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) 1996-2013, 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 "parser/parse_coerce.h"
  25 #include "utils/array.h"
  26 #include "utils/builtins.h"
  27 #include "utils/lsyscache.h"
  28 #include "utils/json.h"
  29 #include "utils/jsonapi.h"
  30 #include "utils/typcache.h"
  31 #include "utils/syscache.h"
  32
  33 /*
  34  * The context of the parser is maintained by the recursive descent
  35  * mechanism, but is passed explicitly to the error reporting routine
  36  * for better diagnostics.
  37  */
  38 typedef enum                                    /* contexts of JSON parser */
  39 {
  40         JSON_PARSE_VALUE,                       /* expecting a value */
  41         JSON_PARSE_STRING,                      /* expecting a string (for a field name) */
  42         JSON_PARSE_ARRAY_START,         /* saw '[', expecting value or ']' */
  43         JSON_PARSE_ARRAY_NEXT,          /* saw array element, expecting ',' or ']' */
  44         JSON_PARSE_OBJECT_START,        /* saw '{', expecting label or '}' */
  45         JSON_PARSE_OBJECT_LABEL,        /* saw object label, expecting ':' */
  46         JSON_PARSE_OBJECT_NEXT,         /* saw object value, expecting ',' or '}' */
  47         JSON_PARSE_OBJECT_COMMA,        /* saw object ',', expecting next label */
  48         JSON_PARSE_END                          /* saw the end of a document, expect nothing */
  49 } JsonParseContext;
  50
  51 static inline void json_lex(JsonLexContext *lex);
  52 static inline void json_lex_string(JsonLexContext *lex);
  53 static inline void json_lex_number(JsonLexContext *lex, char *s);
  54 static inline void parse_scalar(JsonLexContext *lex, JsonSemAction *sem);
  55 static void parse_object_field(JsonLexContext *lex, JsonSemAction *sem);
  56 static void parse_object(JsonLexContext *lex, JsonSemAction *sem);
  57 static void parse_array_element(JsonLexContext *lex, JsonSemAction *sem);
  58 static void parse_array(JsonLexContext *lex, JsonSemAction *sem);
  59 static void report_parse_error(JsonParseContext ctx, JsonLexContext *lex);
  60 static void report_invalid_token(JsonLexContext *lex);
  61 static int      report_json_context(JsonLexContext *lex);
  62 static char *extract_mb_char(char *s);
  63 static void composite_to_json(Datum composite, StringInfo result,
  64                                   bool use_line_feeds);
  65 static void array_dim_to_json(StringInfo result, int dim, int ndims, int *dims,
  66                                   Datum *vals, bool *nulls, int *valcount,
  67                                   TYPCATEGORY tcategory, Oid typoutputfunc,
  68                                   bool use_line_feeds);
  69 static void array_to_json_internal(Datum array, StringInfo result,
  70                                            bool use_line_feeds);
  71
  72 /* the null action object used for pure validation */
  73 static JsonSemAction nullSemAction =
  74 {
  75         NULL, NULL, NULL, NULL, NULL,
  76         NULL, NULL, NULL, NULL, NULL
  77 };
  78
  79 /* Recursive Descent parser support routines */
  80
  81 /*
  82  * lex_peek
  83  *
  84  * what is the current look_ahead token?
  85 */
  86 static inline JsonTokenType
  87 lex_peek(JsonLexContext *lex)
  88 {
  89         return lex->token_type;
  90 }
  91
  92 /*
  93  * lex_accept
  94  *
  95  * accept the look_ahead token and move the lexer to the next token if the
  96  * look_ahead token matches the token parameter. In that case, and if required,
  97  * also hand back the de-escaped lexeme.
  98  *
  99  * returns true if the token matched, false otherwise.
 100  */
 101 static inline bool
 102 lex_accept(JsonLexContext *lex, JsonTokenType token, char **lexeme)
 103 {
 104         if (lex->token_type == token)
 105         {
 106                 if (lexeme != NULL)
 107                 {
 108                         if (lex->token_type == JSON_TOKEN_STRING)
 109                         {
 110                                 if (lex->strval != NULL)
 111                                         *lexeme = pstrdup(lex->strval->data);
 112                         }
 113                         else
 114                         {
 115                                 int                     len = (lex->token_terminator - lex->token_start);
 116                                 char       *tokstr = palloc(len + 1);
 117
 118                                 memcpy(tokstr, lex->token_start, len);
 119                                 tokstr[len] = '\0';
 120                                 *lexeme = tokstr;
 121                         }
 122                 }
 123                 json_lex(lex);
 124                 return true;
 125         }
 126         return false;
 127 }
 128
 129 /*
 130  * lex_accept
 131  *
 132  * move the lexer to the next token if the current look_ahead token matches
 133  * the parameter token. Otherwise, report an error.
 134  */
 135 static inline void
 136 lex_expect(JsonParseContext ctx, JsonLexContext *lex, JsonTokenType token)
 137 {
 138         if (!lex_accept(lex, token, NULL))
 139                 report_parse_error(ctx, lex);;
 140 }
 141
 142 /*
 143  * All the defined      type categories are upper case , so use lower case here
 144  * so we avoid any possible clash.
 145  */
 146 /* fake type category for JSON so we can distinguish it in datum_to_json */
 147 #define TYPCATEGORY_JSON 'j'
 148 /* fake category for types that have a cast to json */
 149 #define TYPCATEGORY_JSON_CAST 'c'
 150 /* letters appearing in numeric output that aren't valid in a JSON number */
 151 #define NON_NUMERIC_LETTER "NnAaIiFfTtYy"
 152 /* chars to consider as part of an alphanumeric token */
 153 #define JSON_ALPHANUMERIC_CHAR(c)  \
 154         (((c) >= 'a' && (c) <= 'z') || \
 155          ((c) >= 'A' && (c) <= 'Z') || \
 156          ((c) >= '0' && (c) <= '9') || \
 157          (c) == '_' || \
 158          IS_HIGHBIT_SET(c))
 159
 160 /*
 161  * Input.
 162  */
 163 Datum
 164 json_in(PG_FUNCTION_ARGS)
 165 {
 166         char       *json = PG_GETARG_CSTRING(0);
 167         text       *result = cstring_to_text(json);
 168         JsonLexContext *lex;
 169
 170         /* validate it */
 171         lex = makeJsonLexContext(result, false);
 172         pg_parse_json(lex, &nullSemAction);
 173
 174         /* Internal representation is the same as text, for now */
 175         PG_RETURN_TEXT_P(result);
 176 }
 177
 178 /*
 179  * Output.
 180  */
 181 Datum
 182 json_out(PG_FUNCTION_ARGS)
 183 {
 184         /* we needn't detoast because text_to_cstring will handle that */
 185         Datum           txt = PG_GETARG_DATUM(0);
 186
 187         PG_RETURN_CSTRING(TextDatumGetCString(txt));
 188 }
 189
 190 /*
 191  * Binary send.
 192  */
 193 Datum
 194 json_send(PG_FUNCTION_ARGS)
 195 {
 196         text       *t = PG_GETARG_TEXT_PP(0);
 197         StringInfoData buf;
 198
 199         pq_begintypsend(&buf);
 200         pq_sendtext(&buf, VARDATA_ANY(t), VARSIZE_ANY_EXHDR(t));
 201         PG_RETURN_BYTEA_P(pq_endtypsend(&buf));
 202 }
 203
 204 /*
 205  * Binary receive.
 206  */
 207 Datum
 208 json_recv(PG_FUNCTION_ARGS)
 209 {
 210         StringInfo      buf = (StringInfo) PG_GETARG_POINTER(0);
 211         text       *result;
 212         char       *str;
 213         int                     nbytes;
 214         JsonLexContext *lex;
 215
 216         str = pq_getmsgtext(buf, buf->len - buf->cursor, &nbytes);
 217
 218         result = palloc(nbytes + VARHDRSZ);
 219         SET_VARSIZE(result, nbytes + VARHDRSZ);
 220         memcpy(VARDATA(result), str, nbytes);
 221
 222         /* Validate it. */
 223         lex = makeJsonLexContext(result, false);
 224         pg_parse_json(lex, &nullSemAction);
 225
 226         PG_RETURN_TEXT_P(result);
 227 }
 228
 229 /*
 230  * makeJsonLexContext
 231  *
 232  * lex constructor, with or without StringInfo object
 233  * for de-escaped lexemes.
