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) 1996-2014, 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_cast.h"
  19 #include "catalog/pg_type.h"
  20 #include "executor/spi.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 static inline void json_lex(JsonLexContext *lex);
  70 static inline void json_lex_string(JsonLexContext *lex);
  71 static inline void json_lex_number(JsonLexContext *lex, char *s, bool *num_err);
  72 static inline void parse_scalar(JsonLexContext *lex, JsonSemAction *sem);
  73 static void parse_object_field(JsonLexContext *lex, JsonSemAction *sem);
  74 static void parse_object(JsonLexContext *lex, JsonSemAction *sem);
  75 static void parse_array_element(JsonLexContext *lex, JsonSemAction *sem);
  76 static void parse_array(JsonLexContext *lex, JsonSemAction *sem);
  77 static void report_parse_error(JsonParseContext ctx, JsonLexContext *lex);
  78 static void report_invalid_token(JsonLexContext *lex);
  79 static int      report_json_context(JsonLexContext *lex);
  80 static char *extract_mb_char(char *s);
  81 static void composite_to_json(Datum composite, StringInfo result,
  82                                   bool use_line_feeds);
  83 static void array_dim_to_json(StringInfo result, int dim, int ndims, int *dims,
  84                                   Datum *vals, bool *nulls, int *valcount,
  85                                   JsonTypeCategory tcategory, Oid outfuncoid,
  86                                   bool use_line_feeds);
  87 static void array_to_json_internal(Datum array, StringInfo result,
  88                                            bool use_line_feeds);
  89 static void json_categorize_type(Oid typoid,
  90                                          JsonTypeCategory *tcategory,
  91                                          Oid *outfuncoid);
  92 static void datum_to_json(Datum val, bool is_null, StringInfo result,
  93                           JsonTypeCategory tcategory, Oid outfuncoid,
  94                           bool key_scalar);
  95 static void add_json(Datum val, bool is_null, StringInfo result,
  96                  Oid val_type, bool key_scalar);
  97 static text *catenate_stringinfo_string(StringInfo buffer, const char *addon);
  98
  99 /* the null action object used for pure validation */
 100 static JsonSemAction nullSemAction =
 101 {
 102         NULL, NULL, NULL, NULL, NULL,
 103         NULL, NULL, NULL, NULL, NULL
 104 };
 105
 106 /* Recursive Descent parser support routines */
 107
 108 /*
 109  * lex_peek
 110  *
 111  * what is the current look_ahead token?
 112 */
 113 static inline JsonTokenType
 114 lex_peek(JsonLexContext *lex)
 115 {
 116         return lex->token_type;
 117 }
 118
 119 /*
 120  * lex_accept
 121  *
 122  * accept the look_ahead token and move the lexer to the next token if the
 123  * look_ahead token matches the token parameter. In that case, and if required,
 124  * also hand back the de-escaped lexeme.
 125  *
 126  * returns true if the token matched, false otherwise.
 127  */
 128 static inline bool
 129 lex_accept(JsonLexContext *lex, JsonTokenType token, char **lexeme)
 130 {
 131         if (lex->token_type == token)
 132         {
 133                 if (lexeme != NULL)
 134                 {
 135                         if (lex->token_type == JSON_TOKEN_STRING)
 136                         {
 137                                 if (lex->strval != NULL)
 138                                         *lexeme = pstrdup(lex->strval->data);
 139                         }
 140                         else
 141                         {
 142                                 int                     len = (lex->token_terminator - lex->token_start);
 143                                 char       *tokstr = palloc(len + 1);
 144
 145                                 memcpy(tokstr, lex->token_start, len);
 146                                 tokstr[len] = '\0';
 147                                 *lexeme = tokstr;
 148                         }
 149                 }
 150                 json_lex(lex);
 151                 return true;
 152         }
 153         return false;
 154 }
 155
 156 /*
 157  * lex_accept
 158  *
 159  * move the lexer to the next token if the current look_ahead token matches
 160  * the parameter token. Otherwise, report an error.
 161  */
 162 static inline void
 163 lex_expect(JsonParseContext ctx, JsonLexContext *lex, JsonTokenType token)
 164 {
 165         if (!lex_accept(lex, token, NULL))
 166                 report_parse_error(ctx, lex);;
 167 }
 168
 169 /* chars to consider as part of an alphanumeric token */
 170 #define JSON_ALPHANUMERIC_CHAR(c)  \
 171         (((c) >= 'a' && (c) <= 'z') || \
 172          ((c) >= 'A' && (c) <= 'Z') || \
 173          ((c) >= '0' && (c) <= '9') || \
 174          (c) == '_' || \
 175          IS_HIGHBIT_SET(c))
 176
 177 /* utility function to check if a string is a valid JSON number */
 178 extern bool
 179 IsValidJsonNumber(const char *str, int len)
 180 {
 181         bool            numeric_error;
 182         JsonLexContext dummy_lex;
 183
 184
 185         /*
 186          * json_lex_number expects a leading  '-' to have been eaten already.
 187          *
 188          * having to cast away the constness of str is ugly, but there's not much
 189          * easy alternative.
 190          */
 191         if (*str == '-')
 192         {
 193                 dummy_lex.input = (char *) str + 1;
 194                 dummy_lex.input_length = len - 1;
 195         }
 196         else
 197         {
 198                 dummy_lex.input = (char *) str;
 199                 dummy_lex.input_length = len;
 200         }
 201
 202         json_lex_number(&dummy_lex, dummy_lex.input, &numeric_error);
 203
 204         return !numeric_error;
 205 }
 206
 207 /*
 208  * Input.
 209  */
 210 Datum
 211 json_in(PG_FUNCTION_ARGS)
 212 {
 213         char       *json = PG_GETARG_CSTRING(0);
 214         text       *result = cstring_to_text(json);
 215         JsonLexContext *lex;
 216
 217         /* validate it */
 218         lex = makeJsonLexContext(result, false);
 219         pg_parse_json(lex, &nullSemAction);
 220
 221         /* Internal representation is the same as text, for now */
 222         PG_RETURN_TEXT_P(result);
 223 }
 224
 225 /*
 226  * Output.
 227  */
 228 Datum
 229 json_out(PG_FUNCTION_ARGS)
 230 {
 231         /* we needn't detoast because text_to_cstring will handle that */
 232         Datum           txt = PG_GETARG_DATUM(0);
 233
 234         PG_RETURN_CSTRING(TextDatumGetCString(txt));
 235 }
 236
 237 /*
 238  * Binary send.
 239  */
 240 Datum
 241 json_send(PG_FUNCTION_ARGS)
 242 {
 243         text       *t = PG_GETARG_TEXT_PP(0);
 244         StringInfoData buf;
 245
 246         pq_begintypsend(&buf);
 247         pq_sendtext(&buf, VARDATA_ANY(t), VARSIZE_ANY_EXHDR(t));
 248         PG_RETURN_BYTEA_P(pq_endtypsend(&buf));
 249 }
 250
 251 /*
 252  * Binary receive.
 253  */
 254 Datum
 255 json_recv(PG_FUNCTION_ARGS)
 256 {
 257         StringInfo      buf = (StringInfo) PG_GETARG_POINTER(0);
 258         char       *str;
 259         int                     nbytes;
 260         JsonLexContext *lex;
 261
 262         str = pq_getmsgtext(buf, buf->len - buf->cursor, &nbytes);
 263
 264         /* Validate it. */
 265         lex = makeJsonLexContextCstringLen(str, nbytes, false);
 266         pg_parse_json(lex, &nullSemAction);
 267
 268         PG_RETURN_TEXT_P(cstring_to_text_with_len(str, nbytes));
 269 }
 270
 271 /*
 272  * makeJsonLexContext
 273  *
 274  * lex constructor, with or without StringInfo object
 275  * for de-escaped lexemes.
 276  *
 277  * Without is better as it makes the processing faster, so only make one
 278  * if really required.
 279  *
 280  * If you already have the json as a text* value, use the first of these
 281  * functions, otherwise use  makeJsonLexContextCstringLen().
 282  */
 283 JsonLexContext *
 284 makeJsonLexContext(text *json, bool need_escapes)
 285 {
 286         return makeJsonLexContextCstringLen(VARDATA(json),
 287                                                                                 VARSIZE(json) - VARHDRSZ,
 288                                                                                 need_escapes);
 289 }
 290
 291 JsonLexContext *
 292 makeJsonLexContextCstringLen(char *json, int len, bool need_escapes)
 293 {
 294         JsonLexContext *lex = palloc0(sizeof(JsonLexContext));
 295
 296         lex->input = lex->token_terminator = lex->line_start = json;
 297         lex->line_number = 1;
 298         lex->input_length = len;
 299         if (need_escapes)
 300                 lex->strval = makeStringInfo();
 301         return lex;
 302 }
 303
 304 /*
 305  * pg_parse_json
 306  *
 307  * Publicly visible entry point for the JSON parser.
 308  *
 309  * lex is a lexing context, set up for the json to be processed by calling
 310  * makeJsonLexContext(). sem is a strucure of function pointers to semantic
 311  * action routines to be called at appropriate spots during parsing, and a
 312  * pointer to a state object to be passed to those routines.
 313  */
 314 void
 315 pg_parse_json(JsonLexContext *lex, JsonSemAction *sem)
 316 {
 317         JsonTokenType tok;
 318
 319         /* get the initial token */
 320         json_lex(lex);
 321
 322         tok = lex_peek(lex);
 323
 324         /* parse by recursive descent */
 325         switch (tok)
 326         {
 327                 case JSON_TOKEN_OBJECT_START:
 328                         parse_object(lex, sem);
 329                         break;
 330                 case JSON_TOKEN_ARRAY_START:
 331                         parse_array(lex, sem);
 332                         break;
 333                 default:
 334                         parse_scalar(lex, sem);         /* json can be a bare scalar */
 335         }
 336
 337         lex_expect(JSON_PARSE_END, lex, JSON_TOKEN_END);
 338
 339 }
 340
 341 /*
 342  *      Recursive Descent parse routines. There is one for each structural
 343  *      element in a json document:
 344  *        - scalar (string, number, true, false, null)
 345  *        - array  ( [ ] )
 346  *        - array element
 347  *        - object ( { } )
 348  *        - object field
 349  */
 350 static inline void
 351 parse_scalar(JsonLexContext *lex, JsonSemAction *sem)
 352 {
 353         char       *val = NULL;
 354         json_scalar_action sfunc = sem->scalar;
 355         char      **valaddr;
 356         JsonTokenType tok = lex_peek(lex);
 357
 358         valaddr = sfunc == NULL ? NULL : &val;
 359
 360         /* a scalar must be a string, a number, true, false, or null */
 361         switch (tok)
 362         {
 363                 case JSON_TOKEN_TRUE:
 364                         lex_accept(lex, JSON_TOKEN_TRUE, valaddr);
 365                         break;
 366                 case JSON_TOKEN_FALSE:
 367                         lex_accept(lex, JSON_TOKEN_FALSE, valaddr);
 368                         break;
 369                 case JSON_TOKEN_NULL:
 370                         lex_accept(lex, JSON_TOKEN_NULL, valaddr);
 371                         break;
 372                 case JSON_TOKEN_NUMBER:
 373                         lex_accept(lex, JSON_TOKEN_NUMBER, valaddr);
 374                         break;
 375                 case JSON_TOKEN_STRING:
 376                         lex_accept(lex, JSON_TOKEN_STRING, valaddr);
 377                         break;
 378                 default:
 379                         report_parse_error(JSON_PARSE_VALUE, lex);
 380         }
 381
 382         if (sfunc != NULL)
 383                 (*sfunc) (sem->semstate, val, tok);
 384 }
 385
 386 static void
 387 parse_object_field(JsonLexContext *lex, JsonSemAction *sem)
 388 {
 389         /*
 390          * An object field is "fieldname" : value where value can be a scalar,
 391          * object or array.  Note: in user-facing docs and error messages, we
 392          * generally call a field name a "key".
