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