 234  *
 235  * Without is better as it makes the processing faster, so only make one
 236  * if really required.
 237  */
 238 JsonLexContext *
 239 makeJsonLexContext(text *json, bool need_escapes)
 240 {
 241         JsonLexContext *lex = palloc0(sizeof(JsonLexContext));
 242
 243         lex->input = lex->token_terminator = lex->line_start = VARDATA(json);
 244         lex->line_number = 1;
 245         lex->input_length = VARSIZE(json) - VARHDRSZ;
 246         if (need_escapes)
 247                 lex->strval = makeStringInfo();
 248         return lex;
 249 }
 250
 251 /*
 252  * pg_parse_json
 253  *
 254  * Publicly visible entry point for the JSON parser.
 255  *
 256  * lex is a lexing context, set up for the json to be processed by calling
 257  * makeJsonLexContext(). sem is a strucure of function pointers to semantic
 258  * action routines to be called at appropriate spots during parsing, and a
 259  * pointer to a state object to be passed to those routines.
 260  */
 261 void
 262 pg_parse_json(JsonLexContext *lex, JsonSemAction *sem)
 263 {
 264         JsonTokenType tok;
 265
 266         /* get the initial token */
 267         json_lex(lex);
 268
 269         tok = lex_peek(lex);
 270
 271         /* parse by recursive descent */
 272         switch (tok)
 273         {
 274                 case JSON_TOKEN_OBJECT_START:
 275                         parse_object(lex, sem);
 276                         break;
 277                 case JSON_TOKEN_ARRAY_START:
 278                         parse_array(lex, sem);
 279                         break;
 280                 default:
 281                         parse_scalar(lex, sem);         /* json can be a bare scalar */
 282         }
 283
 284         lex_expect(JSON_PARSE_END, lex, JSON_TOKEN_END);
 285
 286 }
 287
 288 /*
 289  *      Recursive Descent parse routines. There is one for each structural
 290  *      element in a json document:
 291  *        - scalar (string, number, true, false, null)
 292  *        - array  ( [ ] )
 293  *        - array element
 294  *        - object ( { } )
 295  *        - object field
 296  */
 297 static inline void
 298 parse_scalar(JsonLexContext *lex, JsonSemAction *sem)
 299 {
 300         char       *val = NULL;
 301         json_scalar_action sfunc = sem->scalar;
 302         char      **valaddr;
 303         JsonTokenType tok = lex_peek(lex);
 304
 305         valaddr = sfunc == NULL ? NULL : &val;
 306
 307         /* a scalar must be a string, a number, true, false, or null */
 308         switch (tok)
 309         {
 310                 case JSON_TOKEN_TRUE:
 311                         lex_accept(lex, JSON_TOKEN_TRUE, valaddr);
 312                         break;
 313                 case JSON_TOKEN_FALSE:
 314                         lex_accept(lex, JSON_TOKEN_FALSE, valaddr);
 315                         break;
 316                 case JSON_TOKEN_NULL:
 317                         lex_accept(lex, JSON_TOKEN_NULL, valaddr);
 318                         break;
 319                 case JSON_TOKEN_NUMBER:
 320                         lex_accept(lex, JSON_TOKEN_NUMBER, valaddr);
 321                         break;
 322                 case JSON_TOKEN_STRING:
 323                         lex_accept(lex, JSON_TOKEN_STRING, valaddr);
 324                         break;
 325                 default:
 326                         report_parse_error(JSON_PARSE_VALUE, lex);
 327         }
 328
 329         if (sfunc != NULL)
 330                 (*sfunc) (sem->semstate, val, tok);
 331 }
 332
 333 static void
 334 parse_object_field(JsonLexContext *lex, JsonSemAction *sem)
 335 {
 336         /*
 337          * an object field is "fieldname" : value where value can be a scalar,
 338          * object or array
 339          */
 340
 341         char       *fname = NULL;       /* keep compiler quiet */
 342         json_ofield_action ostart = sem->object_field_start;
 343         json_ofield_action oend = sem->object_field_end;
 344         bool            isnull;
 345         char      **fnameaddr = NULL;
 346         JsonTokenType tok;
 347
 348         if (ostart != NULL || oend != NULL)
 349                 fnameaddr = &fname;
 350
 351         if (!lex_accept(lex, JSON_TOKEN_STRING, fnameaddr))
 352                 report_parse_error(JSON_PARSE_STRING, lex);
 353
 354         lex_expect(JSON_PARSE_OBJECT_LABEL, lex, JSON_TOKEN_COLON);
 355
 356         tok = lex_peek(lex);
 357         isnull = tok == JSON_TOKEN_NULL;
 358
 359         if (ostart != NULL)
 360                 (*ostart) (sem->semstate, fname, isnull);
 361
 362         switch (tok)
 363         {
 364                 case JSON_TOKEN_OBJECT_START:
 365                         parse_object(lex, sem);
 366                         break;
 367                 case JSON_TOKEN_ARRAY_START:
 368                         parse_array(lex, sem);
 369                         break;
 370                 default:
 371                         parse_scalar(lex, sem);
 372         }
 373
 374         if (oend != NULL)
 375                 (*oend) (sem->semstate, fname, isnull);
 376
 377         if (fname != NULL)
 378                 pfree(fname);
 379 }
 380
 381 static void
 382 parse_object(JsonLexContext *lex, JsonSemAction *sem)
 383 {
 384         /*
 385          * an object is a possibly empty sequence of object fields, separated by
 386          * commas and surrounde by curly braces.
 387          */
 388         json_struct_action ostart = sem->object_start;
 389         json_struct_action oend = sem->object_end;
 390         JsonTokenType tok;
 391
 392         if (ostart != NULL)
 393                 (*ostart) (sem->semstate);
 394
 395         /*
 396          * Data inside an object at at a higher nesting level than the object
 397          * itself. Note that we increment this after we call the semantic routine
 398          * for the object start and restore it before we call the routine for the
 399          * object end.
 400          */
 401         lex->lex_level++;
 402
 403         /* we know this will succeeed, just clearing the token */
 404         lex_expect(JSON_PARSE_OBJECT_START, lex, JSON_TOKEN_OBJECT_START);
 405
 406         tok = lex_peek(lex);
 407         switch (tok)
 408         {
 409                 case JSON_TOKEN_STRING:
 410                         parse_object_field(lex, sem);
 411                         while (lex_accept(lex, JSON_TOKEN_COMMA, NULL))
 412                                 parse_object_field(lex, sem);
 413                         break;
 414                 case JSON_TOKEN_OBJECT_END:
 415                         break;
 416                 default:
 417                         /* case of an invalid initial token inside the object */
 418                         report_parse_error(JSON_PARSE_OBJECT_START, lex);
 419         }
 420
 421         lex_expect(JSON_PARSE_OBJECT_NEXT, lex, JSON_TOKEN_OBJECT_END);
 422
 423         lex->lex_level--;
 424
 425         if (oend != NULL)
 426                 (*oend) (sem->semstate);
 427 }
 428
 429 static void
 430 parse_array_element(JsonLexContext *lex, JsonSemAction *sem)
 431 {
 432         json_aelem_action astart = sem->array_element_start;
 433         json_aelem_action aend = sem->array_element_end;
 434         JsonTokenType tok = lex_peek(lex);
 435
 436         bool            isnull;
 437
 438         isnull = tok == JSON_TOKEN_NULL;
 439
 440         if (astart != NULL)
 441                 (*astart) (sem->semstate, isnull);
 442
 443         /* an array element is any object, array or scalar */
 444         switch (tok)
 445         {
 446                 case JSON_TOKEN_OBJECT_START:
 447                         parse_object(lex, sem);
 448                         break;
 449                 case JSON_TOKEN_ARRAY_START:
 450                         parse_array(lex, sem);
 451                         break;
 452                 default:
 453                         parse_scalar(lex, sem);
 454         }
 455
 456         if (aend != NULL)
 457                 (*aend) (sem->semstate, isnull);
 458 }
 459
 460 static void
 461 parse_array(JsonLexContext *lex, JsonSemAction *sem)
 462 {
 463         /*
 464          * an array is a possibly empty sequence of array elements, separated by
 465          * commas and surrounded by square brackets.