 393          */
 394
 395         char       *fname = NULL;       /* keep compiler quiet */
 396         json_ofield_action ostart = sem->object_field_start;
 397         json_ofield_action oend = sem->object_field_end;
 398         bool            isnull;
 399         char      **fnameaddr = NULL;
 400         JsonTokenType tok;
 401
 402         if (ostart != NULL || oend != NULL)
 403                 fnameaddr = &fname;
 404
 405         if (!lex_accept(lex, JSON_TOKEN_STRING, fnameaddr))
 406                 report_parse_error(JSON_PARSE_STRING, lex);
 407
 408         lex_expect(JSON_PARSE_OBJECT_LABEL, lex, JSON_TOKEN_COLON);
 409
 410         tok = lex_peek(lex);
 411         isnull = tok == JSON_TOKEN_NULL;
 412
 413         if (ostart != NULL)
 414                 (*ostart) (sem->semstate, fname, isnull);
 415
 416         switch (tok)
 417         {
 418                 case JSON_TOKEN_OBJECT_START:
 419                         parse_object(lex, sem);
 420                         break;
 421                 case JSON_TOKEN_ARRAY_START:
 422                         parse_array(lex, sem);
 423                         break;
 424                 default:
 425                         parse_scalar(lex, sem);
 426         }
 427
 428         if (oend != NULL)
 429                 (*oend) (sem->semstate, fname, isnull);
 430 }
 431
 432 static void
 433 parse_object(JsonLexContext *lex, JsonSemAction *sem)
 434 {
 435         /*
 436          * an object is a possibly empty sequence of object fields, separated by
 437          * commas and surrounded by curly braces.
 438          */
 439         json_struct_action ostart = sem->object_start;
 440         json_struct_action oend = sem->object_end;
 441         JsonTokenType tok;
 442
 443         if (ostart != NULL)
 444                 (*ostart) (sem->semstate);
 445
 446         /*
 447          * Data inside an object is at a higher nesting level than the object
 448          * itself. Note that we increment this after we call the semantic routine
 449          * for the object start and restore it before we call the routine for the
 450          * object end.
 451          */
 452         lex->lex_level++;
 453
 454         /* we know this will succeeed, just clearing the token */
 455         lex_expect(JSON_PARSE_OBJECT_START, lex, JSON_TOKEN_OBJECT_START);
 456
 457         tok = lex_peek(lex);
 458         switch (tok)
 459         {
 460                 case JSON_TOKEN_STRING:
 461                         parse_object_field(lex, sem);
 462                         while (lex_accept(lex, JSON_TOKEN_COMMA, NULL))
 463                                 parse_object_field(lex, sem);
 464                         break;
 465                 case JSON_TOKEN_OBJECT_END:
 466                         break;
 467                 default:
 468                         /* case of an invalid initial token inside the object */
 469                         report_parse_error(JSON_PARSE_OBJECT_START, lex);
 470         }
 471
 472         lex_expect(JSON_PARSE_OBJECT_NEXT, lex, JSON_TOKEN_OBJECT_END);
 473
 474         lex->lex_level--;
 475
 476         if (oend != NULL)
 477                 (*oend) (sem->semstate);
 478 }
 479
 480 static void
 481 parse_array_element(JsonLexContext *lex, JsonSemAction *sem)
 482 {
 483         json_aelem_action astart = sem->array_element_start;
 484         json_aelem_action aend = sem->array_element_end;
 485         JsonTokenType tok = lex_peek(lex);
 486
 487         bool            isnull;
 488
 489         isnull = tok == JSON_TOKEN_NULL;
 490
 491         if (astart != NULL)
 492                 (*astart) (sem->semstate, isnull);
 493
 494         /* an array element is any object, array or scalar */
 495         switch (tok)
 496         {
 497                 case JSON_TOKEN_OBJECT_START:
 498                         parse_object(lex, sem);
 499                         break;
 500                 case JSON_TOKEN_ARRAY_START:
 501                         parse_array(lex, sem);
 502                         break;
 503                 default:
 504                         parse_scalar(lex, sem);
 505         }
 506
 507         if (aend != NULL)
 508                 (*aend) (sem->semstate, isnull);
 509 }
 510
 511 static void
 512 parse_array(JsonLexContext *lex, JsonSemAction *sem)
 513 {
 514         /*
 515          * an array is a possibly empty sequence of array elements, separated by
 516          * commas and surrounded by square brackets.
 517          */
 518         json_struct_action astart = sem->array_start;
 519         json_struct_action aend = sem->array_end;
 520
 521         if (astart != NULL)
 522                 (*astart) (sem->semstate);
 523
 524         /*
 525          * Data inside an array is at a higher nesting level than the array
 526          * itself. Note that we increment this after we call the semantic routine
 527          * for the array start and restore it before we call the routine for the
 528          * array end.
 529          */
 530         lex->lex_level++;
 531
 532         lex_expect(JSON_PARSE_ARRAY_START, lex, JSON_TOKEN_ARRAY_START);
 533         if (lex_peek(lex) != JSON_TOKEN_ARRAY_END)
 534         {
 535
 536                 parse_array_element(lex, sem);
 537
 538                 while (lex_accept(lex, JSON_TOKEN_COMMA, NULL))
 539                         parse_array_element(lex, sem);
 540         }
 541
 542         lex_expect(JSON_PARSE_ARRAY_NEXT, lex, JSON_TOKEN_ARRAY_END);
 543
 544         lex->lex_level--;
 545
 546         if (aend != NULL)
 547                 (*aend) (sem->semstate);
 548 }
 549
 550 /*
 551  * Lex one token from the input stream.
 552  */
 553 static inline void
 554 json_lex(JsonLexContext *lex)
 555 {
 556         char       *s;
 557         int                     len;
 558
 559         /* Skip leading whitespace. */
 560         s = lex->token_terminator;
 561         len = s - lex->input;
 562         while (len < lex->input_length &&
 563                    (*s == ' ' || *s == '\t' || *s == '\n' || *s == '\r'))
 564         {
 565                 if (*s == '\n')
 566                         ++lex->line_number;
 567                 ++s;
 568                 ++len;
 569         }
 570         lex->token_start = s;
 571
 572         /* Determine token type. */
 573         if (len >= lex->input_length)
 574         {
 575                 lex->token_start = NULL;
 576                 lex->prev_token_terminator = lex->token_terminator;
 577                 lex->token_terminator = s;
 578                 lex->token_type = JSON_TOKEN_END;
 579         }
 580         else
 581                 switch (*s)
 582                 {
 583                                 /* Single-character token, some kind of punctuation mark. */
 584                         case '{':
 585                                 lex->prev_token_terminator = lex->token_terminator;
 586                                 lex->token_terminator = s + 1;
 587                                 lex->token_type = JSON_TOKEN_OBJECT_START;
 588                                 break;
 589                         case '}':
 590                                 lex->prev_token_terminator = lex->token_terminator;
 591                                 lex->token_terminator = s + 1;
 592                                 lex->token_type = JSON_TOKEN_OBJECT_END;
 593                                 break;
 594                         case '[':
 595                                 lex->prev_token_terminator = lex->token_terminator;
 596                                 lex->token_terminator = s + 1;
 597                                 lex->token_type = JSON_TOKEN_ARRAY_START;
 598                                 break;
 599                         case ']':
 600                                 lex->prev_token_terminator = lex->token_terminator;
 601                                 lex->token_terminator = s + 1;
 602                                 lex->token_type = JSON_TOKEN_ARRAY_END;
 603                                 break;
 604                         case ',':
 605                                 lex->prev_token_terminator = lex->token_terminator;
 606                                 lex->token_terminator = s + 1;
 607                                 lex->token_type = JSON_TOKEN_COMMA;
 608                                 break;
 609                         case ':':
 610                                 lex->prev_token_terminator = lex->token_terminator;
 611                                 lex->token_terminator = s + 1;
 612                                 lex->token_type = JSON_TOKEN_COLON;
 613                                 break;
 614                         case '"':
 615                                 /* string */
 616                                 json_lex_string(lex);
 617                                 lex->token_type = JSON_TOKEN_STRING;
 618                                 break;
 619                         case '-':
 620                                 /* Negative number. */
 621                                 json_lex_number(lex, s + 1, NULL);
 622                                 lex->token_type = JSON_TOKEN_NUMBER;
 623                                 break;
 624                         case '0':
 625                         case '1':
 626                         case '2':
 627                         case '3':
 628                         case '4':
 629                         case '5':
 630                         case '6':
 631                         case '7':
 632                         case '8':
 633                         case '9':
 634                                 /* Positive number. */
 635                                 json_lex_number(lex, s, NULL);
 636                                 lex->token_type = JSON_TOKEN_NUMBER;
 637                                 break;
 638                         default:
 639                                 {
 640                                         char       *p;
 641
 642                                         /*
 643                                          * We're not dealing with a string, number, legal
 644                                          * punctuation mark, or end of string.  The only legal
 645                                          * tokens we might find here are true, false, and null,
 646                                          * but for error reporting purposes we scan until we see a
 647                                          * non-alphanumeric character.  That way, we can report
 648                                          * the whole word as an unexpected token, rather than just
 649                                          * some unintuitive prefix thereof.
 650                                          */
 651                                         for (p = s; p - s < lex->input_length - len && JSON_ALPHANUMERIC_CHAR(*p); p++)
 652                                                  /* skip */ ;
 653
 654                                         /*
 655                                          * We got some sort of unexpected punctuation or an
 656                                          * otherwise unexpected character, so just complain about
 657                                          * that one character.
