as is standard with aggregates */
2144         if (state == NULL)
2145                 PG_RETURN_NULL();
2146
2147         /* Else return state with appropriate object terminator added */
2148         PG_RETURN_TEXT_P(catenate_stringinfo_string(state->str, " }"));
2149 }
2150
2151 /*
2152  * Helper function for aggregates: return given StringInfo's contents plus
2153  * specified trailing string, as a text datum.  We need this because aggregate
2154  * final functions are not allowed to modify the aggregate state.
2155  */
2156 static text *
2157 catenate_stringinfo_string(StringInfo buffer, const char *addon)
2158 {
2159         /* custom version of cstring_to_text_with_len */
2160         int                     buflen = buffer->len;
2161         int                     addlen = strlen(addon);
2162         text       *result = (text *) palloc(buflen + addlen + VARHDRSZ);
2163
2164         SET_VARSIZE(result, buflen + addlen + VARHDRSZ);
2165         memcpy(VARDATA(result), buffer->data, buflen);
2166         memcpy(VARDATA(result) + buflen, addon, addlen);
2167
2168         return result;
2169 }
2170
2171 /*
2172  * SQL function json_build_object(variadic "any")
2173  */
2174 Datum
2175 json_build_object(PG_FUNCTION_ARGS)
2176 {
2177         int                     nargs = PG_NARGS();
2178         int                     i;
2179         const char *sep = "";
2180         StringInfo      result;
2181         Datum      *args;
2182         bool       *nulls;
2183         Oid                *types;
2184
2185         /* fetch argument values to build the object */
2186         nargs = extract_variadic_args(fcinfo, 0, false, &args, &types, &nulls);
2187
2188         if (nargs < 0)
2189                 PG_RETURN_NULL();
2190
2191         if (nargs % 2 != 0)
2192                 ereport(ERROR,
2193                                 (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
2194                                  errmsg("argument list must have even number of elements"),
2195                                  /* translator: %s is a SQL function name */
2196                                  errhint("The arguments of %s must consist of alternating keys and values.",
2197                                                  "json_build_object()")));
2198
2199         result = makeStringInfo();
2200
2201         appendStringInfoChar(result, '{');
2202
2203         for (i = 0; i < nargs; i += 2)
2204         {
2205                 appendStringInfoString(result, sep);
2206                 sep = ", ";
2207
2208                 /* process key */
2209                 if (nulls[i])
2210                         ereport(ERROR,
2211                                         (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
2212                                          errmsg("argument %d cannot be null", i + 1),
2213                                          errhint("Object keys should be text.")));
2214
2215                 add_json(args[i], false, result, types[i], true);
2216
2217                 appendStringInfoString(result, " : ");
2218
2219                 /* process value */
2220                 add_json(args[i + 1], nulls[i + 1], result, types[i + 1], false);
2221         }
2222
2223         appendStringInfoChar(result, '}');
2224
2225         PG_RETURN_TEXT_P(cstring_to_text_with_len(result->data, result->len));
2226 }
2227
2228 /*
2229  * degenerate case of json_build_object where it gets 0 arguments.
2230  */
2231 Datum
2232 json_build_object_noargs(PG_FUNCTION_ARGS)
2233 {
2234         PG_RETURN_TEXT_P(cstring_to_text_with_len("{}", 2));
2235 }
2236
2237 /*
2238  * SQL function json_build_array(variadic "any")
2239  */
2240 Datum
2241 json_build_array(PG_FUNCTION_ARGS)
2242 {
2243         int                     nargs;
2244         int                     i;
2245         const char *sep = "";
2246         StringInfo      result;
2247         Datum      *args;
2248         bool       *nulls;
2249         Oid                *types;
2250
2251         /* fetch argument values to build the array */
2252         nargs = extract_variadic_args(fcinfo, 0, false, &args, &types, &nulls);
2253
2254         if (nargs < 0)
2255                 PG_RETURN_NULL();
2256
2257         result = makeStringInfo();
2258
2259         appendStringInfoChar(result, '[');
2260
2261         for (i = 0; i < nargs; i++)
2262         {
2263                 appendStringInfoString(result, sep);
2264                 sep = ", ";
2265                 add_json(args[i], nulls[i], result, types[i], false);
2266         }
2267
2268         appendStringInfoChar(result, ']');
2269
2270         PG_RETURN_TEXT_P(cstring_to_text_with_len(result->data, result->len));
2271 }
2272
2273 /*
2274  * degenerate case of json_build_array where it gets 0 arguments.