 466          */
 467         json_struct_action astart = sem->array_start;
 468         json_struct_action aend = sem->array_end;
 469
 470         if (astart != NULL)
 471                 (*astart) (sem->semstate);
 472
 473         /*
 474          * Data inside an array at at a higher nesting level than the array
 475          * itself. Note that we increment this after we call the semantic routine
 476          * for the array start and restore it before we call the routine for the
 477          * array end.
 478          */
 479         lex->lex_level++;
 480
 481         lex_expect(JSON_PARSE_ARRAY_START, lex, JSON_TOKEN_ARRAY_START);
 482         if (lex_peek(lex) != JSON_TOKEN_ARRAY_END)
 483         {
 484
 485                 parse_array_element(lex, sem);
 486
 487                 while (lex_accept(lex, JSON_TOKEN_COMMA, NULL))
 488                         parse_array_element(lex, sem);
 489         }
 490
 491         lex_expect(JSON_PARSE_ARRAY_NEXT, lex, JSON_TOKEN_ARRAY_END);
 492
 493         lex->lex_level--;
 494
 495         if (aend != NULL)
 496                 (*aend) (sem->semstate);
 497 }
 498
 499 /*
 500  * Lex one token from the input stream.
 501  */
 502 static inline void
 503 json_lex(JsonLexContext *lex)
 504 {
 505         char       *s;
 506         int                     len;
 507
 508         /* Skip leading whitespace. */
 509         s = lex->token_terminator;
 510         len = s - lex->input;
 511         while (len < lex->input_length &&
 512                    (*s == ' ' || *s == '\t' || *s == '\n' || *s == '\r'))
 513         {
 514                 if (*s == '\n')
 515                         ++lex->line_number;
 516                 ++s;
 517                 ++len;
 518         }
 519         lex->token_start = s;
 520
 521         /* Determine token type. */
 522         if (len >= lex->input_length)
 523         {
 524                 lex->token_start = NULL;
 525                 lex->prev_token_terminator = lex->token_terminator;
 526                 lex->token_terminator = s;
 527                 lex->token_type = JSON_TOKEN_END;
 528         }
 529         else
 530                 switch (*s)
 531                 {
 532                                 /* Single-character token, some kind of punctuation mark. */
 533                         case '{':
 534                                 lex->prev_token_terminator = lex->token_terminator;
 535                                 lex->token_terminator = s + 1;
 536                                 lex->token_type = JSON_TOKEN_OBJECT_START;
 537                                 break;
 538                         case '}':
 539                                 lex->prev_token_terminator = lex->token_terminator;
 540                                 lex->token_terminator = s + 1;
 541                                 lex->token_type = JSON_TOKEN_OBJECT_END;
 542                                 break;
 543                         case '[':
 544                                 lex->prev_token_terminator = lex->token_terminator;
 545                                 lex->token_terminator = s + 1;
 546                                 lex->token_type = JSON_TOKEN_ARRAY_START;
 547                                 break;
 548                         case ']':
 549                                 lex->prev_token_terminator = lex->token_terminator;
 550                                 lex->token_terminator = s + 1;
 551                                 lex->token_type = JSON_TOKEN_ARRAY_END;
 552                                 break;
 553                         case ',':
 554                                 lex->prev_token_terminator = lex->token_terminator;
 555                                 lex->token_terminator = s + 1;
 556                                 lex->token_type = JSON_TOKEN_COMMA;
 557                                 break;
 558                         case ':':
 559                                 lex->prev_token_terminator = lex->token_terminator;
 560                                 lex->token_terminator = s + 1;
 561                                 lex->token_type = JSON_TOKEN_COLON;
 562                                 break;
 563                         case '"':
 564                                 /* string */
 565                                 json_lex_string(lex);
 566                                 lex->token_type = JSON_TOKEN_STRING;
 567                                 break;
 568                         case '-':
 569                                 /* Negative number. */
 570                                 json_lex_number(lex, s + 1);
 571                                 lex->token_type = JSON_TOKEN_NUMBER;
 572                                 break;
 573                         case '0':
 574                         case '1':
 575                         case '2':
 576                         case '3':
 577                         case '4':
 578                         case '5':
 579                         case '6':
 580                         case '7':
 581                         case '8':
 582                         case '9':
 583                                 /* Positive number. */
 584                                 json_lex_number(lex, s);
 585                                 lex->token_type = JSON_TOKEN_NUMBER;
 586                                 break;
 587                         default:
 588                                 {
 589                                         char       *p;
 590
 591                                         /*
 592                                          * We're not dealing with a string, number, legal
 593                                          * punctuation mark, or end of string.  The only legal
 594                                          * tokens we might find here are true, false, and null,
 595                                          * but for error reporting purposes we scan until we see a
 596                                          * non-alphanumeric character.  That way, we can report
 597                                          * the whole word as an unexpected token, rather than just
 598                                          * some unintuitive prefix thereof.
 599                                          */
 600                                         for (p = s; p - s < lex->input_length - len && JSON_ALPHANUMERIC_CHAR(*p); p++)
 601                                                  /* skip */ ;
 602
 603                                         /*
 604                                          * We got some sort of unexpected punctuation or an
 605                                          * otherwise unexpected character, so just complain about
 606                                          * that one character.
 607                                          */
 608                                         if (p == s)
 609                                         {
 610                                                 lex->prev_token_terminator = lex->token_terminator;
 611                                                 lex->token_terminator = s + 1;
 612                                                 report_invalid_token(lex);
 613                                         }
 614
 615                                         /*
 616                                          * We've got a real alphanumeric token here.  If it
 617                                          * happens to be true, false, or null, all is well.  If
 618                                          * not, error out.
 619                                          */
 620                                         lex->prev_token_terminator = lex->token_terminator;
 621                                         lex->token_terminator = p;
 622                                         if (p - s == 4)
 623                                         {
 624                                                 if (memcmp(s, "true", 4) == 0)
 625                                                         lex->token_type = JSON_TOKEN_TRUE;
 626                                                 else if (memcmp(s, "null", 4) == 0)
 627                                                         lex->token_type = JSON_TOKEN_NULL;
 628                                                 else
 629                                                         report_invalid_token(lex);
 630                                         }
 631                                         else if (p - s == 5 && memcmp(s, "false", 5) == 0)
 632                                                 lex->token_type = JSON_TOKEN_FALSE;
 633                                         else
 634                                                 report_invalid_token(lex);
 635
 636                                 }
 637                 }                                               /* end of switch */
 638 }
 639
 640 /*
 641  * The next token in the input stream is known to be a string; lex it.