 658                                          */
 659                                         if (p == s)
 660                                         {
 661                                                 lex->prev_token_terminator = lex->token_terminator;
 662                                                 lex->token_terminator = s + 1;
 663                                                 report_invalid_token(lex);
 664                                         }
 665
 666                                         /*
 667                                          * We've got a real alphanumeric token here.  If it
 668                                          * happens to be true, false, or null, all is well.  If
 669                                          * not, error out.
 670                                          */
 671                                         lex->prev_token_terminator = lex->token_terminator;
 672                                         lex->token_terminator = p;
 673                                         if (p - s == 4)
 674                                         {
 675                                                 if (memcmp(s, "true", 4) == 0)
 676                                                         lex->token_type = JSON_TOKEN_TRUE;
 677                                                 else if (memcmp(s, "null", 4) == 0)
 678                                                         lex->token_type = JSON_TOKEN_NULL;
 679                                                 else
 680                                                         report_invalid_token(lex);
 681                                         }
 682                                         else if (p - s == 5 && memcmp(s, "false", 5) == 0)
 683                                                 lex->token_type = JSON_TOKEN_FALSE;
 684                                         else
 685                                                 report_invalid_token(lex);
 686
 687                                 }
 688                 }                                               /* end of switch */
 689 }
 690
 691 /*
 692  * The next token in the input stream is known to be a string; lex it.
 693  */
 694 static inline void
 695 json_lex_string(JsonLexContext *lex)
 696 {
 697         char       *s;
 698         int                     len;
 699         int                     hi_surrogate = -1;
 700
 701         if (lex->strval != NULL)
 702                 resetStringInfo(lex->strval);
 703
 704         Assert(lex->input_length > 0);
 705         s = lex->token_start;
 706         len = lex->token_start - lex->input;
 707         for (;;)
 708         {
 709                 s++;
 710                 len++;
 711                 /* Premature end of the string. */
 712                 if (len >= lex->input_length)
 713                 {
 714                         lex->token_terminator = s;
 715                         report_invalid_token(lex);
 716                 }
 717                 else if (*s == '"')
 718                         break;
 719                 else if ((unsigned char) *s < 32)
 720                 {
 721                         /* Per RFC4627, these characters MUST be escaped. */
 722                         /* Since *s isn't printable, exclude it from the context string */
 723                         lex->token_terminator = s;
 724                         ereport(ERROR,
 725                                         (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
 726                                          errmsg("invalid input syntax for type json"),
 727                                          errdetail("Character with value 0x%02x must be escaped.",
 728                                                            (unsigned char) *s),
 729                                          report_json_context(lex)));
 730                 }
 731                 else if (*s == '\\')
 732                 {
 733                         /* OK, we have an escape character. */
 734                         s++;
 735                         len++;
 736                         if (len >= lex->input_length)
 737                         {
 738                                 lex->token_terminator = s;
 739                                 report_invalid_token(lex);
 740                         }
 741                         else if (*s == 'u')
 742                         {
 743                                 int                     i;
 744                                 int                     ch = 0;
 745
 746                                 for (i = 1; i <= 4; i++)
 747                                 {
 748                                         s++;
 749                                         len++;
 750                                         if (len >= lex->input_length)
 751                                         {
 752                                                 lex->token_terminator = s;
 753                                                 report_invalid_token(lex);
 754                                         }
 755                                         else if (*s >= '0' && *s <= '9')
 756                                                 ch = (ch * 16) + (*s - '0');
 757                                         else if (*s >= 'a' && *s <= 'f')
 758                                                 ch = (ch * 16) + (*s - 'a') + 10;
 759                                         else if (*s >= 'A' && *s <= 'F')
 760                                                 ch = (ch * 16) + (*s - 'A') + 10;
 761                                         else
 762                                         {
 763                                                 lex->token_terminator = s + pg_mblen(s);
 764                                                 ereport(ERROR,
 765                                                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
 766                                                                  errmsg("invalid input syntax for type json"),
 767                                                                  errdetail("\"\\u\" must be followed by four hexadecimal digits."),
 768                                                                  report_json_context(lex)));
 769                                         }
 770                                 }
 771                                 if (lex->strval != NULL)
 772                                 {
 773                                         char            utf8str[5];
 774                                         int                     utf8len;
 775
 776                                         if (ch >= 0xd800 && ch <= 0xdbff)
 777                                         {
 778                                                 if (hi_surrogate != -1)
 779                                                         ereport(ERROR,
 780                                                            (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
 781                                                                 errmsg("invalid input syntax for type json"),
 782                                                                 errdetail("Unicode high surrogate must not follow a high surrogate."),
 783                                                                 report_json_context(lex)));
 784                                                 hi_surrogate = (ch & 0x3ff) << 10;
 785                                                 continue;
 786                                         }
 787                                         else if (ch >= 0xdc00 && ch <= 0xdfff)
 788                                         {
 789                                                 if (hi_surrogate == -1)
 790                                                         ereport(ERROR,
 791                                                            (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
 792                                                                 errmsg("invalid input syntax for type json"),
 793                                                                 errdetail("Unicode low surrogate must follow a high surrogate."),
 794                                                                 report_json_context(lex)));
 795                                                 ch = 0x10000 + hi_surrogate + (ch & 0x3ff);
 796                                                 hi_surrogate = -1;
 797                                         }
 798
 799                                         if (hi_surrogate != -1)
 800                                                 ereport(ERROR,
 801                                                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
 802                                                                  errmsg("invalid input syntax for type json"),
 803                                                                  errdetail("Unicode low surrogate must follow a high surrogate."),
 804                                                                  report_json_context(lex)));
 805
 806                                         /*
 807                                          * For UTF8, replace the escape sequence by the actual
 808                                          * utf8 character in lex->strval. Do this also for other
 809                                          * encodings if the escape designates an ASCII character,
 810                                          * otherwise raise an error. We don't ever unescape a
 811                                          * \u0000, since that would result in an impermissible nul
 812                                          * byte.
 813                                          */
 814
 815                                         if (ch == 0)
 816                                         {
 817                                                 appendStringInfoString(lex->strval, "\\u0000");
 818                                         }
 819                                         else if (GetDatabaseEncoding() == PG_UTF8)
 820                                         {
 821                                                 unicode_to_utf8(ch, (unsigned char *) utf8str);
 822                                                 utf8len = pg_utf_mblen((unsigned char *) utf8str);
 823                                                 appendBinaryStringInfo(lex->strval, utf8str, utf8len);
 824                                         }
 825                                         else if (ch <= 0x007f)
 826                                         {
 827                                                 /*
 828                                                  * This is the only way to designate things like a
 829                                                  * form feed character in JSON, so it's useful in all
 830                                                  * encodings.
 831                                                  */
 832                                                 appendStringInfoChar(lex->strval, (char) ch);
 833                                         }
 834                                         else
 835                                         {
 836                                                 ereport(ERROR,
 837                                                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
 838                                                                  errmsg("invalid input syntax for type json"),
 839                                                                  errdetail("Unicode escape values cannot be used for code point values above 007F when the server encoding is not UTF8."),
 840                                                                  report_json_context(lex)));
 841                                         }
 842
 843                                 }
 844                         }
 845                         else if (lex->strval != NULL)
 846                         {
 847                                 if (hi_surrogate != -1)
 848                                         ereport(ERROR,
 849                                                         (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
 850                                                          errmsg("invalid input syntax for type json"),
 851                                                          errdetail("Unicode low surrogate must follow a high surrogate."),
 852                                                          report_json_context(lex)));
 853
 854                                 switch (*s)
 855                                 {
 856                                         case '"':
 857                                         case '\\':
 858                                         case '/':
 859                                                 appendStringInfoChar(lex->strval, *s);
 860                                                 break;
 861                                         case 'b':
 862                                                 appendStringInfoChar(lex->strval, '\b');
 863                                                 break;
 864                                         case 'f':
 865                                                 appendStringInfoChar(lex->strval, '\f');
 866                                                 break;
 867                                         case 'n':
 868                                                 appendStringInfoChar(lex->strval, '\n');
 869                                                 break;
 870                                         case 'r':
 871                                                 appendStringInfoChar(lex->strval, '\r');
 872                                                 break;
 873                                         case 't':
 874                                                 appendStringInfoChar(lex->strval, '\t');
 875                                                 break;
 876                                         default:
 877                                                 /* Not a valid string escape, so error out. */
 878                                                 lex->token_terminator = s + pg_mblen(s);
 879                                                 ereport(ERROR,
 880                                                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
 881                                                                  errmsg("invalid input syntax for type json"),
 882                                                         errdetail("Escape sequence \"\\%s\" is invalid.",
 883                                                                           extract_mb_char(s)),
 884                                                                  report_json_context(lex)));
 885                                 }
 886                         }
 887                         else if (strchr("\"\\/bfnrt", *s) == NULL)
 888                         {
 889                                 /*
 890                                  * Simpler processing if we're not bothered about de-escaping
 891                                  *
 892                                  * It's very tempting to remove the strchr() call here and
 893                                  * replace it with a switch statement, but testing so far has
 894                                  * shown it's not a performance win.
 895                                  */
 896                                 lex->token_terminator = s + pg_mblen(s);
 897                                 ereport(ERROR,
 898                                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
 899                                                  errmsg("invalid input syntax for type json"),
 900                                                  errdetail("Escape sequence \"\\%s\" is invalid.",
 901                                                                    extract_mb_char(s)),
 902                                                  report_json_context(lex)));
 903                         }
 904
 905                 }
 906                 else if (lex->strval != NULL)
 907                 {
 908                         if (hi_surrogate != -1)
 909                                 ereport(ERROR,
 910                                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
 911                                                  errmsg("invalid input syntax for type json"),
 912                                                  errdetail("Unicode low surrogate must follow a high surrogate."),
 913                                                  report_json_context(lex)));
 914
 915                         appendStringInfoChar(lex->strval, *s);
 916                 }
 917
 918         }
 919
 920         if (hi_surrogate != -1)
 921                 ereport(ERROR,
 922                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
 923                                  errmsg("invalid input syntax for type json"),
 924                         errdetail("Unicode low surrogate must follow a high surrogate."),
 925                                  report_json_context(lex)));
 926
 927         /* Hooray, we found the end of the string! */
 928         lex->prev_token_terminator = lex->token_terminator;
 929         lex->token_terminator = s + 1;
 930 }
 931
 932 /*-------------------------------------------------------------------------
 933  * The next token in the input stream is known to be a number; lex it.
 934  *
 935  * In JSON, a number consists of four parts:
 936  *
 937  * (1) An optional minus sign ('-').
 938  *
 939  * (2) Either a single '0', or a string of one or more digits that does not
 940  *         begin with a '0'.
