1 /*-------------------------------------------------------------------------
4 * JSON data type support.
6 * Portions Copyright (c) 1996-2016, PostgreSQL Global Development Group
7 * Portions Copyright (c) 1994, Regents of the University of California
10 * src/backend/utils/adt/json.c
12 *-------------------------------------------------------------------------
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"
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.
40 typedef enum /* contexts of JSON parser */
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 */
53 typedef enum /* type categories for datum_to_json */
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 */
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 */
68 typedef struct JsonAggState
71 JsonTypeCategory key_category;
73 JsonTypeCategory val_category;
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,
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,
96 static void array_to_json_internal(Datum array, StringInfo result,
98 static void json_categorize_type(Oid typoid,
99 JsonTypeCategory *tcategory,
101 static void datum_to_json(Datum val, bool is_null, StringInfo result,
102 JsonTypeCategory tcategory, Oid outfuncoid,
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);
108 /* the null action object used for pure validation */
109 static JsonSemAction nullSemAction =
111 NULL, NULL, NULL, NULL, NULL,
112 NULL, NULL, NULL, NULL, NULL
115 /* Recursive Descent parser support routines */
120 * what is the current look_ahead token?
122 static inline JsonTokenType
123 lex_peek(JsonLexContext *lex)
125 return lex->token_type;
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.
135 * returns true if the token matched, false otherwise.
138 lex_accept(JsonLexContext *lex, JsonTokenType token, char **lexeme)
140 if (lex->token_type == token)
144 if (lex->token_type == JSON_TOKEN_STRING)
146 if (lex->strval != NULL)
147 *lexeme = pstrdup(lex->strval->data);
151 int len = (lex->token_terminator - lex->token_start);
152 char *tokstr = palloc(len + 1);
154 memcpy(tokstr, lex->token_start, len);
168 * move the lexer to the next token if the current look_ahead token matches
169 * the parameter token. Otherwise, report an error.
172 lex_expect(JsonParseContext ctx, JsonLexContext *lex, JsonTokenType token)
174 if (!lex_accept(lex, token, NULL))
175 report_parse_error(ctx, lex);
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') || \
187 * Utility function to check if a string is a valid JSON number.
189 * str is of length len, and need not be null-terminated.
192 IsValidJsonNumber(const char *str, int len)
196 JsonLexContext dummy_lex;
202 * json_lex_number expects a leading '-' to have been eaten already.
204 * having to cast away the constness of str is ugly, but there's not much
209 dummy_lex.input = (char *) str + 1;
210 dummy_lex.input_length = len - 1;
214 dummy_lex.input = (char *) str;
215 dummy_lex.input_length = len;
218 json_lex_number(&dummy_lex, dummy_lex.input, &numeric_error, &total_len);
220 return (!numeric_error) && (total_len == dummy_lex.input_length);
227 json_in(PG_FUNCTION_ARGS)
229 char *json = PG_GETARG_CSTRING(0);
230 text *result = cstring_to_text(json);
234 lex = makeJsonLexContext(result, false);
235 pg_parse_json(lex, &nullSemAction);
237 /* Internal representation is the same as text, for now */
238 PG_RETURN_TEXT_P(result);
245 json_out(PG_FUNCTION_ARGS)
247 /* we needn't detoast because text_to_cstring will handle that */
248 Datum txt = PG_GETARG_DATUM(0);
250 PG_RETURN_CSTRING(TextDatumGetCString(txt));
257 json_send(PG_FUNCTION_ARGS)
259 text *t = PG_GETARG_TEXT_PP(0);
262 pq_begintypsend(&buf);
263 pq_sendtext(&buf, VARDATA_ANY(t), VARSIZE_ANY_EXHDR(t));
264 PG_RETURN_BYTEA_P(pq_endtypsend(&buf));
271 json_recv(PG_FUNCTION_ARGS)
273 StringInfo buf = (StringInfo) PG_GETARG_POINTER(0);
278 str = pq_getmsgtext(buf, buf->len - buf->cursor, &nbytes);
281 lex = makeJsonLexContextCstringLen(str, nbytes, false);
282 pg_parse_json(lex, &nullSemAction);
284 PG_RETURN_TEXT_P(cstring_to_text_with_len(str, nbytes));
290 * lex constructor, with or without StringInfo object
291 * for de-escaped lexemes.