 642  */
 643 static inline void
 644 json_lex_string(JsonLexContext *lex)
 645 {
 646         char       *s;
 647         int                     len;
 648         int                     hi_surrogate = -1;
 649
 650         if (lex->strval != NULL)
 651                 resetStringInfo(lex->strval);
 652
 653         Assert(lex->input_length > 0);
 654         s = lex->token_start;
 655         len = lex->token_start - lex->input;
 656         for (;;)
 657         {
 658                 s++;
 659                 len++;
 660                 /* Premature end of the string. */
 661                 if (len >= lex->input_length)
 662                 {
 663                         lex->token_terminator = s;
 664                         report_invalid_token(lex);
 665                 }
 666                 else if (*s == '"')
 667                         break;
 668                 else if ((unsigned char) *s < 32)
 669                 {
 670                         /* Per RFC4627, these characters MUST be escaped. */
 671                         /* Since *s isn't printable, exclude it from the context string */
 672                         lex->token_terminator = s;
 673                         ereport(ERROR,
 674                                         (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
 675                                          errmsg("invalid input syntax for type json"),
 676                                          errdetail("Character with value 0x%02x must be escaped.",
 677                                                            (unsigned char) *s),
 678                                          report_json_context(lex)));
 679                 }
 680                 else if (*s == '\\')
 681                 {
 682                         /* OK, we have an escape character. */
 683                         s++;
 684                         len++;
 685                         if (len >= lex->input_length)
 686                         {
 687                                 lex->token_terminator = s;
 688                                 report_invalid_token(lex);
 689                         }
 690                         else if (*s == 'u')
 691                         {
 692                                 int                     i;
 693                                 int                     ch = 0;
 694
 695                                 for (i = 1; i <= 4; i++)
 696                                 {
 697                                         s++;
 698                                         len++;
 699                                         if (len >= lex->input_length)
 700                                         {
 701                                                 lex->token_terminator = s;
 702                                                 report_invalid_token(lex);
 703                                         }
 704                                         else if (*s >= '0' && *s <= '9')
 705                                                 ch = (ch * 16) + (*s - '0');
 706                                         else if (*s >= 'a' && *s <= 'f')
 707                                                 ch = (ch * 16) + (*s - 'a') + 10;
 708                                         else if (*s >= 'A' && *s <= 'F')
 709                                                 ch = (ch * 16) + (*s - 'A') + 10;
 710                                         else
 711                                         {
 712                                                 lex->token_terminator = s + pg_mblen(s);
 713                                                 ereport(ERROR,
 714                                                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
 715                                                                  errmsg("invalid input syntax for type json"),
 716                                                                  errdetail("\"\\u\" must be followed by four hexadecimal digits."),
 717                                                                  report_json_context(lex)));
 718                                         }
 719                                 }
 720                                 if (lex->strval != NULL)
 721                                 {
 722                                         char            utf8str[5];
 723                                         int                     utf8len;
 724
 725                                         if (ch >= 0xd800 && ch <= 0xdbff)
 726                                         {
 727                                                 if (hi_surrogate != -1)
 728                                                         ereport(ERROR,
 729                                                            (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
 730                                                                 errmsg("invalid input syntax for type json"),
 731                                                                 errdetail("Unicode high surrogate must not follow a high surrogate."),
 732                                                                 report_json_context(lex)));
 733                                                 hi_surrogate = (ch & 0x3ff) << 10;
 734                                                 continue;
 735                                         }
 736                                         else if (ch >= 0xdc00 && ch <= 0xdfff)
 737                                         {
 738                                                 if (hi_surrogate == -1)
 739                                                         ereport(ERROR,
 740                                                            (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
 741                                                                 errmsg("invalid input syntax for type json"),
 742                                                                 errdetail("Unicode low surrogate must follow a high surrogate."),
 743                                                                 report_json_context(lex)));
 744                                                 ch = 0x10000 + hi_surrogate + (ch & 0x3ff);
 745                                                 hi_surrogate = -1;
 746                                         }
 747
 748                                         if (hi_surrogate != -1)
 749                                                 ereport(ERROR,
 750                                                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
 751                                                                  errmsg("invalid input syntax for type json"),
 752                                                                  errdetail("Unicode low surrogate must follow a high surrogate."),
 753                                                                  report_json_context(lex)));
 754
 755                                         /*
 756                                          * For UTF8, replace the escape sequence by the actual utf8
 757                                          * character in lex->strval. Do this also for other encodings
 758                                          * if the escape designates an ASCII character, otherwise
 759                                          * raise an error. We don't ever unescape a \u0000, since that
 760                                          * would result in an impermissible nul byte.
 761                                          */
 762
 763                                         if (ch == 0)
 764                                         {
 765                                                 appendStringInfoString(lex->strval, "\\u0000");
 766                                         }
 767                                         else if (GetDatabaseEncoding() == PG_UTF8)
 768                                         {
 769                                                 unicode_to_utf8(ch, (unsigned char *) utf8str);
 770                                                 utf8len = pg_utf_mblen((unsigned char *) utf8str);
 771                                                 appendBinaryStringInfo(lex->strval, utf8str, utf8len);
 772                                         }
 773                                         else if (ch <= 0x007f)
 774                                         {
 775                                                 /*
 776                                                  * This is the only way to designate things like a form feed
 777                                                  * character in JSON, so it's useful in all encodings.
 778                                                  */
 779                                                 appendStringInfoChar(lex->strval, (char) ch);
 780                                         }
 781                                         else
 782                                         {
 783                                                 ereport(ERROR,
 784                                                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
 785                                                                  errmsg("invalid input syntax for type json"),
 786                                                                  errdetail("Unicode escape values cannot be used for code point values above 007F when the server encoding is not UTF8."),
 787                                                                  report_json_context(lex)));
 788                                         }
 789
 790                                 }
 791                         }
 792                         else if (lex->strval != NULL)
 793                         {
 794                                 if (hi_surrogate != -1)
 795                                         ereport(ERROR,
 796                                                         (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
 797                                                          errmsg("invalid input syntax for type json"),
 798                                                          errdetail("Unicode low surrogate must follow a high surrogate."),
 799                                                          report_json_context(lex)));
 800
 801                                 switch (*s)
 802                                 {
 803                                         case '"':
 804                                         case '\\':
 805                                         case '/':
 806                                                 appendStringInfoChar(lex->strval, *s);
 807                                                 break;
 808                                         case 'b':
 809                                                 appendStringInfoChar(lex->strval, '\b');
 810                                                 break;
 811                                         case 'f':
 812                                                 appendStringInfoChar(lex->strval, '\f');
 813                                                 break;
 814                                         case 'n':
 815                                                 appendStringInfoChar(lex->strval, '\n');
 816                                                 break;
 817                                         case 'r':
 818                                                 appendStringInfoChar(lex->strval, '\r');
 819                                                 break;
 820                                         case 't':
 821                                                 appendStringInfoChar(lex->strval, '\t');
 822                                                 break;
 823                                         default:
 824                                                 /* Not a valid string escape, so error out. */
 825                                                 lex->token_terminator = s + pg_mblen(s);
 826                                                 ereport(ERROR,
 827                                                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
 828                                                                  errmsg("invalid input syntax for type json"),
 829                                                         errdetail("Escape sequence \"\\%s\" is invalid.",
 830                                                                           extract_mb_char(s)),
 831                                                                  report_json_context(lex)));
 832                                 }
 833                         }
 834                         else if (strchr("\"\\/bfnrt", *s) == NULL)
 835                         {
 836                                 /*
 837                                  * Simpler processing if we're not bothered about de-escaping
 838                                  *
 839                                  * It's very tempting to remove the strchr() call here and
 840                                  * replace it with a switch statement, but testing so far has
 841                                  * shown it's not a performance win.
 842                                  */
 843                                 lex->token_terminator = s + pg_mblen(s);
 844                                 ereport(ERROR,
 845                                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
 846                                                  errmsg("invalid input syntax for type json"),
 847                                                  errdetail("Escape sequence \"\\%s\" is invalid.",
 848                                                                    extract_mb_char(s)),
 849                                                  report_json_context(lex)));
 850                         }
 851
 852                 }
 853                 else if (lex->strval != NULL)
 854                 {
 855                         if (hi_surrogate != -1)
 856                                 ereport(ERROR,
 857                                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
 858                                                  errmsg("invalid input syntax for type json"),
 859                                                  errdetail("Unicode low surrogate must follow a high surrogate."),
 860                                                  report_json_context(lex)));
 861
 862                         appendStringInfoChar(lex->strval, *s);
 863                 }
 864
 865         }
 866
 867         if (hi_surrogate != -1)
 868                 ereport(ERROR,
 869                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
 870                                  errmsg("invalid input syntax for type json"),
 871                 errdetail("Unicode low surrogate must follow a high surrogate."),
 872                                  report_json_context(lex)));
 873
 874         /* Hooray, we found the end of the string! */
 875         lex->prev_token_terminator = lex->token_terminator;
 876         lex->token_terminator = s + 1;
 877 }
 878
 879 /*-------------------------------------------------------------------------
 880  * The next token in the input stream is known to be a number; lex it.
 881  *
 882  * In JSON, a number consists of four parts:
 883  *
 884  * (1) An optional minus sign ('-').
 885  *
 886  * (2) Either a single '0', or a string of one or more digits that does not
 887  *         begin with a '0'.
 888  *
 889  * (3) An optional decimal part, consisting of a period ('.') followed by
 890  *         one or more digits.  (Note: While this part can be omitted
 891  *         completely, it's not OK to have only the decimal point without
 892  *         any digits afterwards.)
 893  *
 894  * (4) An optional exponent part, consisting of 'e' or 'E', optionally
 895  *         followed by '+' or '-', followed by one or more digits.      (Note:
 896  *         As with the decimal part, if 'e' or 'E' is present, it must be
 897  *         followed by at least one digit.)