 941  *
 942  * (3) An optional decimal part, consisting of a period ('.') followed by
 943  *         one or more digits.  (Note: While this part can be omitted
 944  *         completely, it's not OK to have only the decimal point without
 945  *         any digits afterwards.)
 946  *
 947  * (4) An optional exponent part, consisting of 'e' or 'E', optionally
 948  *         followed by '+' or '-', followed by one or more digits.  (Note:
 949  *         As with the decimal part, if 'e' or 'E' is present, it must be
 950  *         followed by at least one digit.)
 951  *
 952  * The 's' argument to this function points to the ostensible beginning
 953  * of part 2 - i.e. the character after any optional minus sign, and the
 954  * first character of the string if there is none.
 955  *
 956  *-------------------------------------------------------------------------
 957  */
 958 static inline void
 959 json_lex_number(JsonLexContext *lex, char *s, bool *num_err)
 960 {
 961         bool            error = false;
 962         char       *p;
 963         int                     len;
 964
 965         len = s - lex->input;
 966         /* Part (1): leading sign indicator. */
 967         /* Caller already did this for us; so do nothing. */
 968
 969         /* Part (2): parse main digit string. */
 970         if (*s == '0')
 971         {
 972                 s++;
 973                 len++;
 974         }
 975         else if (*s >= '1' && *s <= '9')
 976         {
 977                 do
 978                 {
 979                         s++;
 980                         len++;
 981                 } while (len < lex->input_length && *s >= '0' && *s <= '9');
 982         }
 983         else
 984                 error = true;
 985
 986         /* Part (3): parse optional decimal portion. */
 987         if (len < lex->input_length && *s == '.')
 988         {
 989                 s++;
 990                 len++;
 991                 if (len == lex->input_length || *s < '0' || *s > '9')
 992                         error = true;
 993                 else
 994                 {
 995                         do
 996                         {
 997                                 s++;
 998                                 len++;
 999                         } while (len < lex->input_length && *s >= '0' && *s <= '9');
1000                 }
1001         }
1002
1003         /* Part (4): parse optional exponent. */
1004         if (len < lex->input_length && (*s == 'e' || *s == 'E'))
1005         {
1006                 s++;
1007                 len++;
1008                 if (len < lex->input_length && (*s == '+' || *s == '-'))
1009                 {
1010                         s++;
1011                         len++;
1012                 }
1013                 if (len == lex->input_length || *s < '0' || *s > '9')
1014                         error = true;
1015                 else
1016                 {
1017                         do
1018                         {
1019                                 s++;
1020                                 len++;
1021                         } while (len < lex->input_length && *s >= '0' && *s <= '9');
1022                 }
1023         }
1024
1025         /*
1026          * Check for trailing garbage.  As in json_lex(), any alphanumeric stuff
1027          * here should be considered part of the token for error-reporting
1028          * purposes.
1029          */
1030         for (p = s; len < lex->input_length && JSON_ALPHANUMERIC_CHAR(*p); p++, len++)
1031                 error = true;
1032
1033         if (num_err != NULL)
1034         {
1035                 /* let the caller handle the error */
1036                 *num_err = error;
1037         }
1038         else
1039         {
1040                 lex->prev_token_terminator = lex->token_terminator;
1041                 lex->token_terminator = p;
1042                 if (error)
1043                         report_invalid_token(lex);
1044         }
1045 }
1046
1047 /*
1048  * Report a parse error.
1049  *
1050  * lex->token_start and lex->token_terminator must identify the current token.
1051  */
1052 static void
1053 report_parse_error(JsonParseContext ctx, JsonLexContext *lex)
1054 {
1055         char       *token;
1056         int                     toklen;
1057
1058         /* Handle case where the input ended prematurely. */
1059         if (lex->token_start == NULL || lex->token_type == JSON_TOKEN_END)
1060                 ereport(ERROR,
1061                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
1062                                  errmsg("invalid input syntax for type json"),
1063                                  errdetail("The input string ended unexpectedly."),
1064                                  report_json_context(lex)));
1065
1066         /* Separate out the current token. */
1067         toklen = lex->token_terminator - lex->token_start;
1068         token = palloc(toklen + 1);
1069         memcpy(token, lex->token_start, toklen);
1070         token[toklen] = '\0';
1071
1072         /* Complain, with the appropriate detail message. */
1073         if (ctx == JSON_PARSE_END)
1074                 ereport(ERROR,
1075                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
1076                                  errmsg("invalid input syntax for type json"),
1077                                  errdetail("Expected end of input, but found \"%s\".",
1078                                                    token),
1079                                  report_json_context(lex)));
1080         else
1081         {
1082                 switch (ctx)
1083                 {
1084                         case JSON_PARSE_VALUE:
1085                                 ereport(ERROR,
1086                                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
1087                                                  errmsg("invalid input syntax for type json"),
1088                                                  errdetail("Expected JSON value, but found \"%s\".",
1089                                                                    token),
1090                                                  report_json_context(lex)));
1091                                 break;
1092                         case JSON_PARSE_STRING:
1093                                 ereport(ERROR,
1094                                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
1095                                                  errmsg("invalid input syntax for type json"),
1096                                                  errdetail("Expected string, but found \"%s\".",
1097                                                                    token),
1098                                                  report_json_context(lex)));
1099                                 break;
1100                         case JSON_PARSE_ARRAY_START:
1101                                 ereport(ERROR,
1102                                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
1103                                                  errmsg("invalid input syntax for type json"),
1104                                                  errdetail("Expected array element or \"]\", but found \"%s\".",
1105                                                                    token),
1106                                                  report_json_context(lex)));
1107                                 break;
1108                         case JSON_PARSE_ARRAY_NEXT:
1109                                 ereport(ERROR,
1110                                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
1111                                                  errmsg("invalid input syntax for type json"),
1112                                           errdetail("Expected \",\" or \"]\", but found \"%s\".",
1113                                                                 token),
1114                                                  report_json_context(lex)));
1115                                 break;
1116                         case JSON_PARSE_OBJECT_START:
1117                                 ereport(ERROR,
1118                                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
1119                                                  errmsg("invalid input syntax for type json"),
1120                                          errdetail("Expected string or \"}\", but found \"%s\".",
1121                                                            token),
1122                                                  report_json_context(lex)));
1123                                 break;
1124                         case JSON_PARSE_OBJECT_LABEL:
1125                                 ereport(ERROR,
1126                                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
1127                                                  errmsg("invalid input syntax for type json"),
1128                                                  errdetail("Expected \":\", but found \"%s\".",
1129                                                                    token),
1130                                                  report_json_context(lex)));
1131                                 break;
1132                         case JSON_PARSE_OBJECT_NEXT:
1133                                 ereport(ERROR,
1134                                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
1135                                                  errmsg("invalid input syntax for type json"),
1136                                           errdetail("Expected \",\" or \"}\", but found \"%s\".",
1137                                                                 token),
1138                                                  report_json_context(lex)));
1139                                 break;
1140                         case JSON_PARSE_OBJECT_COMMA:
1141                                 ereport(ERROR,
1142                                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
1143                                                  errmsg("invalid input syntax for type json"),
1144                                                  errdetail("Expected string, but found \"%s\".",
1145                                                                    token),
1146                                                  report_json_context(lex)));
1147                                 break;
1148                         default:
1149                                 elog(ERROR, "unexpected json parse state: %d", ctx);
1150                 }
1151         }
1152 }
1153
1154 /*
1155  * Report an invalid input token.
1156  *
1157  * lex->token_start and lex->token_terminator must identify the token.
1158  */
1159 static void
1160 report_invalid_token(JsonLexContext *lex)
1161 {
1162         char       *token;
1163         int                     toklen;
1164
1165         /* Separate out the offending token. */
1166         toklen = lex->token_terminator - lex->token_start;
1167         token = palloc(toklen + 1);
1168         memcpy(token, lex->token_start, toklen);
1169         token[toklen] = '\0';
1170
1171         ereport(ERROR,
1172                         (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
1173                          errmsg("invalid input syntax for type json"),
1174                          errdetail("Token \"%s\" is invalid.", token),
1175                          report_json_context(lex)));
1176 }
1177
1178 /*
1179  * Report a CONTEXT line for bogus JSON input.
1180  *
1181  * lex->token_terminator must be set to identify the spot where we detected
1182  * the error.  Note that lex->token_start might be NULL, in case we recognized
1183  * error at EOF.
1184  *
1185  * The return value isn't meaningful, but we make it non-void so that this
1186  * can be invoked inside ereport().
1187  */
1188 static int
1189 report_json_context(JsonLexContext *lex)
1190 {
1191         const char *context_start;
1192         const char *context_end;
1193         const char *line_start;
1194         int                     line_number;
1195         char       *ctxt;
1196         int                     ctxtlen;
1197         const char *prefix;
1198         const char *suffix;
1199
1200         /* Choose boundaries for the part of the input we will display */
1201         context_start = lex->input;
1202         context_end = lex->token_terminator;
1203         line_start = context_start;
1204         line_number = 1;
1205         for (;;)
1206         {
1207                 /* Always advance over newlines */
1208                 if (context_start < context_end && *context_start == '\n')
1209                 {
1210                         context_start++;
1211                         line_start = context_start;
1212                         line_number++;
1213                         continue;
1214                 }
1215                 /* Otherwise, done as soon as we are close enough to context_end */
1216                 if (context_end - context_start < 50)
1217                         break;
1218                 /* Advance to next multibyte character */
1219                 if (IS_HIGHBIT_SET(*context_start))
1220                         context_start += pg_mblen(context_start);
1221                 else
1222                         context_start++;
1223         }
1224
1225         /*
1226          * We add "..." to indicate that the excerpt doesn't start at the
1227          * beginning of the line ... but if we're within 3 characters of the
1228          * beginning of the line, we might as well just show the whole line.
1229          */
1230         if (context_start - line_start <= 3)
1231                 context_start = line_start;
1232
1233         /* Get a null-terminated copy of the data to present */
1234         ctxtlen = context_end - context_start;
1235         ctxt = palloc(ctxtlen + 1);
1236         memcpy(ctxt, context_start, ctxtlen);
1237         ctxt[ctxtlen] = '\0';
1238
1239         /*
1240          * Show the context, prefixing "..." if not starting at start of line, and
1241          * suffixing "..." if not ending at end of line.
1242          */
1243         prefix = (context_start > line_start) ? "..." : "";
1244         suffix = (lex->token_type != JSON_TOKEN_END && context_end - lex->input < lex->input_length && *context_end != '\n' && *context_end != '\r') ? "..." : "";
1245
1246         return errcontext("JSON data, line %d: %s%s%s",
1247                                           line_number, prefix, ctxt, suffix);
1248 }
1249
1250 /*
1251  * Extract a single, possibly multi-byte char from the input string.