2275  */
2276 Datum
2277 json_build_array_noargs(PG_FUNCTION_ARGS)
2278 {
2279         PG_RETURN_TEXT_P(cstring_to_text_with_len("[]", 2));
2280 }
2281
2282 /*
2283  * SQL function json_object(text[])
2284  *
2285  * take a one or two dimensional array of text as key/value pairs
2286  * for a json object.
2287  */
2288 Datum
2289 json_object(PG_FUNCTION_ARGS)
2290 {
2291         ArrayType  *in_array = PG_GETARG_ARRAYTYPE_P(0);
2292         int                     ndims = ARR_NDIM(in_array);
2293         StringInfoData result;
2294         Datum      *in_datums;
2295         bool       *in_nulls;
2296         int                     in_count,
2297                                 count,
2298                                 i;
2299         text       *rval;
2300         char       *v;
2301
2302         switch (ndims)
2303         {
2304                 case 0:
2305                         PG_RETURN_DATUM(CStringGetTextDatum("{}"));
2306                         break;
2307
2308                 case 1:
2309                         if ((ARR_DIMS(in_array)[0]) % 2)
2310                                 ereport(ERROR,
2311                                                 (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
2312                                                  errmsg("array must have even number of elements")));
2313                         break;
2314
2315                 case 2:
2316                         if ((ARR_DIMS(in_array)[1]) != 2)
2317                                 ereport(ERROR,
2318                                                 (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
2319                                                  errmsg("array must have two columns")));
2320                         break;
2321
2322                 default:
2323                         ereport(ERROR,
2324                                         (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
2325                                          errmsg("wrong number of array subscripts")));
2326         }
2327
2328         deconstruct_array(in_array,
2329                                           TEXTOID, -1, false, 'i',
2330                                           &in_datums, &in_nulls, &in_count);
2331
2332         count = in_count / 2;
2333
2334         initStringInfo(&result);
2335
2336         appendStringInfoChar(&result, '{');
2337
2338         for (i = 0; i < count; ++i)
2339         {
2340                 if (in_nulls[i * 2])
2341                         ereport(ERROR,
2342                                         (errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED),
2343                                          errmsg("null value not allowed for object key")));
2344
2345                 v = TextDatumGetCString(in_datums[i * 2]);
2346                 if (i > 0)
2347                         appendStringInfoString(&result, ", ");
2348                 escape_json(&result, v);
2349                 appendStringInfoString(&result, " : ");
2350                 pfree(v);
2351                 if (in_nulls[i * 2 + 1])
2352                         appendStringInfoString(&result, "null");
2353                 else
2354                 {
2355                         v = TextDatumGetCString(in_datums[i * 2 + 1]);
2356                         escape_json(&result, v);
2357                         pfree(v);
2358                 }
2359         }
2360
2361         appendStringInfoChar(&result, '}');
2362
2363         pfree(in_datums);
2364         pfree(in_nulls);
2365
2366         rval = cstring_to_text_with_len(result.data, result.len);
2367         pfree(result.data);
2368
2369         PG_RETURN_TEXT_P(rval);
2370
2371 }
2372
2373 /*
2374  * SQL function json_object(text[], text[])
2375  *
2376  * take separate key and value arrays of text to construct a json object
2377  * pairwise.
2378  */
2379 Datum
2380 json_object_two_arg(PG_FUNCTION_ARGS)
2381 {
2382         ArrayType  *key_array = PG_GETARG_ARRAYTYPE_P(0);
2383         ArrayType  *val_array = PG_GETARG_ARRAYTYPE_P(1);
2384         int                     nkdims = ARR_NDIM(key_array);
2385         int                     nvdims = ARR_NDIM(val_array);
2386         StringInfoData result;
2387         Datum      *key_datums,
2388                            *val_datums;
2389         bool       *key_nulls,
2390                            *val_nulls;
2391         int                     key_count,
2392                                 val_count,
2393                                 i;
2394         text       *rval;
2395         char       *v;
2396
2397         if (nkdims > 1 || nkdims != nvdims)
2398                 ereport(ERROR,
2399                                 (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
2400                                  errmsg("wrong number of array subscripts")));
2401
2402         if (nkdims == 0)
2403                 PG_RETURN_DATUM(CStringGetTextDatum("{}"));
2404
2405         deconstruct_array(key_array,
2406                                           TEXTOID, -1, false, 'i',
2407                                           &key_datums, &key_nulls, &key_count);
2408
2409         deconstruct_array(val_array,
2410                                           TEXTOID, -1, false, 'i',
2411                                           &val_datums, &val_nulls, &val_count);
2412
2413         if (key_count != val_count)
2414                 ereport(ERROR,
2415                                 (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
2416                                  errmsg("mismatched array dimensions")));
2417
2418         initStringInfo(&result);
2419
2420         appendStringInfoChar(&result, '{');
2421
2422         for (i = 0; i < key_count; ++i)
2423         {
2424                 if (key_nulls[i])
2425                         ereport(ERROR,
2426                                         (errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED),
2427                                          errmsg("null value not allowed for object key")));
2428
2429                 v = TextDatumGetCString(key_datums[i]);
2430                 if (i > 0)
2431                         appendStringInfoString(&result, ", ");
2432                 escape_json(&result, v);
2433                 appendStringInfoString(&result, " : ");
2434                 pfree(v);
2435                 if (val_nulls[i])
2436                         appendStringInfoString(&result, "null");
2437                 else
2438                 {
2439                         v = TextDatumGetCString(val_datums[i]);
2440                         escape_json(&result, v);
2441                         pfree(v);
2442                 }
2443         }
2444
2445         appendStringInfoChar(&result, '}');
2446
2447         pfree(key_datums);
2448         pfree(key_nulls);
2449         pfree(val_datums);
2450         pfree(val_nulls);
2451
2452         rval = cstring_to_text_with_len(result.data, result.len);
2453         pfree(result.data);
2454
2455         PG_RETURN_TEXT_P(rval);
2456 }
2457
2458
2459 /*
2460  * Produce a JSON string literal, properly escaping characters in the text.