293 * Without is better as it makes the processing faster, so only make one
294 * if really required.
296 * If you already have the json as a text* value, use the first of these
297 * functions, otherwise use makeJsonLexContextCstringLen().
300 makeJsonLexContext(text *json, bool need_escapes)
302 return makeJsonLexContextCstringLen(VARDATA(json),
303 VARSIZE(json) - VARHDRSZ,
308 makeJsonLexContextCstringLen(char *json, int len, bool need_escapes)
310 JsonLexContext *lex = palloc0(sizeof(JsonLexContext));
312 lex->input = lex->token_terminator = lex->line_start = json;
313 lex->line_number = 1;
314 lex->input_length = len;
316 lex->strval = makeStringInfo();
323 * Publicly visible entry point for the JSON parser.
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.
331 pg_parse_json(JsonLexContext *lex, JsonSemAction *sem)
335 /* get the initial token */
340 /* parse by recursive descent */
343 case JSON_TOKEN_OBJECT_START:
344 parse_object(lex, sem);
346 case JSON_TOKEN_ARRAY_START:
347 parse_array(lex, sem);
350 parse_scalar(lex, sem); /* json can be a bare scalar */
353 lex_expect(JSON_PARSE_END, lex, JSON_TOKEN_END);
358 * json_count_array_elements
360 * Returns number of array elements in lex context at start of array token
361 * until end of array token at same nesting level.
363 * Designed to be called from array_start routines.
366 json_count_array_elements(JsonLexContext *lex)
368 JsonLexContext copylex;
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.
376 memcpy(©lex, lex, sizeof(JsonLexContext));
377 copylex.strval = NULL; /* not interested in values here */
381 lex_expect(JSON_PARSE_ARRAY_START, ©lex, JSON_TOKEN_ARRAY_START);
382 if (lex_peek(©lex) != JSON_TOKEN_ARRAY_END)
387 parse_array_element(©lex, &nullSemAction);
389 while (lex_accept(©lex, JSON_TOKEN_COMMA, NULL));
391 lex_expect(JSON_PARSE_ARRAY_NEXT, ©lex, JSON_TOKEN_ARRAY_END);
397 * Recursive Descent parse routines. There is one for each structural
398 * element in a json document:
399 * - scalar (string, number, true, false, null)
406 parse_scalar(JsonLexContext *lex, JsonSemAction *sem)
409 json_scalar_action sfunc = sem->scalar;
411 JsonTokenType tok = lex_peek(lex);
413 valaddr = sfunc == NULL ? NULL : &val;
415 /* a scalar must be a string, a number, true, false, or null */
418 case JSON_TOKEN_TRUE:
419 lex_accept(lex, JSON_TOKEN_TRUE, valaddr);
421 case JSON_TOKEN_FALSE:
422 lex_accept(lex, JSON_TOKEN_FALSE, valaddr);
424 case JSON_TOKEN_NULL:
425 lex_accept(lex, JSON_TOKEN_NULL, valaddr);
427 case JSON_TOKEN_NUMBER:
428 lex_accept(lex, JSON_TOKEN_NUMBER, valaddr);
430 case JSON_TOKEN_STRING:
431 lex_accept(lex, JSON_TOKEN_STRING, valaddr);
434 report_parse_error(JSON_PARSE_VALUE, lex);
438 (*sfunc) (sem->semstate, val, tok);
442 parse_object_field(JsonLexContext *lex, JsonSemAction *sem)
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".