 898  *
 899  * The 's' argument to this function points to the ostensible beginning
 900  * of part 2 - i.e. the character after any optional minus sign, and the
 901  * first character of the string if there is none.
 902  *
 903  *-------------------------------------------------------------------------
 904  */
 905 static inline void
 906 json_lex_number(JsonLexContext *lex, char *s)
 907 {
 908         bool            error = false;
 909         char       *p;
 910         int                     len;
 911
 912         len = s - lex->input;
 913         /* Part (1): leading sign indicator. */
 914         /* Caller already did this for us; so do nothing. */
 915
 916         /* Part (2): parse main digit string. */
 917         if (*s == '0')
 918         {
 919                 s++;
 920                 len++;
 921         }
 922         else if (*s >= '1' && *s <= '9')
 923         {
 924                 do
 925                 {
 926                         s++;
 927                         len++;
 928                 } while (len < lex->input_length && *s >= '0' && *s <= '9');
 929         }
 930         else
 931                 error = true;
 932
 933         /* Part (3): parse optional decimal portion. */
 934         if (len < lex->input_length && *s == '.')
 935         {
 936                 s++;
 937                 len++;
 938                 if (len == lex->input_length || *s < '0' || *s > '9')
 939                         error = true;
 940                 else
 941                 {
 942                         do
 943                         {
 944                                 s++;
 945                                 len++;
 946                         } while (len < lex->input_length && *s >= '0' && *s <= '9');
 947                 }
 948         }
 949
 950         /* Part (4): parse optional exponent. */
 951         if (len < lex->input_length && (*s == 'e' || *s == 'E'))
 952         {
 953                 s++;
 954                 len++;
 955                 if (len < lex->input_length && (*s == '+' || *s == '-'))
 956                 {
 957                         s++;
 958                         len++;
 959                 }
 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         /*
 973          * Check for trailing garbage.  As in json_lex(), any alphanumeric stuff
 974          * here should be considered part of the token for error-reporting
 975          * purposes.
 976          */
 977         for (p = s; len < lex->input_length && JSON_ALPHANUMERIC_CHAR(*p); p++, len++)
 978                 error = true;
 979         lex->prev_token_terminator = lex->token_terminator;
 980         lex->token_terminator = p;
 981         if (error)
 982                 report_invalid_token(lex);
 983 }
 984
 985 /*
 986  * Report a parse error.
 987  *
 988  * lex->token_start and lex->token_terminator must identify the current token.
 989  */
 990 static void
 991 report_parse_error(JsonParseContext ctx, JsonLexContext *lex)
 992 {
 993         char       *token;
 994         int                     toklen;
 995
 996         /* Handle case where the input ended prematurely. */
 997         if (lex->token_start == NULL || lex->token_type == JSON_TOKEN_END)
 998                 ereport(ERROR,
 999                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
1000                                  errmsg("invalid input syntax for type json"),
1001                                  errdetail("The input string ended unexpectedly."),
1002                                  report_json_context(lex)));
1003
1004         /* Separate out the current token. */
1005         toklen = lex->token_terminator - lex->token_start;
1006         token = palloc(toklen + 1);
1007         memcpy(token, lex->token_start, toklen);
1008         token[toklen] = '\0';
1009
1010         /* Complain, with the appropriate detail message. */
1011         if (ctx == JSON_PARSE_END)
1012                 ereport(ERROR,
1013                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
1014                                  errmsg("invalid input syntax for type json"),
1015                                  errdetail("Expected end of input, but found \"%s\".",
1016                                                    token),
1017                                  report_json_context(lex)));
1018         else
1019         {
1020                 switch (ctx)
1021                 {
1022                         case JSON_PARSE_VALUE:
1023                                 ereport(ERROR,
1024                                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
1025                                                  errmsg("invalid input syntax for type json"),
1026                                                  errdetail("Expected JSON value, but found \"%s\".",
1027                                                                    token),
1028                                                  report_json_context(lex)));
1029                                 break;
1030                         case JSON_PARSE_STRING:
1031                                 ereport(ERROR,
1032                                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
1033                                                  errmsg("invalid input syntax for type json"),
1034                                                  errdetail("Expected string, but found \"%s\".",
1035                                                                    token),
1036                                                  report_json_context(lex)));
1037                                 break;
1038                         case JSON_PARSE_ARRAY_START:
1039                                 ereport(ERROR,
1040                                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
1041                                                  errmsg("invalid input syntax for type json"),
1042                                                  errdetail("Expected array element or \"]\", but found \"%s\".",
1043                                                                    token),
1044                                                  report_json_context(lex)));
1045                                 break;
1046                         case JSON_PARSE_ARRAY_NEXT:
1047                                 ereport(ERROR,
1048                                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
1049                                                  errmsg("invalid input syntax for type json"),
1050                                           errdetail("Expected \",\" or \"]\", but found \"%s\".",
1051                                                                 token),
1052                                                  report_json_context(lex)));
1053                                 break;
1054                         case JSON_PARSE_OBJECT_START:
1055                                 ereport(ERROR,
1056                                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
1057                                                  errmsg("invalid input syntax for type json"),
1058                                          errdetail("Expected string or \"}\", but found \"%s\".",
1059                                                            token),
1060                                                  report_json_context(lex)));
1061                                 break;
1062                         case JSON_PARSE_OBJECT_LABEL:
1063                                 ereport(ERROR,
1064                                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
1065                                                  errmsg("invalid input syntax for type json"),
1066                                                  errdetail("Expected \":\", but found \"%s\".",
1067                                                                    token),
1068                                                  report_json_context(lex)));
1069                                 break;
1070                         case JSON_PARSE_OBJECT_NEXT:
1071                                 ereport(ERROR,
1072                                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
1073                                                  errmsg("invalid input syntax for type json"),
1074                                           errdetail("Expected \",\" or \"}\", but found \"%s\".",
1075                                                                 token),
1076                                                  report_json_context(lex)));
1077                                 break;
1078                         case JSON_PARSE_OBJECT_COMMA:
1079                                 ereport(ERROR,
1080                                                 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
1081                                                  errmsg("invalid input syntax for type json"),
1082                                                  errdetail("Expected string, but found \"%s\".",
1083                                                                    token),
1084                                                  report_json_context(lex)));
1085                                 break;
1086                         default:
1087                                 elog(ERROR, "unexpected json parse state: %d", ctx);
1088                 }
1089         }
1090 }
1091
1092 /*
1093  * Report an invalid input token.
1094  *
1095  * lex->token_start and lex->token_terminator must identify the token.
1096  */
1097 static void
1098 report_invalid_token(JsonLexContext *lex)
1099 {
1100         char       *token;
1101         int                     toklen;
1102
1103         /* Separate out the offending token. */
1104         toklen = lex->token_terminator - lex->token_start;
1105         token = palloc(toklen + 1);
1106         memcpy(token, lex->token_start, toklen);
1107         token[toklen] = '\0';
1108
1109         ereport(ERROR,
1110                         (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
1111                          errmsg("invalid input syntax for type json"),
1112                          errdetail("Token \"%s\" is invalid.", token),
1113                          report_json_context(lex)));
1114 }
1115
1116 /*
1117  * Report a CONTEXT line for bogus JSON input.
1118  *
1119  * lex->token_terminator must be set to identify the spot where we detected
1120  * the error.  Note that lex->token_start might be NULL, in case we recognized
1121  * error at EOF.
1122  *
1123  * The return value isn't meaningful, but we make it non-void so that this
1124  * can be invoked inside ereport().
1125  */
1126 static int
1127 report_json_context(JsonLexContext *lex)
1128 {
1129         const char *context_start;
1130         const char *context_end;
1131         const char *line_start;
1132         int                     line_number;
1133         char       *ctxt;
1134         int                     ctxtlen;
1135         const char *prefix;
1136         const char *suffix;
1137
1138         /* Choose boundaries for the part of the input we will display */
1139         context_start = lex->input;
1140         context_end = lex->token_terminator;
1141         line_start = context_start;
1142         line_number = 1;
1143         for (;;)
1144         {
1145                 /* Always advance over newlines */
1146                 if (context_start < context_end && *context_start == '\n')
1147                 {
1148                         context_start++;
1149                         line_start = context_start;
1150                         line_number++;
1151                         continue;
1152                 }
1153                 /* Otherwise, done as soon as we are close enough to context_end */
1154                 if (context_end - context_start < 50)
1155                         break;
1156                 /* Advance to next multibyte character */
1157                 if (IS_HIGHBIT_SET(*context_start))
1158                         context_start += pg_mblen(context_start);
1159                 else
1160                         context_start++;
1161         }
1162
1163         /*
1164          * We add "..." to indicate that the excerpt doesn't start at the
1165          * beginning of the line ... but if we're within 3 characters of the
1166          * beginning of the line, we might as well just show the whole line.