1252  */
1253 static char *
1254 extract_mb_char(char *s)
1255 {
1256         char       *res;
1257         int                     len;
1258
1259         len = pg_mblen(s);
1260         res = palloc(len + 1);
1261         memcpy(res, s, len);
1262         res[len] = '\0';
1263
1264         return res;
1265 }
1266
1267 /*
1268  * Determine how we want to print values of a given type in datum_to_json.
1269  *
1270  * Given the datatype OID, return its JsonTypeCategory, as well as the type's
1271  * output function OID.  If the returned category is JSONTYPE_CAST, we
1272  * return the OID of the type->JSON cast function instead.
1273  */
1274 static void
1275 json_categorize_type(Oid typoid,
1276                                          JsonTypeCategory *tcategory,
1277                                          Oid *outfuncoid)
1278 {
1279         bool            typisvarlena;
1280
1281         /* Look through any domain */
1282         typoid = getBaseType(typoid);
1283
1284         /* We'll usually need to return the type output function */
1285         getTypeOutputInfo(typoid, outfuncoid, &typisvarlena);
1286
1287         /* Check for known types */
1288         switch (typoid)
1289         {
1290                 case BOOLOID:
1291                         *tcategory = JSONTYPE_BOOL;
1292                         break;
1293
1294                 case INT2OID:
1295                 case INT4OID:
1296                 case INT8OID:
1297                 case FLOAT4OID:
1298                 case FLOAT8OID:
1299                 case NUMERICOID:
1300                         *tcategory = JSONTYPE_NUMERIC;
1301                         break;
1302
1303                 case DATEOID:
1304                         *tcategory = JSONTYPE_DATE;
1305                         break;
1306
1307                 case TIMESTAMPOID:
1308                         *tcategory = JSONTYPE_TIMESTAMP;
1309                         break;
1310
1311                 case TIMESTAMPTZOID:
1312                         *tcategory = JSONTYPE_TIMESTAMPTZ;
1313                         break;
1314
1315                 case JSONOID:
1316                 case JSONBOID:
1317                         *tcategory = JSONTYPE_JSON;
1318                         break;
1319
1320                 default:
1321                         /* Check for arrays and composites */
1322                         if (OidIsValid(get_element_type(typoid)))
1323                                 *tcategory = JSONTYPE_ARRAY;
1324                         else if (type_is_rowtype(typoid))
1325                                 *tcategory = JSONTYPE_COMPOSITE;
1326                         else
1327                         {
1328                                 /* It's probably the general case ... */
1329                                 *tcategory = JSONTYPE_OTHER;
1330                                 /* but let's look for a cast to json, if it's not built-in */
1331                                 if (typoid >= FirstNormalObjectId)
1332                                 {
1333                                         HeapTuple       tuple;
1334
1335                                         tuple = SearchSysCache2(CASTSOURCETARGET,
1336                                                                                         ObjectIdGetDatum(typoid),
1337                                                                                         ObjectIdGetDatum(JSONOID));
1338                                         if (HeapTupleIsValid(tuple))
1339                                         {
1340                                                 Form_pg_cast castForm = (Form_pg_cast) GETSTRUCT(tuple);
1341
1342                                                 if (castForm->castmethod == COERCION_METHOD_FUNCTION)
1343                                                 {
1344                                                         *tcategory = JSONTYPE_CAST;
1345                                                         *outfuncoid = castForm->castfunc;
1346                                                 }
1347
1348                                                 ReleaseSysCache(tuple);
1349                                         }
1350                                 }
1351                         }
1352                         break;
1353         }
1354 }
1355
1356 /*
1357  * Turn a Datum into JSON text, appending the string to "result".
1358  *
1359  * tcategory and outfuncoid are from a previous call to json_categorize_type,
1360  * except that if is_null is true then they can be invalid.
1361  *
1362  * If key_scalar is true, the value is being printed as a key, so insist
1363  * it's of an acceptable type, and force it to be quoted.
1364  */
1365 static void
1366 datum_to_json(Datum val, bool is_null, StringInfo result,
1367                           JsonTypeCategory tcategory, Oid outfuncoid,
1368                           bool key_scalar)
1369 {
1370         char       *outputstr;
1371         text       *jsontext;
1372
1373         /* callers are expected to ensure that null keys are not passed in */
1374         Assert(!(key_scalar && is_null));
1375
1376         if (is_null)
1377         {
1378                 appendStringInfoString(result, "null");
1379                 return;
1380         }
1381
1382         if (key_scalar &&
1383                 (tcategory == JSONTYPE_ARRAY ||
1384                  tcategory == JSONTYPE_COMPOSITE ||
1385                  tcategory == JSONTYPE_JSON ||
1386                  tcategory == JSONTYPE_CAST))
1387                 ereport(ERROR,
1388                                 (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1389                  errmsg("key value must be scalar, not array, composite, or json")));
1390
1391         switch (tcategory)
1392         {
1393                 case JSONTYPE_ARRAY:
1394                         array_to_json_internal(val, result, false);
1395                         break;
1396                 case JSONTYPE_COMPOSITE:
1397                         composite_to_json(val, result, false);
1398                         break;
1399                 case JSONTYPE_BOOL:
1400                         outputstr = DatumGetBool(val) ? "true" : "false";
1401                         if (key_scalar)
1402                                 escape_json(result, outputstr);
1403                         else
1404                                 appendStringInfoString(result, outputstr);
1405                         break;
1406                 case JSONTYPE_NUMERIC:
1407                         outputstr = OidOutputFunctionCall(outfuncoid, val);
1408
1409                         /*
1410                          * Don't call escape_json for a non-key if it's a valid JSON
1411                          * number.
1412                          */
1413                         if (!key_scalar && IsValidJsonNumber(outputstr, strlen(outputstr)))
1414                                 appendStringInfoString(result, outputstr);
1415                         else
1416                                 escape_json(result, outputstr);
1417                         pfree(outputstr);
1418                         break;
1419                 case JSONTYPE_DATE:
1420                         {
1421                                 DateADT         date;
1422                                 struct pg_tm tm;
1423                                 char            buf[MAXDATELEN + 1];
1424
1425                                 date = DatumGetDateADT(val);
1426
1427                                 /* XSD doesn't support infinite values */
1428                                 if (DATE_NOT_FINITE(date))
1429                                         ereport(ERROR,
1430                                                         (errcode(ERRCODE_DATETIME_VALUE_OUT_OF_RANGE),
1431                                                          errmsg("date out of range"),
1432                                                          errdetail("JSON does not support infinite date values.")));
1433                                 else
1434                                 {
1435                                         j2date(date + POSTGRES_EPOCH_JDATE,
1436                                                    &(tm.tm_year), &(tm.tm_mon), &(tm.tm_mday));
1437                                         EncodeDateOnly(&tm, USE_XSD_DATES, buf);
1438                                 }
1439
1440                                 appendStringInfo(result, "\"%s\"", buf);
1441                         }
1442                         break;
1443                 case JSONTYPE_TIMESTAMP:
1444                         {
1445                                 Timestamp       timestamp;
1446                                 struct pg_tm tm;
1447                                 fsec_t          fsec;
1448                                 char            buf[MAXDATELEN + 1];
1449
1450                                 timestamp = DatumGetTimestamp(val);
1451
1452                                 /* XSD doesn't support infinite values */
1453                                 if (TIMESTAMP_NOT_FINITE(timestamp))
1454                                         ereport(ERROR,
1455                                                         (errcode(ERRCODE_DATETIME_VALUE_OUT_OF_RANGE),
1456                                                          errmsg("timestamp out of range"),
1457                                                          errdetail("JSON does not support infinite timestamp values.")));
1458                                 else if (timestamp2tm(timestamp, NULL, &tm, &fsec, NULL, NULL) == 0)
1459                                         EncodeDateTime(&tm, fsec, false, 0, NULL, USE_XSD_DATES, buf);
1460                                 else
1461                                         ereport(ERROR,
1462                                                         (errcode(ERRCODE_DATETIME_VALUE_OUT_OF_RANGE),
1463                                                          errmsg("timestamp out of range")));
1464
1465                                 appendStringInfo(result, "\"%s\"", buf);
1466                         }
1467                         break;
1468                 case JSONTYPE_TIMESTAMPTZ:
1469                         {
1470                                 TimestampTz timestamp;
1471                                 struct pg_tm tm;
1472                                 int                     tz;
1473                                 fsec_t          fsec;
1474                                 const char *tzn = NULL;
1475                                 char            buf[MAXDATELEN + 1];
1476
1477                                 timestamp = DatumGetTimestamp(val);
1478
1479                                 /* XSD doesn't support infinite values */
1480                                 if (TIMESTAMP_NOT_FINITE(timestamp))
1481                                         ereport(ERROR,
1482                                                         (errcode(ERRCODE_DATETIME_VALUE_OUT_OF_RANGE),
1483                                                          errmsg("timestamp out of range"),
1484                                                          errdetail("JSON does not support infinite timestamp values.")));
1485                                 else if (timestamp2tm(timestamp, &tz, &tm, &fsec, &tzn, NULL) == 0)
1486                                         EncodeDateTime(&tm, fsec, true, tz, tzn, USE_XSD_DATES, buf);
1487                                 else
1488                                         ereport(ERROR,
1489                                                         (errcode(ERRCODE_DATETIME_VALUE_OUT_OF_RANGE),
1490                                                          errmsg("timestamp out of range")));
1491
1492                                 appendStringInfo(result, "\"%s\"", buf);
1493                         }
1494                         break;
1495                 case JSONTYPE_JSON:
1496                         /* JSON and JSONB output will already be escaped */
1497                         outputstr = OidOutputFunctionCall(outfuncoid, val);
1498                         appendStringInfoString(result, outputstr);
1499                         pfree(outputstr);
1500                         break;
1501                 case JSONTYPE_CAST:
1502                         /* outfuncoid refers to a cast function, not an output function */
1503                         jsontext = DatumGetTextP(OidFunctionCall1(outfuncoid, val));
1504                         outputstr = text_to_cstring(jsontext);
1505                         appendStringInfoString(result, outputstr);
1506                         pfree(outputstr);
1507                         pfree(jsontext);
1508                         break;
1509                 default:
1510                         outputstr = OidOutputFunctionCall(outfuncoid, val);
1511                         escape_json(result, outputstr);
1512                         pfree(outputstr);
1513                         break;
1514         }
1515 }
1516
1517 /*
1518  * Process a single dimension of an array.
1519  * If it's the innermost dimension, output the values, otherwise call
1520  * ourselves recursively to process the next dimension.