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;
454 char **fnameaddr = NULL;
457 if (ostart != NULL || oend != NULL)
460 if (!lex_accept(lex, JSON_TOKEN_STRING, fnameaddr))
461 report_parse_error(JSON_PARSE_STRING, lex);
463 lex_expect(JSON_PARSE_OBJECT_LABEL, lex, JSON_TOKEN_COLON);
466 isnull = tok == JSON_TOKEN_NULL;
469 (*ostart) (sem->semstate, fname, isnull);
473 case JSON_TOKEN_OBJECT_START:
474 parse_object(lex, sem);
476 case JSON_TOKEN_ARRAY_START:
477 parse_array(lex, sem);
480 parse_scalar(lex, sem);
484 (*oend) (sem->semstate, fname, isnull);
488 parse_object(JsonLexContext *lex, JsonSemAction *sem)
491 * an object is a possibly empty sequence of object fields, separated by
492 * commas and surrounded by curly braces.
494 json_struct_action ostart = sem->object_start;
495 json_struct_action oend = sem->object_end;
501 (*ostart) (sem->semstate);
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
511 /* we know this will succeeed, just clearing the token */
512 lex_expect(JSON_PARSE_OBJECT_START, lex, JSON_TOKEN_OBJECT_START);
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);
522 case JSON_TOKEN_OBJECT_END:
525 /* case of an invalid initial token inside the object */
526 report_parse_error(JSON_PARSE_OBJECT_START, lex);
529 lex_expect(JSON_PARSE_OBJECT_NEXT, lex, JSON_TOKEN_OBJECT_END);
534 (*oend) (sem->semstate);
538 parse_array_element(JsonLexContext *lex, JsonSemAction *sem)
540 json_aelem_action astart = sem->array_element_start;
541 json_aelem_action aend = sem->array_element_end;
542 JsonTokenType tok = lex_peek(lex);
546 isnull = tok == JSON_TOKEN_NULL;
549 (*astart) (sem->semstate, isnull);
551 /* an array element is any object, array or scalar */
554 case JSON_TOKEN_OBJECT_START:
555 parse_object(lex, sem);
557 case JSON_TOKEN_ARRAY_START:
558 parse_array(lex, sem);
561 parse_scalar(lex, sem);
565 (*aend) (sem->semstate, isnull);
569 parse_array(JsonLexContext *lex, JsonSemAction *sem)
572 * an array is a possibly empty sequence of array elements, separated by
573 * commas and surrounded by square brackets.
575 json_struct_action astart = sem->array_start;
576 json_struct_action aend = sem->array_end;
581 (*astart) (sem->semstate);
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
591 lex_expect(JSON_PARSE_ARRAY_START, lex, JSON_TOKEN_ARRAY_START);
592 if (lex_peek(lex) != JSON_TOKEN_ARRAY_END)
595 parse_array_element(lex, sem);
597 while (lex_accept(lex, JSON_TOKEN_COMMA, NULL))
598 parse_array_element(lex, sem);
601 lex_expect(JSON_PARSE_ARRAY_NEXT, lex, JSON_TOKEN_ARRAY_END);
606 (*aend) (sem->semstate);
610 * Lex one token from the input stream.
613 json_lex(JsonLexContext *lex)
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'))
629 lex->token_start = s;
631 /* Determine token type. */
632 if (len >= lex->input_length)
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;
642 /* Single-character token, some kind of punctuation mark. */
644 lex->prev_token_terminator = lex->token_terminator;
645 lex->token_terminator = s + 1;
646 lex->token_type = JSON_TOKEN_OBJECT_START;
649 lex->prev_token_terminator = lex->token_terminator;
650 lex->token_terminator = s + 1;
651 lex->token_type = JSON_TOKEN_OBJECT_END;
654 lex->prev_token_terminator = lex->token_terminator;
655 lex->token_terminator = s + 1;
656 lex->token_type = JSON_TOKEN_ARRAY_START;
659 lex->prev_token_terminator = lex->token_terminator;
660 lex->token_terminator = s + 1;
661 lex->token_type = JSON_TOKEN_ARRAY_END;
664 lex->prev_token_terminator = lex->token_terminator;
665 lex->token_terminator = s + 1;
666 lex->token_type = JSON_TOKEN_COMMA;
669 lex->prev_token_terminator = lex->token_terminator;
670 lex->token_terminator = s + 1;
671 lex->token_type = JSON_TOKEN_COLON;
675 json_lex_string(lex);
676 lex->token_type = JSON_TOKEN_STRING;
679 /* Negative number. */
680 json_lex_number(lex, s + 1, NULL, NULL);
681 lex->token_type = JSON_TOKEN_NUMBER;
693 /* Positive number. */
694 json_lex_number(lex, s, NULL, NULL);
695 lex->token_type = JSON_TOKEN_NUMBER;
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.