1167          */
1168         if (context_start - line_start <= 3)
1169                 context_start = line_start;
1170
1171         /* Get a null-terminated copy of the data to present */
1172         ctxtlen = context_end - context_start;
1173         ctxt = palloc(ctxtlen + 1);
1174         memcpy(ctxt, context_start, ctxtlen);
1175         ctxt[ctxtlen] = '\0';
1176
1177         /*
1178          * Show the context, prefixing "..." if not starting at start of line, and
1179          * suffixing "..." if not ending at end of line.
1180          */
1181         prefix = (context_start > line_start) ? "..." : "";
1182         suffix = (lex->token_type != JSON_TOKEN_END && context_end - lex->input < lex->input_length && *context_end != '\n' && *context_end != '\r') ? "..." : "";
1183
1184         return errcontext("JSON data, line %d: %s%s%s",
1185                                           line_number, prefix, ctxt, suffix);
1186 }
1187
1188 /*
1189  * Extract a single, possibly multi-byte char from the input string.
1190  */
1191 static char *
1192 extract_mb_char(char *s)
1193 {
1194         char       *res;
1195         int                     len;
1196
1197         len = pg_mblen(s);
1198         res = palloc(len + 1);
1199         memcpy(res, s, len);
1200         res[len] = '\0';
1201
1202         return res;
1203 }
1204
1205 /*
1206  * Turn a scalar Datum into JSON, appending the string to "result".
1207  *
1208  * Hand off a non-scalar datum to composite_to_json or array_to_json_internal
1209  * as appropriate.
1210  */
1211 static void
1212 datum_to_json(Datum val, bool is_null, StringInfo result,
1213                           TYPCATEGORY tcategory, Oid typoutputfunc)
1214 {
1215         char       *outputstr;
1216         text       *jsontext;
1217
1218         if (is_null)
1219         {
1220                 appendStringInfoString(result, "null");
1221                 return;
1222         }
1223
1224         switch (tcategory)
1225         {
1226                 case TYPCATEGORY_ARRAY:
1227                         array_to_json_internal(val, result, false);
1228                         break;
1229                 case TYPCATEGORY_COMPOSITE:
1230                         composite_to_json(val, result, false);
1231                         break;
1232                 case TYPCATEGORY_BOOLEAN:
1233                         if (DatumGetBool(val))
1234                                 appendStringInfoString(result, "true");
1235                         else
1236                                 appendStringInfoString(result, "false");
1237                         break;
1238                 case TYPCATEGORY_NUMERIC:
1239                         outputstr = OidOutputFunctionCall(typoutputfunc, val);
1240
1241                         /*
1242                          * Don't call escape_json here if it's a valid JSON number.
1243                          * Numeric output should usually be a valid JSON number and JSON
1244                          * numbers shouldn't be quoted. Quote cases like "Nan" and
1245                          * "Infinity", however.
1246                          */
1247                         if (strpbrk(outputstr, NON_NUMERIC_LETTER) == NULL)
1248                                 appendStringInfoString(result, outputstr);
1249                         else
1250                                 escape_json(result, outputstr);
1251                         pfree(outputstr);
1252                         break;
1253                 case TYPCATEGORY_JSON:
1254                         /* JSON will already be escaped */
1255                         outputstr = OidOutputFunctionCall(typoutputfunc, val);
1256                         appendStringInfoString(result, outputstr);
1257                         pfree(outputstr);
1258                         break;
1259                 case TYPCATEGORY_JSON_CAST:
1260                         jsontext = DatumGetTextP(OidFunctionCall1(typoutputfunc, val));
1261                         outputstr = text_to_cstring(jsontext);
1262                         appendStringInfoString(result, outputstr);
1263                         pfree(outputstr);
1264                         pfree(jsontext);
1265                         break;
1266                 default:
1267                         outputstr = OidOutputFunctionCall(typoutputfunc, val);
1268                         escape_json(result, outputstr);
1269                         pfree(outputstr);
1270                         break;
1271         }
1272 }
1273
1274 /*
1275  * Process a single dimension of an array.
1276  * If it's the innermost dimension, output the values, otherwise call
1277  * ourselves recursively to process the next dimension.
1278  */
1279 static void
1280 array_dim_to_json(StringInfo result, int dim, int ndims, int *dims, Datum *vals,
1281                                   bool *nulls, int *valcount, TYPCATEGORY tcategory,
1282                                   Oid typoutputfunc, bool use_line_feeds)
1283 {
1284         int                     i;
1285         const char *sep;
1286
1287         Assert(dim < ndims);
1288
1289         sep = use_line_feeds ? ",\n " : ",";
1290
1291         appendStringInfoChar(result, '[');
1292
1293         for (i = 1; i <= dims[dim]; i++)
1294         {
1295                 if (i > 1)
1296                         appendStringInfoString(result, sep);
1297
1298                 if (dim + 1 == ndims)
1299                 {
1300                         datum_to_json(vals[*valcount], nulls[*valcount], result, tcategory,
1301                                                   typoutputfunc);
1302                         (*valcount)++;
1303                 }
1304                 else
1305                 {
1306                         /*
1307                          * Do we want line feeds on inner dimensions of arrays? For now
1308                          * we'll say no.
1309                          */
1310                         array_dim_to_json(result, dim + 1, ndims, dims, vals, nulls,
1311                                                           valcount, tcategory, typoutputfunc, false);
1312                 }
1313         }
1314
1315         appendStringInfoChar(result, ']');
1316 }
1317
1318 /*
1319  * Turn an array into JSON.
1320  */
1321 static void
1322 array_to_json_internal(Datum array, StringInfo result, bool use_line_feeds)
1323 {
1324         ArrayType  *v = DatumGetArrayTypeP(array);
1325         Oid                     element_type = ARR_ELEMTYPE(v);
1326         int                *dim;
1327         int                     ndim;
1328         int                     nitems;
1329         int                     count = 0;
1330         Datum      *elements;
1331         bool       *nulls;
1332         int16           typlen;
1333         bool            typbyval;
1334         char            typalign,
1335                                 typdelim;
1336         Oid                     typioparam;
1337         Oid                     typoutputfunc;
1338         TYPCATEGORY tcategory;
1339         Oid                     castfunc = InvalidOid;
1340
1341         ndim = ARR_NDIM(v);
1342         dim = ARR_DIMS(v);
1343         nitems = ArrayGetNItems(ndim, dim);
1344
1345         if (nitems <= 0)
1346         {
1347                 appendStringInfoString(result, "[]");
1348                 return;
1349         }
1350
1351         get_type_io_data(element_type, IOFunc_output,
1352                                          &typlen, &typbyval, &typalign,
1353                                          &typdelim, &typioparam, &typoutputfunc);
1354
1355         if (element_type > FirstNormalObjectId)
1356         {
1357                 HeapTuple       tuple;
1358                 Form_pg_cast castForm;
1359
1360                 tuple = SearchSysCache2(CASTSOURCETARGET,
1361                                                                 ObjectIdGetDatum(element_type),
1362                                                                 ObjectIdGetDatum(JSONOID));
1363                 if (HeapTupleIsValid(tuple))
1364                 {
1365                         castForm = (Form_pg_cast) GETSTRUCT(tuple);
1366
1367                         if (castForm->castmethod == COERCION_METHOD_FUNCTION)
1368                                 castfunc = typoutputfunc = castForm->castfunc;
1369
1370                         ReleaseSysCache(tuple);
1371                 }
1372         }
1373
1374         deconstruct_array(v, element_type, typlen, typbyval,
1375                                           typalign, &elements, &nulls,
1376                                           &nitems);
1377
1378         if (castfunc != InvalidOid)
1379                 tcategory = TYPCATEGORY_JSON_CAST;
1380         else if (element_type == RECORDOID)
1381                 tcategory = TYPCATEGORY_COMPOSITE;
1382         else if (element_type == JSONOID)
1383                 tcategory = TYPCATEGORY_JSON;
1384         else
1385                 tcategory = TypeCategory(element_type);
1386
1387         array_dim_to_json(result, 0, ndim, dim, elements, nulls, &count, tcategory,
1388                                           typoutputfunc, use_line_feeds);
1389
1390         pfree(elements);
1391         pfree(nulls);
1392 }
1393
1394 /*
1395  * Turn a composite / record into JSON.