1521  */
1522 static void
1523 array_dim_to_json(StringInfo result, int dim, int ndims, int *dims, Datum *vals,
1524                                   bool *nulls, int *valcount, JsonTypeCategory tcategory,
1525                                   Oid outfuncoid, bool use_line_feeds)
1526 {
1527         int                     i;
1528         const char *sep;
1529
1530         Assert(dim < ndims);
1531
1532         sep = use_line_feeds ? ",\n " : ",";
1533
1534         appendStringInfoChar(result, '[');
1535
1536         for (i = 1; i <= dims[dim]; i++)
1537         {
1538                 if (i > 1)
1539                         appendStringInfoString(result, sep);
1540
1541                 if (dim + 1 == ndims)
1542                 {
1543                         datum_to_json(vals[*valcount], nulls[*valcount], result, tcategory,
1544                                                   outfuncoid, false);
1545                         (*valcount)++;
1546                 }
1547                 else
1548                 {
1549                         /*
1550                          * Do we want line feeds on inner dimensions of arrays? For now
1551                          * we'll say no.
1552                          */
1553                         array_dim_to_json(result, dim + 1, ndims, dims, vals, nulls,
1554                                                           valcount, tcategory, outfuncoid, false);
1555                 }
1556         }
1557
1558         appendStringInfoChar(result, ']');
1559 }
1560
1561 /*
1562  * Turn an array into JSON.
1563  */
1564 static void
1565 array_to_json_internal(Datum array, StringInfo result, bool use_line_feeds)
1566 {
1567         ArrayType  *v = DatumGetArrayTypeP(array);
1568         Oid                     element_type = ARR_ELEMTYPE(v);
1569         int                *dim;
1570         int                     ndim;
1571         int                     nitems;
1572         int                     count = 0;
1573         Datum      *elements;
1574         bool       *nulls;
1575         int16           typlen;
1576         bool            typbyval;
1577         char            typalign;
1578         JsonTypeCategory tcategory;
1579         Oid                     outfuncoid;
1580
1581         ndim = ARR_NDIM(v);
1582         dim = ARR_DIMS(v);
1583         nitems = ArrayGetNItems(ndim, dim);
1584
1585         if (nitems <= 0)
1586         {
1587                 appendStringInfoString(result, "[]");
1588                 return;
1589         }
1590
1591         get_typlenbyvalalign(element_type,
1592                                                  &typlen, &typbyval, &typalign);
1593
1594         json_categorize_type(element_type,
1595                                                  &tcategory, &outfuncoid);
1596
1597         deconstruct_array(v, element_type, typlen, typbyval,
1598                                           typalign, &elements, &nulls,
1599                                           &nitems);
1600
1601         array_dim_to_json(result, 0, ndim, dim, elements, nulls, &count, tcategory,
1602                                           outfuncoid, use_line_feeds);
1603
1604         pfree(elements);
1605         pfree(nulls);
1606 }
1607
1608 /*
1609  * Turn a composite / record into JSON.
1610  */
1611 static void
1612 composite_to_json(Datum composite, StringInfo result, bool use_line_feeds)
1613 {
1614         HeapTupleHeader td;
1615         Oid                     tupType;
1616         int32           tupTypmod;
1617         TupleDesc       tupdesc;
1618         HeapTupleData tmptup,
1619                            *tuple;
1620         int                     i;
1621         bool            needsep = false;
1622         const char *sep;
1623
1624         sep = use_line_feeds ? ",\n " : ",";
1625
1626         td = DatumGetHeapTupleHeader(composite);
1627
1628         /* Extract rowtype info and find a tupdesc */
1629         tupType = HeapTupleHeaderGetTypeId(td);
1630         tupTypmod = HeapTupleHeaderGetTypMod(td);
1631         tupdesc = lookup_rowtype_tupdesc(tupType, tupTypmod);
1632
1633         /* Build a temporary HeapTuple control structure */
1634         tmptup.t_len = HeapTupleHeaderGetDatumLength(td);
1635         tmptup.t_data = td;
1636         tuple = &tmptup;
1637
1638         appendStringInfoChar(result, '{');
1639
1640         for (i = 0; i < tupdesc->natts; i++)
1641         {
1642                 Datum           val;
1643                 bool            isnull;
1644                 char       *attname;
1645                 JsonTypeCategory tcategory;
1646                 Oid                     outfuncoid;
1647
1648                 if (tupdesc->attrs[i]->attisdropped)
1649                         continue;
1650
1651                 if (needsep)
1652                         appendStringInfoString(result, sep);
1653                 needsep = true;
1654
1655                 attname = NameStr(tupdesc->attrs[i]->attname);
1656                 escape_json(result, attname);
1657                 appendStringInfoChar(result, ':');
1658
1659                 val = heap_getattr(tuple, i + 1, tupdesc, &isnull);
1660
1661                 if (isnull)
1662                 {
1663                         tcategory = JSONTYPE_NULL;
1664                         outfuncoid = InvalidOid;
1665                 }
1666                 else
1667                         json_categorize_type(tupdesc->attrs[i]->atttypid,
1668                                                                  &tcategory, &outfuncoid);
1669
1670                 datum_to_json(val, isnull, result, tcategory, outfuncoid, false);
1671         }
1672
1673         appendStringInfoChar(result, '}');
1674         ReleaseTupleDesc(tupdesc);
1675 }
1676
1677 /*
1678  * Append JSON text for "val" to "result".
1679  *
1680  * This is just a thin wrapper around datum_to_json.  If the same type will be
1681  * printed many times, avoid using this; better to do the json_categorize_type
1682  * lookups only once.
1683  */
1684 static void
1685 add_json(Datum val, bool is_null, StringInfo result,
1686                  Oid val_type, bool key_scalar)
1687 {
1688         JsonTypeCategory tcategory;
1689         Oid                     outfuncoid;
1690
1691         if (val_type == InvalidOid)
1692                 ereport(ERROR,
1693                                 (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1694                                  errmsg("could not determine input data type")));
1695
1696         if (is_null)
1697         {
1698                 tcategory = JSONTYPE_NULL;
1699                 outfuncoid = InvalidOid;
1700         }
1701         else
1702                 json_categorize_type(val_type,
1703                                                          &tcategory, &outfuncoid);
1704
1705         datum_to_json(val, is_null, result, tcategory, outfuncoid, key_scalar);
1706 }
1707
1708 /*
1709  * SQL function array_to_json(row)
1710  */
1711 extern Datum
1712 array_to_json(PG_FUNCTION_ARGS)
1713 {
1714         Datum           array = PG_GETARG_DATUM(0);
1715         StringInfo      result;
1716
1717         result = makeStringInfo();
1718
1719         array_to_json_internal(array, result, false);
1720
1721         PG_RETURN_TEXT_P(cstring_to_text_with_len(result->data, result->len));
1722 }
1723
1724 /*
1725  * SQL function array_to_json(row, prettybool)
1726  */
1727 extern Datum
1728 array_to_json_pretty(PG_FUNCTION_ARGS)
1729 {
1730         Datum           array = PG_GETARG_DATUM(0);
1731         bool            use_line_feeds = PG_GETARG_BOOL(1);
1732         StringInfo      result;
1733
1734         result = makeStringInfo();
1735
1736         array_to_json_internal(array, result, use_line_feeds);
1737
1738         PG_RETURN_TEXT_P(cstring_to_text_with_len(result->data, result->len));
1739 }
1740
1741 /*
1742  * SQL function row_to_json(row)
1743  */
1744 extern Datum
1745 row_to_json(PG_FUNCTION_ARGS)
1746 {
1747         Datum           array = PG_GETARG_DATUM(0);
1748         StringInfo      result;
1749
1750         result = makeStringInfo();
1751
1752         composite_to_json(array, result, false);
1753
1754         PG_RETURN_TEXT_P(cstring_to_text_with_len(result->data, result->len));
1755 }
1756
1757 /*
1758  * SQL function row_to_json(row, prettybool)
1759  */
1760 extern Datum
1761 row_to_json_pretty(PG_FUNCTION_ARGS)
1762 {
1763         Datum           array = PG_GETARG_DATUM(0);
1764         bool            use_line_feeds = PG_GETARG_BOOL(1);
1765         StringInfo      result;
1766
1767         result = makeStringInfo();
1768
1769         composite_to_json(array, result, use_line_feeds);
1770
1771         PG_RETURN_TEXT_P(cstring_to_text_with_len(result->data, result->len));
1772 }
1773
1774 /*
1775  * SQL function to_json(anyvalue)
1776  */
1777 Datum
1778 to_json(PG_FUNCTION_ARGS)
1779 {
1780         Datum           val = PG_GETARG_DATUM(0);
1781         Oid                     val_type = get_fn_expr_argtype(fcinfo->flinfo, 0);
1782         StringInfo      result;
1783         JsonTypeCategory tcategory;
1784         Oid                     outfuncoid;
1785
1786         if (val_type == InvalidOid)
1787                 ereport(ERROR,
1788                                 (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1789                                  errmsg("could not determine input data type")));
1790
1791         json_categorize_type(val_type,
1792                                                  &tcategory, &outfuncoid);
1793
1794         result = makeStringInfo();
1795
1796         datum_to_json(val, false, result, tcategory, outfuncoid, false);
1797
1798         PG_RETURN_TEXT_P(cstring_to_text_with_len(result->data, result->len));
1799 }
1800
1801 /*
1802  * json_agg transition function
1803  *
1804  * aggregate input column as a json array value.
1805  */
1806 Datum
1807 json_agg_transfn(PG_FUNCTION_ARGS)
1808 {
1809         Oid                     val_type = get_fn_expr_argtype(fcinfo->flinfo, 1);
1810         MemoryContext aggcontext,
1811                                 oldcontext;
1812         StringInfo      state;
1813         Datum           val;
1814         JsonTypeCategory tcategory;
1815         Oid                     outfuncoid;
1816
1817         if (val_type == InvalidOid)
1818                 ereport(ERROR,
1819                                 (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1820                                  errmsg("could not determine input data type")));
1821
1822         if (!AggCheckCallContext(fcinfo, &aggcontext))
1823         {
1824                 /* cannot be called directly because of internal-type argument */
1825                 elog(ERROR, "json_agg_transfn called in non-aggregate context");
1826         }
1827
1828         if (PG_ARGISNULL(0))
1829         {
1830                 /*
1831                  * Make this StringInfo in a context where it will persist for the
1832                  * duration of the aggregate call.  MemoryContextSwitchTo is only
1833                  * needed the first time, as the StringInfo routines make sure they
1834                  * use the right context to enlarge the object if necessary.