710 for (p = s; p - s < lex->input_length - len && JSON_ALPHANUMERIC_CHAR(*p); p++)
714 * We got some sort of unexpected punctuation or an
715 * otherwise unexpected character, so just complain about
716 * that one character.
720 lex->prev_token_terminator = lex->token_terminator;
721 lex->token_terminator = s + 1;
722 report_invalid_token(lex);
726 * We've got a real alphanumeric token here. If it
727 * happens to be true, false, or null, all is well. If
730 lex->prev_token_terminator = lex->token_terminator;
731 lex->token_terminator = p;
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;
739 report_invalid_token(lex);
741 else if (p - s == 5 && memcmp(s, "false", 5) == 0)
742 lex->token_type = JSON_TOKEN_FALSE;
744 report_invalid_token(lex);
747 } /* end of switch */
751 * The next token in the input stream is known to be a string; lex it.
754 json_lex_string(JsonLexContext *lex)
758 int hi_surrogate = -1;
760 if (lex->strval != NULL)
761 resetStringInfo(lex->strval);
763 Assert(lex->input_length > 0);
764 s = lex->token_start;
765 len = lex->token_start - lex->input;
770 /* Premature end of the string. */
771 if (len >= lex->input_length)
773 lex->token_terminator = s;
774 report_invalid_token(lex);
778 else if ((unsigned char) *s < 32)
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;
784 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
785 errmsg("invalid input syntax for type json"),
786 errdetail("Character with value 0x%02x must be escaped.",
788 report_json_context(lex)));
792 /* OK, we have an escape character. */
795 if (len >= lex->input_length)
797 lex->token_terminator = s;
798 report_invalid_token(lex);
805 for (i = 1; i <= 4; i++)
809 if (len >= lex->input_length)
811 lex->token_terminator = s;
812 report_invalid_token(lex);
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;
822 lex->token_terminator = s + pg_mblen(s);
824 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
825 errmsg("invalid input syntax for type json"),
826 errdetail("\"\\u\" must be followed by four hexadecimal digits."),
827 report_json_context(lex)));
830 if (lex->strval != NULL)
835 if (ch >= 0xd800 && ch <= 0xdbff)
837 if (hi_surrogate != -1)
839 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
840 errmsg("invalid input syntax for type json"),
841 errdetail("Unicode high surrogate must not follow a high surrogate."),
842 report_json_context(lex)));
843 hi_surrogate = (ch & 0x3ff) << 10;
846 else if (ch >= 0xdc00 && ch <= 0xdfff)
848 if (hi_surrogate == -1)
850 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
851 errmsg("invalid input syntax for type json"),
852 errdetail("Unicode low surrogate must follow a high surrogate."),
853 report_json_context(lex)));
854 ch = 0x10000 + hi_surrogate + (ch & 0x3ff);
858 if (hi_surrogate != -1)
860 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
861 errmsg("invalid input syntax for type json"),
862 errdetail("Unicode low surrogate must follow a high surrogate."),
863 report_json_context(lex)));
866 * For UTF8, replace the escape sequence by the actual
867 * utf8 character in lex->strval. Do this also for other
868 * encodings if the escape designates an ASCII character,
869 * otherwise raise an error.