1396  */
1397 static void
1398 composite_to_json(Datum composite, StringInfo result, bool use_line_feeds)
1399 {
1400         HeapTupleHeader td;
1401         Oid                     tupType;
1402         int32           tupTypmod;
1403         TupleDesc       tupdesc;
1404         HeapTupleData tmptup,
1405                            *tuple;
1406         int                     i;
1407         bool            needsep = false;
1408         const char *sep;
1409
1410         sep = use_line_feeds ? ",\n " : ",";
1411
1412         td = DatumGetHeapTupleHeader(composite);
1413
1414         /* Extract rowtype info and find a tupdesc */
1415         tupType = HeapTupleHeaderGetTypeId(td);
1416         tupTypmod = HeapTupleHeaderGetTypMod(td);
1417         tupdesc = lookup_rowtype_tupdesc(tupType, tupTypmod);
1418
1419         /* Build a temporary HeapTuple control structure */
1420         tmptup.t_len = HeapTupleHeaderGetDatumLength(td);
1421         tmptup.t_data = td;
1422         tuple = &tmptup;
1423
1424         appendStringInfoChar(result, '{');
1425
1426         for (i = 0; i < tupdesc->natts; i++)
1427         {
1428                 Datum           val,
1429                                         origval;
1430                 bool            isnull;
1431                 char       *attname;
1432                 TYPCATEGORY tcategory;
1433                 Oid                     typoutput;
1434                 bool            typisvarlena;
1435                 Oid                     castfunc = InvalidOid;
1436
1437                 if (tupdesc->attrs[i]->attisdropped)
1438                         continue;
1439
1440                 if (needsep)
1441                         appendStringInfoString(result, sep);
1442                 needsep = true;
1443
1444                 attname = NameStr(tupdesc->attrs[i]->attname);
1445                 escape_json(result, attname);
1446                 appendStringInfoChar(result, ':');
1447
1448                 origval = heap_getattr(tuple, i + 1, tupdesc, &isnull);
1449
1450                 getTypeOutputInfo(tupdesc->attrs[i]->atttypid,
1451                                                   &typoutput, &typisvarlena);
1452
1453                 if (tupdesc->attrs[i]->atttypid > FirstNormalObjectId)
1454                 {
1455                         HeapTuple       cast_tuple;
1456                         Form_pg_cast castForm;
1457
1458                         cast_tuple = SearchSysCache2(CASTSOURCETARGET,
1459                                                            ObjectIdGetDatum(tupdesc->attrs[i]->atttypid),
1460                                                                                  ObjectIdGetDatum(JSONOID));
1461                         if (HeapTupleIsValid(cast_tuple))
1462                         {
1463                                 castForm = (Form_pg_cast) GETSTRUCT(cast_tuple);
1464
1465                                 if (castForm->castmethod == COERCION_METHOD_FUNCTION)
1466                                         castfunc = typoutput = castForm->castfunc;
1467
1468                                 ReleaseSysCache(cast_tuple);
1469                         }
1470                 }
1471
1472                 if (castfunc != InvalidOid)
1473                         tcategory = TYPCATEGORY_JSON_CAST;
1474                 else if (tupdesc->attrs[i]->atttypid == RECORDARRAYOID)
1475                         tcategory = TYPCATEGORY_ARRAY;
1476                 else if (tupdesc->attrs[i]->atttypid == RECORDOID)
1477                         tcategory = TYPCATEGORY_COMPOSITE;
1478                 else if (tupdesc->attrs[i]->atttypid == JSONOID)
1479                         tcategory = TYPCATEGORY_JSON;
1480                 else
1481                         tcategory = TypeCategory(tupdesc->attrs[i]->atttypid);
1482
1483                 /*
1484                  * If we have a toasted datum, forcibly detoast it here to avoid
1485                  * memory leakage inside the type's output routine.
1486                  */
1487                 if (typisvarlena && !isnull)
1488                         val = PointerGetDatum(PG_DETOAST_DATUM(origval));
1489                 else
1490                         val = origval;
1491
1492                 datum_to_json(val, isnull, result, tcategory, typoutput);
1493
1494                 /* Clean up detoasted copy, if any */
1495                 if (val != origval)
1496                         pfree(DatumGetPointer(val));
1497         }
1498
1499         appendStringInfoChar(result, '}');
1500         ReleaseTupleDesc(tupdesc);
1501 }
1502
1503 /*
1504  * SQL function array_to_json(row)
1505  */
1506 extern Datum
1507 array_to_json(PG_FUNCTION_ARGS)
1508 {
1509         Datum           array = PG_GETARG_DATUM(0);
1510         StringInfo      result;
1511
1512         result = makeStringInfo();
1513
1514         array_to_json_internal(array, result, false);
1515
1516         PG_RETURN_TEXT_P(cstring_to_text(result->data));
1517 }
1518
1519 /*
1520  * SQL function array_to_json(row, prettybool)
1521  */
1522 extern Datum
1523 array_to_json_pretty(PG_FUNCTION_ARGS)
1524 {
1525         Datum           array = PG_GETARG_DATUM(0);
1526         bool            use_line_feeds = PG_GETARG_BOOL(1);
1527         StringInfo      result;
1528
1529         result = makeStringInfo();
1530
1531         array_to_json_internal(array, result, use_line_feeds);
1532
1533         PG_RETURN_TEXT_P(cstring_to_text(result->data));
1534 }
1535
1536 /*
1537  * SQL function row_to_json(row)
1538  */
1539 extern Datum
1540 row_to_json(PG_FUNCTION_ARGS)
1541 {
1542         Datum           array = PG_GETARG_DATUM(0);
1543         StringInfo      result;
1544
1545         result = makeStringInfo();
1546
1547         composite_to_json(array, result, false);
1548
1549         PG_RETURN_TEXT_P(cstring_to_text(result->data));
1550 }
1551
1552 /*
1553  * SQL function row_to_json(row, prettybool)
1554  */
1555 extern Datum
1556 row_to_json_pretty(PG_FUNCTION_ARGS)
1557 {
1558         Datum           array = PG_GETARG_DATUM(0);
1559         bool            use_line_feeds = PG_GETARG_BOOL(1);
1560         StringInfo      result;
1561
1562         result = makeStringInfo();
1563
1564         composite_to_json(array, result, use_line_feeds);
1565
1566         PG_RETURN_TEXT_P(cstring_to_text(result->data));
1567 }
1568
1569 /*
1570  * SQL function to_json(anyvalue)
1571  */
1572 Datum
1573 to_json(PG_FUNCTION_ARGS)
1574 {
1575         Oid                     val_type = get_fn_expr_argtype(fcinfo->flinfo, 0);
1576         StringInfo      result;
1577         Datum           orig_val,
1578                                 val;
1579         TYPCATEGORY tcategory;
1580         Oid                     typoutput;
1581         bool            typisvarlena;
1582         Oid                     castfunc = InvalidOid;
1583
1584         if (val_type == InvalidOid)
1585                 ereport(ERROR,
1586                                 (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1587                                  errmsg("could not determine input data type")));
1588
1589
1590         result = makeStringInfo();
1591
1592         orig_val = PG_ARGISNULL(0) ? (Datum) 0 : PG_GETARG_DATUM(0);
1593
1594         getTypeOutputInfo(val_type, &typoutput, &typisvarlena);
1595
1596         if (val_type > FirstNormalObjectId)
1597         {
1598                 HeapTuple       tuple;
1599                 Form_pg_cast castForm;
1600
1601                 tuple = SearchSysCache2(CASTSOURCETARGET,
1602                                                                 ObjectIdGetDatum(val_type),
1603                                                                 ObjectIdGetDatum(JSONOID));
1604                 if (HeapTupleIsValid(tuple))
1605                 {
1606                         castForm = (Form_pg_cast) GETSTRUCT(tuple);
1607
1608                         if (castForm->castmethod == COERCION_METHOD_FUNCTION)
1609                                 castfunc = typoutput = castForm->castfunc;
1610
1611                         ReleaseSysCache(tuple);
1612                 }
1613         }
1614
1615         if (castfunc != InvalidOid)
1616                 tcategory = TYPCATEGORY_JSON_CAST;
1617         else if (val_type == RECORDARRAYOID)
1618                 tcategory = TYPCATEGORY_ARRAY;
1619         else if (val_type == RECORDOID)
1620                 tcategory = TYPCATEGORY_COMPOSITE;
1621         else if (val_type == JSONOID)
1622                 tcategory = TYPCATEGORY_JSON;
1623         else
1624                 tcategory = TypeCategory(val_type);
1625
1626         /*
1627          * If we have a toasted datum, forcibly detoast it here to avoid memory
1628          * leakage inside the type's output routine.