1835                  */
1836                 oldcontext = MemoryContextSwitchTo(aggcontext);
1837                 state = makeStringInfo();
1838                 MemoryContextSwitchTo(oldcontext);
1839
1840                 appendStringInfoChar(state, '[');
1841         }
1842         else
1843         {
1844                 state = (StringInfo) PG_GETARG_POINTER(0);
1845                 appendStringInfoString(state, ", ");
1846         }
1847
1848         /* fast path for NULLs */
1849         if (PG_ARGISNULL(1))
1850         {
1851                 datum_to_json((Datum) 0, true, state, JSONTYPE_NULL, InvalidOid, false);
1852                 PG_RETURN_POINTER(state);
1853         }
1854
1855         val = PG_GETARG_DATUM(1);
1856
1857         /* XXX we do this every time?? */
1858         json_categorize_type(val_type,
1859                                                  &tcategory, &outfuncoid);
1860
1861         /* add some whitespace if structured type and not first item */
1862         if (!PG_ARGISNULL(0) &&
1863                 (tcategory == JSONTYPE_ARRAY || tcategory == JSONTYPE_COMPOSITE))
1864         {
1865                 appendStringInfoString(state, "\n ");
1866         }
1867
1868         datum_to_json(val, false, state, tcategory, outfuncoid, false);
1869
1870         /*
1871          * The transition type for array_agg() is declared to be "internal", which
1872          * is a pass-by-value type the same size as a pointer.  So we can safely
1873          * pass the ArrayBuildState pointer through nodeAgg.c's machinations.
1874          */
1875         PG_RETURN_POINTER(state);
1876 }
1877
1878 /*
1879  * json_agg final function
1880  */
1881 Datum
1882 json_agg_finalfn(PG_FUNCTION_ARGS)
1883 {
1884         StringInfo      state;
1885
1886         /* cannot be called directly because of internal-type argument */
1887         Assert(AggCheckCallContext(fcinfo, NULL));
1888
1889         state = PG_ARGISNULL(0) ? NULL : (StringInfo) PG_GETARG_POINTER(0);
1890
1891         /* NULL result for no rows in, as is standard with aggregates */
1892         if (state == NULL)
1893                 PG_RETURN_NULL();
1894
1895         /* Else return state with appropriate array terminator added */
1896         PG_RETURN_TEXT_P(catenate_stringinfo_string(state, "]"));
1897 }
1898
1899 /*
1900  * json_object_agg transition function.
1901  *
1902  * aggregate two input columns as a single json object value.
1903  */
1904 Datum
1905 json_object_agg_transfn(PG_FUNCTION_ARGS)
1906 {
1907         Oid                     val_type;
1908         MemoryContext aggcontext,
1909                                 oldcontext;
1910         StringInfo      state;
1911         Datum           arg;
1912
1913         if (!AggCheckCallContext(fcinfo, &aggcontext))
1914         {
1915                 /* cannot be called directly because of internal-type argument */
1916                 elog(ERROR, "json_object_agg_transfn called in non-aggregate context");
1917         }
1918
1919         if (PG_ARGISNULL(0))
1920         {
1921                 /*
1922                  * Make the StringInfo in a context where it will persist for the
1923                  * duration of the aggregate call. Switching context is only needed
1924                  * for this initial step, as the StringInfo routines make sure they
1925                  * use the right context to enlarge the object if necessary.
1926                  */
1927                 oldcontext = MemoryContextSwitchTo(aggcontext);
1928                 state = makeStringInfo();
1929                 MemoryContextSwitchTo(oldcontext);
1930
1931                 appendStringInfoString(state, "{ ");
1932         }
1933         else
1934         {
1935                 state = (StringInfo) PG_GETARG_POINTER(0);
1936                 appendStringInfoString(state, ", ");
1937         }
1938
1939         /*
1940          * Note: since json_object_agg() is declared as taking type "any", the
1941          * parser will not do any type conversion on unknown-type literals (that
1942          * is, undecorated strings or NULLs).  Such values will arrive here as
1943          * type UNKNOWN, which fortunately does not matter to us, since
1944          * unknownout() works fine.
1945          */
1946         val_type = get_fn_expr_argtype(fcinfo->flinfo, 1);
1947
1948         if (val_type == InvalidOid)
1949                 ereport(ERROR,
1950                                 (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1951                                  errmsg("could not determine data type for argument %d", 1)));
1952
1953         if (PG_ARGISNULL(1))
1954                 ereport(ERROR,
1955                                 (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1956                                  errmsg("field name must not be null")));
1957
1958         arg = PG_GETARG_DATUM(1);
1959
1960         add_json(arg, false, state, val_type, true);
1961
1962         appendStringInfoString(state, " : ");
1963
1964         val_type = get_fn_expr_argtype(fcinfo->flinfo, 2);
1965
1966         if (val_type == InvalidOid)
1967                 ereport(ERROR,
1968                                 (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1969                                  errmsg("could not determine data type for argument %d", 2)));
1970
1971         if (PG_ARGISNULL(2))
1972                 arg = (Datum) 0;
1973         else
1974                 arg = PG_GETARG_DATUM(2);
1975
1976         add_json(arg, PG_ARGISNULL(2), state, val_type, false);
1977
1978         PG_RETURN_POINTER(state);
1979 }
1980
1981 /*
1982  * json_object_agg final function.
1983  */
1984 Datum
1985 json_object_agg_finalfn(PG_FUNCTION_ARGS)
1986 {
1987         StringInfo      state;
1988
1989         /* cannot be called directly because of internal-type argument */
1990         Assert(AggCheckCallContext(fcinfo, NULL));
1991
1992         state = PG_ARGISNULL(0) ? NULL : (StringInfo) PG_GETARG_POINTER(0);
1993
1994         /* NULL result for no rows in, as is standard with aggregates */
1995         if (state == NULL)
1996                 PG_RETURN_NULL();
1997
1998         /* Else return state with appropriate object terminator added */
1999         PG_RETURN_TEXT_P(catenate_stringinfo_string(state, " }"));
2000 }
2001
2002 /*
2003  * Helper function for aggregates: return given StringInfo's contents plus
2004  * specified trailing string, as a text datum.  We need this because aggregate
2005  * final functions are not allowed to modify the aggregate state.
2006  */
2007 static text *
2008 catenate_stringinfo_string(StringInfo buffer, const char *addon)
2009 {
2010         /* custom version of cstring_to_text_with_len */
2011         int                     buflen = buffer->len;
2012         int                     addlen = strlen(addon);
2013         text       *result = (text *) palloc(buflen + addlen + VARHDRSZ);
2014
2015         SET_VARSIZE(result, buflen + addlen + VARHDRSZ);
2016         memcpy(VARDATA(result), buffer->data, buflen);
2017         memcpy(VARDATA(result) + buflen, addon, addlen);
2018
2019         return result;
2020 }
2021
2022 /*
2023  * SQL function json_build_object(variadic "any")
2024  */
2025 Datum
2026 json_build_object(PG_FUNCTION_ARGS)
2027 {
2028         int                     nargs = PG_NARGS();
2029         int                     i;
2030         Datum           arg;
2031         const char *sep = "";
2032         StringInfo      result;
2033         Oid                     val_type;
2034
2035         if (nargs % 2 != 0)
2036                 ereport(ERROR,
2037                                 (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
2038                                  errmsg("argument list must have even number of elements"),
2039                                  errhint("The arguments of json_build_object() must consist of alternating keys and values.")));
2040
2041         result = makeStringInfo();
2042
2043         appendStringInfoChar(result, '{');
2044
2045         for (i = 0; i < nargs; i += 2)
2046         {
2047                 /*
2048                  * Note: since json_build_object() is declared as taking type "any",
2049                  * the parser will not do any type conversion on unknown-type literals
2050                  * (that is, undecorated strings or NULLs).  Such values will arrive
2051                  * here as type UNKNOWN, which fortunately does not matter to us,
2052                  * since unknownout() works fine.
2053                  */
2054                 appendStringInfoString(result, sep);
2055                 sep = ", ";
2056
2057                 /* process key */
2058                 val_type = get_fn_expr_argtype(fcinfo->flinfo, i);
2059
2060                 if (val_type == InvalidOid)
2061                         ereport(ERROR,
2062                                         (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
2063                                          errmsg("could not determine data type for argument %d",
2064                                                         i + 1)));
2065
2066                 if (PG_ARGISNULL(i))
2067                         ereport(ERROR,
2068                                         (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
2069                                          errmsg("argument %d cannot be null", i + 1),
2070                                          errhint("Object keys should be text.")));
2071
2072                 arg = PG_GETARG_DATUM(i);
2073
2074                 add_json(arg, false, result, val_type, true);
2075
2076                 appendStringInfoString(result, " : ");
2077
2078                 /* process value */
2079                 val_type = get_fn_expr_argtype(fcinfo->flinfo, i + 1);
2080
2081                 if (val_type == InvalidOid)
2082                         ereport(ERROR,
2083                                         (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
2084                                          errmsg("could not determine data type for argument %d",
2085                                                         i + 2)));
2086
2087                 if (PG_ARGISNULL(i + 1))
2088                         arg = (Datum) 0;
2089                 else
2090                         arg = PG_GETARG_DATUM(i + 1);
2091
2092                 add_json(arg, PG_ARGISNULL(i + 1), result, val_type, false);
2093         }
2094
2095         appendStringInfoChar(result, '}');
2096
2097         PG_RETURN_TEXT_P(cstring_to_text_with_len(result->data, result->len));
2098 }
2099
2100 /*
2101  * degenerate case of json_build_object where it gets 0 arguments.
2102  */
2103 Datum
2104 json_build_object_noargs(PG_FUNCTION_ARGS)
2105 {
2106         PG_RETURN_TEXT_P(cstring_to_text_with_len("{}", 2));
2107 }
2108
2109 /*
2110  * SQL function json_build_array(variadic "any")
2111  */
2112 Datum
2113 json_build_array(PG_FUNCTION_ARGS)
2114 {
2115         int                     nargs = PG_NARGS();
2116         int                     i;
2117         Datum           arg;
2118         const char *sep = "";
2119         StringInfo      result;
2120         Oid                     val_type;
2121
2122         result = makeStringInfo();
2123
2124         appendStringInfoChar(result, '[');
2125
2126         for (i = 0; i < nargs; i++)
2127         {
2128                 /*
2129                  * Note: since json_build_array() is declared as taking type "any",
2130                  * the parser will not do any type conversion on unknown-type literals
2131                  * (that is, undecorated strings or NULLs).  Such values will arrive
2132                  * here as type UNKNOWN, which fortunately does not matter to us,
2133                  * since unknownout() works fine.
2134                  */
2135                 appendStringInfoString(result, sep);
2136                 sep = ", ";
2137
2138                 val_type = get_fn_expr_argtype(fcinfo->flinfo, i);
2139
2140                 if (val_type == InvalidOid)
2141                         ereport(ERROR,
2142                                         (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
2143                                          errmsg("could not determine data type for argument %d",
2144                                                         i + 1)));
2145
2146                 if (PG_ARGISNULL(i))
2147                         arg = (Datum) 0;
2148                 else
2149                         arg = PG_GETARG_DATUM(i);
2150
2151                 add_json(arg, PG_ARGISNULL(i), result, val_type, false);
2152         }
2153
2154         appendStringInfoChar(result, ']');
2155
2156         PG_RETURN_TEXT_P(cstring_to_text_with_len(result->data, result->len));
2157 }
2158
2159 /*
2160  * degenerate case of json_build_array where it gets 0 arguments.