874 /* We can't allow this, since our TEXT type doesn't */
876 (errcode(ERRCODE_UNTRANSLATABLE_CHARACTER),
877 errmsg("unsupported Unicode escape sequence"),
878 errdetail("\\u0000 cannot be converted to text."),
879 report_json_context(lex)));
881 else if (GetDatabaseEncoding() == PG_UTF8)
883 unicode_to_utf8(ch, (unsigned char *) utf8str);
884 utf8len = pg_utf_mblen((unsigned char *) utf8str);
885 appendBinaryStringInfo(lex->strval, utf8str, utf8len);
887 else if (ch <= 0x007f)
890 * This is the only way to designate things like a
891 * form feed character in JSON, so it's useful in all
894 appendStringInfoChar(lex->strval, (char) ch);
899 (errcode(ERRCODE_UNTRANSLATABLE_CHARACTER),
900 errmsg("unsupported Unicode escape sequence"),
901 errdetail("Unicode escape values cannot be used for code point values above 007F when the server encoding is not UTF8."),
902 report_json_context(lex)));
907 else if (lex->strval != NULL)
909 if (hi_surrogate != -1)
911 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
912 errmsg("invalid input syntax for type json"),
913 errdetail("Unicode low surrogate must follow a high surrogate."),
914 report_json_context(lex)));
921 appendStringInfoChar(lex->strval, *s);
924 appendStringInfoChar(lex->strval, '\b');
927 appendStringInfoChar(lex->strval, '\f');
930 appendStringInfoChar(lex->strval, '\n');
933 appendStringInfoChar(lex->strval, '\r');
936 appendStringInfoChar(lex->strval, '\t');
939 /* Not a valid string escape, so error out. */
940 lex->token_terminator = s + pg_mblen(s);
942 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
943 errmsg("invalid input syntax for type json"),
944 errdetail("Escape sequence \"\\%s\" is invalid.",
946 report_json_context(lex)));
949 else if (strchr("\"\\/bfnrt", *s) == NULL)
952 * Simpler processing if we're not bothered about de-escaping
954 * It's very tempting to remove the strchr() call here and
955 * replace it with a switch statement, but testing so far has
956 * shown it's not a performance win.
958 lex->token_terminator = s + pg_mblen(s);
960 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
961 errmsg("invalid input syntax for type json"),
962 errdetail("Escape sequence \"\\%s\" is invalid.",
964 report_json_context(lex)));
968 else if (lex->strval != NULL)
970 if (hi_surrogate != -1)
972 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
973 errmsg("invalid input syntax for type json"),
974 errdetail("Unicode low surrogate must follow a high surrogate."),
975 report_json_context(lex)));
977 appendStringInfoChar(lex->strval, *s);
982 if (hi_surrogate != -1)
984 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
985 errmsg("invalid input syntax for type json"),
986 errdetail("Unicode low surrogate must follow a high surrogate."),
987 report_json_context(lex)));
989 /* Hooray, we found the end of the string! */
990 lex->prev_token_terminator = lex->token_terminator;
991 lex->token_terminator = s + 1;
995 * The next token in the input stream is known to be a number; lex it.
997 * In JSON, a number consists of four parts:
999 * (1) An optional minus sign ('-').
1001 * (2) Either a single '0', or a string of one or more digits that does not
1004 * (3) An optional decimal part, consisting of a period ('.') followed by
1005 * one or more digits. (Note: While this part can be omitted
1006 * completely, it's not OK to have only the decimal point without
1007 * any digits afterwards.)
1009 * (4) An optional exponent part, consisting of 'e' or 'E', optionally
1010 * followed by '+' or '-', followed by one or more digits. (Note:
1011 * As with the decimal part, if 'e' or 'E' is present, it must be
1012 * followed by at least one digit.)
1014 * The 's' argument to this function points to the ostensible beginning
1015 * of part 2 - i.e. the character after any optional minus sign, or the
1016 * first character of the string if there is none.
1018 * If num_err is not NULL, we return an error flag to *num_err rather than
1019 * raising an error for a badly-formed number. Also, if total_len is not NULL
1020 * the distance from lex->input to the token end+1 is returned to *total_len.