1629          */
1630         if (typisvarlena && orig_val != (Datum) 0)
1631                 val = PointerGetDatum(PG_DETOAST_DATUM(orig_val));
1632         else
1633                 val = orig_val;
1634
1635         datum_to_json(val, false, result, tcategory, typoutput);
1636
1637         /* Clean up detoasted copy, if any */
1638         if (val != orig_val)
1639                 pfree(DatumGetPointer(val));
1640
1641         PG_RETURN_TEXT_P(cstring_to_text(result->data));
1642 }
1643
1644 /*
1645  * json_agg transition function
1646  */
1647 Datum
1648 json_agg_transfn(PG_FUNCTION_ARGS)
1649 {
1650         Oid                     val_type = get_fn_expr_argtype(fcinfo->flinfo, 1);
1651         MemoryContext aggcontext,
1652                                 oldcontext;
1653         StringInfo      state;
1654         Datum           orig_val,
1655                                 val;
1656         TYPCATEGORY tcategory;
1657         Oid                     typoutput;
1658         bool            typisvarlena;
1659         Oid                     castfunc = InvalidOid;
1660
1661         if (val_type == InvalidOid)
1662                 ereport(ERROR,
1663                                 (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1664                                  errmsg("could not determine input data type")));
1665
1666         if (!AggCheckCallContext(fcinfo, &aggcontext))
1667         {
1668                 /* cannot be called directly because of internal-type argument */
1669                 elog(ERROR, "json_agg_transfn called in non-aggregate context");
1670         }
1671
1672         if (PG_ARGISNULL(0))
1673         {
1674                 /*
1675                  * Make this StringInfo in a context where it will persist for the
1676                  * duration off the aggregate call. It's only needed for this initial
1677                  * piece, as the StringInfo routines make sure they use the right
1678                  * context to enlarge the object if necessary.
1679                  */
1680                 oldcontext = MemoryContextSwitchTo(aggcontext);
1681                 state = makeStringInfo();
1682                 MemoryContextSwitchTo(oldcontext);
1683
1684                 appendStringInfoChar(state, '[');
1685         }
1686         else
1687         {
1688                 state = (StringInfo) PG_GETARG_POINTER(0);
1689                 appendStringInfoString(state, ", ");
1690         }
1691
1692         /* fast path for NULLs */
1693         if (PG_ARGISNULL(1))
1694         {
1695                 orig_val = (Datum) 0;
1696                 datum_to_json(orig_val, true, state, 0, InvalidOid);
1697                 PG_RETURN_POINTER(state);
1698         }
1699
1700
1701         orig_val = PG_GETARG_DATUM(1);
1702
1703         getTypeOutputInfo(val_type, &typoutput, &typisvarlena);
1704
1705         if (val_type > FirstNormalObjectId)
1706         {
1707                 HeapTuple       tuple;
1708                 Form_pg_cast castForm;
1709
1710                 tuple = SearchSysCache2(CASTSOURCETARGET,
1711                                                                 ObjectIdGetDatum(val_type),
1712                                                                 ObjectIdGetDatum(JSONOID));
1713                 if (HeapTupleIsValid(tuple))
1714                 {
1715                         castForm = (Form_pg_cast) GETSTRUCT(tuple);
1716
1717                         if (castForm->castmethod == COERCION_METHOD_FUNCTION)
1718                                 castfunc = typoutput = castForm->castfunc;
1719
1720                         ReleaseSysCache(tuple);
1721                 }
1722         }
1723
1724         if (castfunc != InvalidOid)
1725                 tcategory = TYPCATEGORY_JSON_CAST;
1726         else if (val_type == RECORDARRAYOID)
1727                 tcategory = TYPCATEGORY_ARRAY;
1728         else if (val_type == RECORDOID)
1729                 tcategory = TYPCATEGORY_COMPOSITE;
1730         else if (val_type == JSONOID)
1731                 tcategory = TYPCATEGORY_JSON;
1732         else
1733                 tcategory = TypeCategory(val_type);
1734
1735         /*
1736          * If we have a toasted datum, forcibly detoast it here to avoid memory
1737          * leakage inside the type's output routine.
1738          */
1739         if (typisvarlena)
1740                 val = PointerGetDatum(PG_DETOAST_DATUM(orig_val));
1741         else
1742                 val = orig_val;
1743
1744         if (!PG_ARGISNULL(0) &&
1745           (tcategory == TYPCATEGORY_ARRAY || tcategory == TYPCATEGORY_COMPOSITE))
1746         {
1747                 appendStringInfoString(state, "\n ");
1748         }
1749
1750         datum_to_json(val, false, state, tcategory, typoutput);
1751
1752         /* Clean up detoasted copy, if any */
1753         if (val != orig_val)
1754                 pfree(DatumGetPointer(val));
1755
1756         /*
1757          * The transition type for array_agg() is declared to be "internal", which
1758          * is a pass-by-value type the same size as a pointer.  So we can safely
1759          * pass the ArrayBuildState pointer through nodeAgg.c's machinations.
1760          */
1761         PG_RETURN_POINTER(state);
1762 }
1763
1764 /*
1765  * json_agg final function
1766  */
1767 Datum
1768 json_agg_finalfn(PG_FUNCTION_ARGS)
1769 {
1770         StringInfo      state;
1771
1772         /* cannot be called directly because of internal-type argument */
1773         Assert(AggCheckCallContext(fcinfo, NULL));
1774
1775         state = PG_ARGISNULL(0) ? NULL : (StringInfo) PG_GETARG_POINTER(0);
1776
1777         if (state == NULL)
1778                 PG_RETURN_NULL();
1779
1780         appendStringInfoChar(state, ']');
1781
1782         PG_RETURN_TEXT_P(cstring_to_text(state->data));
1783 }
1784
1785 /*
1786  * Produce a JSON string literal, properly escaping characters in the text.
1787  */
1788 void
1789 escape_json(StringInfo buf, const char *str)
1790 {
1791         const char *p;
1792
1793         appendStringInfoCharMacro(buf, '\"');
1794         for (p = str; *p; p++)
1795         {
1796                 switch (*p)
1797                 {
1798                         case '\b':
1799                                 appendStringInfoString(buf, "\\b");
1800                                 break;
1801                         case '\f':
1802                                 appendStringInfoString(buf, "\\f");
1803                                 break;
1804                         case '\n':
1805                                 appendStringInfoString(buf, "\\n");
1806                                 break;
1807                         case '\r':
1808                                 appendStringInfoString(buf, "\\r");
1809                                 break;
1810                         case '\t':
1811                                 appendStringInfoString(buf, "\\t");
1812                                 break;
1813                         case '"':
1814                                 appendStringInfoString(buf, "\\\"");
1815                                 break;
1816                         case '\\':
1817                                 appendStringInfoString(buf, "\\\\");
1818                                 break;
1819                         default:
1820                                 if ((unsigned char) *p < ' ')
1821                                         appendStringInfo(buf, "\\u%04x", (int) *p);
1822                                 else
1823                                         appendStringInfoCharMacro(buf, *p);
1824                                 break;
1825                 }
1826         }
1827         appendStringInfoCharMacro(buf, '\"');
1828 }