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