2161  */
2162 Datum
2163 json_build_array_noargs(PG_FUNCTION_ARGS)
2164 {
2165         PG_RETURN_TEXT_P(cstring_to_text_with_len("[]", 2));
2166 }
2167
2168 /*
2169  * SQL function json_object(text[])
2170  *
2171  * take a one or two dimensional array of text as key/value pairs
2172  * for a json object.
2173  */
2174 Datum
2175 json_object(PG_FUNCTION_ARGS)
2176 {
2177         ArrayType  *in_array = PG_GETARG_ARRAYTYPE_P(0);
2178         int                     ndims = ARR_NDIM(in_array);
2179         StringInfoData result;
2180         Datum      *in_datums;
2181         bool       *in_nulls;
2182         int                     in_count,
2183                                 count,
2184                                 i;
2185         text       *rval;
2186         char       *v;
2187
2188         switch (ndims)
2189         {
2190                 case 0:
2191                         PG_RETURN_DATUM(CStringGetTextDatum("{}"));
2192                         break;
2193
2194                 case 1:
2195                         if ((ARR_DIMS(in_array)[0]) % 2)
2196                                 ereport(ERROR,
2197                                                 (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
2198                                                  errmsg("array must have even number of elements")));
2199                         break;
2200
2201                 case 2:
2202                         if ((ARR_DIMS(in_array)[1]) != 2)
2203                                 ereport(ERROR,
2204                                                 (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
2205                                                  errmsg("array must have two columns")));
2206                         break;
2207
2208                 default:
2209                         ereport(ERROR,
2210                                         (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
2211                                          errmsg("wrong number of array subscripts")));
2212         }
2213
2214         deconstruct_array(in_array,
2215                                           TEXTOID, -1, false, 'i',
2216                                           &in_datums, &in_nulls, &in_count);
2217
2218         count = in_count / 2;
2219
2220         initStringInfo(&result);
2221
2222         appendStringInfoChar(&result, '{');
2223
2224         for (i = 0; i < count; ++i)
2225         {
2226                 if (in_nulls[i * 2])
2227                         ereport(ERROR,
2228                                         (errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED),
2229                                          errmsg("null value not allowed for object key")));
2230
2231                 v = TextDatumGetCString(in_datums[i * 2]);
2232                 if (i > 0)
2233                         appendStringInfoString(&result, ", ");
2234                 escape_json(&result, v);
2235                 appendStringInfoString(&result, " : ");
2236                 pfree(v);
2237                 if (in_nulls[i * 2 + 1])
2238                         appendStringInfoString(&result, "null");
2239                 else
2240                 {
2241                         v = TextDatumGetCString(in_datums[i * 2 + 1]);
2242                         escape_json(&result, v);
2243                         pfree(v);
2244                 }
2245         }
2246
2247         appendStringInfoChar(&result, '}');
2248
2249         pfree(in_datums);
2250         pfree(in_nulls);
2251
2252         rval = cstring_to_text_with_len(result.data, result.len);
2253         pfree(result.data);
2254
2255         PG_RETURN_TEXT_P(rval);
2256
2257 }
2258
2259 /*
2260  * SQL function json_object(text[], text[])
2261  *
2262  * take separate key and value arrays of text to construct a json object
2263  * pairwise.
2264  */
2265 Datum
2266 json_object_two_arg(PG_FUNCTION_ARGS)
2267 {
2268         ArrayType  *key_array = PG_GETARG_ARRAYTYPE_P(0);
2269         ArrayType  *val_array = PG_GETARG_ARRAYTYPE_P(1);
2270         int                     nkdims = ARR_NDIM(key_array);
2271         int                     nvdims = ARR_NDIM(val_array);
2272         StringInfoData result;
2273         Datum      *key_datums,
2274                            *val_datums;
2275         bool       *key_nulls,
2276                            *val_nulls;
2277         int                     key_count,
2278                                 val_count,
2279                                 i;
2280         text       *rval;
2281         char       *v;
2282
2283         if (nkdims > 1 || nkdims != nvdims)
2284                 ereport(ERROR,
2285                                 (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
2286                                  errmsg("wrong number of array subscripts")));
2287
2288         if (nkdims == 0)
2289                 PG_RETURN_DATUM(CStringGetTextDatum("{}"));
2290
2291         deconstruct_array(key_array,
2292                                           TEXTOID, -1, false, 'i',
2293                                           &key_datums, &key_nulls, &key_count);
2294
2295         deconstruct_array(val_array,
2296                                           TEXTOID, -1, false, 'i',
2297                                           &val_datums, &val_nulls, &val_count);
2298
2299         if (key_count != val_count)
2300                 ereport(ERROR,
2301                                 (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
2302                                  errmsg("mismatched array dimensions")));
2303
2304         initStringInfo(&result);
2305
2306         appendStringInfoChar(&result, '{');
2307
2308         for (i = 0; i < key_count; ++i)
2309         {
2310                 if (key_nulls[i])
2311                         ereport(ERROR,
2312                                         (errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED),
2313                                          errmsg("null value not allowed for object key")));
2314
2315                 v = TextDatumGetCString(key_datums[i]);
2316                 if (i > 0)
2317                         appendStringInfoString(&result, ", ");
2318                 escape_json(&result, v);
2319                 appendStringInfoString(&result, " : ");
2320                 pfree(v);
2321                 if (val_nulls[i])
2322                         appendStringInfoString(&result, "null");
2323                 else
2324                 {
2325                         v = TextDatumGetCString(val_datums[i]);
2326                         escape_json(&result, v);
2327                         pfree(v);
2328                 }
2329         }
2330
2331         appendStringInfoChar(&result, '}');
2332
2333         pfree(key_datums);
2334         pfree(key_nulls);
2335         pfree(val_datums);
2336         pfree(val_nulls);
2337
2338         rval = cstring_to_text_with_len(result.data, result.len);
2339         pfree(result.data);
2340
2341         PG_RETURN_TEXT_P(rval);
2342 }
2343
2344
2345 /*
2346  * Produce a JSON string literal, properly escaping characters in the text.
2347  */
2348 void
2349 escape_json(StringInfo buf, const char *str)
2350 {
2351         const char *p;
2352
2353         appendStringInfoCharMacro(buf, '\"');
2354         for (p = str; *p; p++)
2355         {
2356                 switch (*p)
2357                 {
2358                         case '\b':
2359                                 appendStringInfoString(buf, "\\b");
2360                                 break;
2361                         case '\f':
2362                                 appendStringInfoString(buf, "\\f");
2363                                 break;
2364                         case '\n':
2365                                 appendStringInfoString(buf, "\\n");
2366                                 break;
2367                         case '\r':
2368                                 appendStringInfoString(buf, "\\r");
2369                                 break;
2370                         case '\t':
2371                                 appendStringInfoString(buf, "\\t");
2372                                 break;
2373                         case '"':
2374                                 appendStringInfoString(buf, "\\\"");
2375                                 break;
2376                         case '\\':
2377
2378                                 /*
2379                                  * Unicode escapes are passed through as is. There is no
2380                                  * requirement that they denote a valid character in the
2381                                  * server encoding - indeed that is a big part of their
2382                                  * usefulness.
2383                                  *
2384                                  * All we require is that they consist of \uXXXX where the Xs
2385                                  * are hexadecimal digits. It is the responsibility of the
2386                                  * caller of, say, to_json() to make sure that the unicode
2387                                  * escape is valid.
2388                                  *
2389                                  * In the case of a jsonb string value being escaped, the only
2390                                  * unicode escape that should be present is \u0000, all the
2391                                  * other unicode escapes will have been resolved.
2392                                  */
2393                                 if (p[1] == 'u' &&
2394                                         isxdigit((unsigned char) p[2]) &&
2395                                         isxdigit((unsigned char) p[3]) &&
2396                                         isxdigit((unsigned char) p[4]) &&
2397                                         isxdigit((unsigned char) p[5]))
2398                                         appendStringInfoCharMacro(buf, *p);
2399                                 else
2400                                         appendStringInfoString(buf, "\\\\");
2401                                 break;
2402                         default:
2403                                 if ((unsigned char) *p < ' ')
2404                                         appendStringInfo(buf, "\\u%04x", (int) *p);
2405                                 else
2406                                         appendStringInfoCharMacro(buf, *p);
2407                                 break;
2408                 }
2409         }
2410         appendStringInfoCharMacro(buf, '\"');
2411 }
2412
2413 /*
2414  * SQL function json_typeof(json) -> text
2415  *
2416  * Returns the type of the outermost JSON value as TEXT.  Possible types are
2417  * "object", "array", "string", "number", "boolean", and "null".
2418  *
2419  * Performs a single call to json_lex() to get the first token of the supplied
2420  * value.  This initial token uniquely determines the value's type.  As our
2421  * input must already have been validated by json_in() or json_recv(), the
2422  * initial token should never be JSON_TOKEN_OBJECT_END, JSON_TOKEN_ARRAY_END,
2423  * JSON_TOKEN_COLON, JSON_TOKEN_COMMA, or JSON_TOKEN_END.
2424  */
2425 Datum
2426 json_typeof(PG_FUNCTION_ARGS)
2427 {
2428         text       *json;
2429
2430         JsonLexContext *lex;
2431         JsonTokenType tok;
2432         char       *type;
2433
2434         json = PG_GETARG_TEXT_P(0);
2435         lex = makeJsonLexContext(json, false);
2436
2437         /* Lex exactly one token from the input and check its type. */
2438         json_lex(lex);
2439         tok = lex_peek(lex);
2440         switch (tok)
2441         {
2442                 case JSON_TOKEN_OBJECT_START:
2443                         type = "object";
2444                         break;
2445                 case JSON_TOKEN_ARRAY_START:
2446                         type = "array";
2447                         break;
2448                 case JSON_TOKEN_STRING:
2449                         type = "string";
2450                         break;
2451                 case JSON_TOKEN_NUMBER:
2452                         type = "number";
2453                         break;
2454                 case JSON_TOKEN_TRUE:
2455                 case JSON_TOKEN_FALSE:
2456                         type = "boolean";
2457                         break;
2458                 case JSON_TOKEN_NULL:
2459                         type = "null";
2460                         break;
2461                 default:
2462                         elog(ERROR, "unexpected json token: %d", tok);
2463         }
2464
2465         PG_RETURN_TEXT_P(cstring_to_text(type));
2466 }