2461  */
2462 void
2463 escape_json(StringInfo buf, const char *str)
2464 {
2465         const char *p;
2466
2467         appendStringInfoCharMacro(buf, '"');
2468         for (p = str; *p; p++)
2469         {
2470                 switch (*p)
2471                 {
2472                         case '\b':
2473                                 appendStringInfoString(buf, "\\b");
2474                                 break;
2475                         case '\f':
2476                                 appendStringInfoString(buf, "\\f");
2477                                 break;
2478                         case '\n':
2479                                 appendStringInfoString(buf, "\\n");
2480                                 break;
2481                         case '\r':
2482                                 appendStringInfoString(buf, "\\r");
2483                                 break;
2484                         case '\t':
2485                                 appendStringInfoString(buf, "\\t");
2486                                 break;
2487                         case '"':
2488                                 appendStringInfoString(buf, "\\\"");
2489                                 break;
2490                         case '\\':
2491                                 appendStringInfoString(buf, "\\\\");
2492                                 break;
2493                         default:
2494                                 if ((unsigned char) *p < ' ')
2495                                         appendStringInfo(buf, "\\u%04x", (int) *p);
2496                                 else
2497                                         appendStringInfoCharMacro(buf, *p);
2498                                 break;
2499                 }
2500         }
2501         appendStringInfoCharMacro(buf, '"');
2502 }
2503
2504 /*
2505  * SQL function json_typeof(json) -> text
2506  *
2507  * Returns the type of the outermost JSON value as TEXT.  Possible types are
2508  * "object", "array", "string", "number", "boolean", and "null".
2509  *
2510  * Performs a single call to json_lex() to get the first token of the supplied
2511  * value.  This initial token uniquely determines the value's type.  As our
2512  * input must already have been validated by json_in() or json_recv(), the
2513  * initial token should never be JSON_TOKEN_OBJECT_END, JSON_TOKEN_ARRAY_END,
2514  * JSON_TOKEN_COLON, JSON_TOKEN_COMMA, or JSON_TOKEN_END.
2515  */
2516 Datum
2517 json_typeof(PG_FUNCTION_ARGS)
2518 {
2519         text       *json;
2520
2521         JsonLexContext *lex;
2522         JsonTokenType tok;
2523         char       *type;
2524
2525         json = PG_GETARG_TEXT_PP(0);
2526         lex = makeJsonLexContext(json, false);
2527
2528         /* Lex exactly one token from the input and check its type. */
2529         json_lex(lex);
2530         tok = lex_peek(lex);
2531         switch (tok)
2532         {
2533                 case JSON_TOKEN_OBJECT_START:
2534                         type = "object";
2535                         break;
2536                 case JSON_TOKEN_ARRAY_START:
2537                         type = "array";
2538                         break;
2539                 case JSON_TOKEN_STRING:
2540                         type = "string";
2541                         break;
2542                 case JSON_TOKEN_NUMBER:
2543                         type = "number";
2544                         break;
2545                 case JSON_TOKEN_TRUE:
2546                 case JSON_TOKEN_FALSE:
2547                         type = "boolean";
2548                         break;
2549                 case JSON_TOKEN_NULL:
2550                         type = "null";
2551                         break;
2552                 default:
2553                         elog(ERROR, "unexpected json token: %d", tok);
2554         }
2555
2556         PG_RETURN_TEXT_P(cstring_to_text(type));
2557 }