1023 json_lex_number(JsonLexContext *lex, char *s,
1024 bool *num_err, int *total_len)
1027 int len = s - lex->input;
1029 /* Part (1): leading sign indicator. */
1030 /* Caller already did this for us; so do nothing. */
1032 /* Part (2): parse main digit string. */
1033 if (len < lex->input_length && *s == '0')
1038 else if (len < lex->input_length && *s >= '1' && *s <= '9')
1044 } while (len < lex->input_length && *s >= '0' && *s <= '9');
1049 /* Part (3): parse optional decimal portion. */
1050 if (len < lex->input_length && *s == '.')
1054 if (len == lex->input_length || *s < '0' || *s > '9')
1062 } while (len < lex->input_length && *s >= '0' && *s <= '9');
1066 /* Part (4): parse optional exponent. */
1067 if (len < lex->input_length && (*s == 'e' || *s == 'E'))
1071 if (len < lex->input_length && (*s == '+' || *s == '-'))
1076 if (len == lex->input_length || *s < '0' || *s > '9')
1084 } while (len < lex->input_length && *s >= '0' && *s <= '9');
1089 * Check for trailing garbage. As in json_lex(), any alphanumeric stuff
1090 * here should be considered part of the token for error-reporting
1093 for (; len < lex->input_length && JSON_ALPHANUMERIC_CHAR(*s); s++, len++)
1096 if (total_len != NULL)
1099 if (num_err != NULL)
1101 /* let the caller handle any error */
1106 /* return token endpoint */
1107 lex->prev_token_terminator = lex->token_terminator;
1108 lex->token_terminator = s;
1109 /* handle error if any */
1111 report_invalid_token(lex);
1116 * Report a parse error.
1118 * lex->token_start and lex->token_terminator must identify the current token.
1121 report_parse_error(JsonParseContext ctx, JsonLexContext *lex)
1126 /* Handle case where the input ended prematurely. */
1127 if (lex->token_start == NULL || lex->token_type == JSON_TOKEN_END)
1129 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
1130 errmsg("invalid input syntax for type json"),
1131 errdetail("The input string ended unexpectedly."),
1132 report_json_context(lex)));
1134 /* Separate out the current token. */
1135 toklen = lex->token_terminator - lex->token_start;
1136 token = palloc(toklen + 1);
1137 memcpy(token, lex->token_start, toklen);
1138 token[toklen] = '\0';
1140 /* Complain, with the appropriate detail message. */
1141 if (ctx == JSON_PARSE_END)
1143 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
1144 errmsg("invalid input syntax for type json"),
1145 errdetail("Expected end of input, but found \"%s\".",
1147 report_json_context(lex)));
1152 case JSON_PARSE_VALUE:
1154 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
1155 errmsg("invalid input syntax for type json"),
1156 errdetail("Expected JSON value, but found \"%s\".",
1158 report_json_context(lex)));
1160 case JSON_PARSE_STRING:
1162 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
1163 errmsg("invalid input syntax for type json"),
1164 errdetail("Expected string, but found \"%s\".",
1166 report_json_context(lex)));
1168 case JSON_PARSE_ARRAY_START:
1170 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
1171 errmsg("invalid input syntax for type json"),
1172 errdetail("Expected array element or \"]\", but found \"%s\".",
1174 report_json_context(lex)));
1176 case JSON_PARSE_ARRAY_NEXT:
1178 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
1179 errmsg("invalid input syntax for type json"),
1180 errdetail("Expected \",\" or \"]\", but found \"%s\".",
1182 report_json_context(lex)));
1184 case JSON_PARSE_OBJECT_START:
1186 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
1187 errmsg("invalid input syntax for type json"),
1188 errdetail("Expected string or \"}\", but found \"%s\".",
1190 report_json_context(lex)));
1192 case JSON_PARSE_OBJECT_LABEL:
1194 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
1195 errmsg("invalid input syntax for type json"),
1196 errdetail("Expected \":\", but found \"%s\".",
1198 report_json_context(lex)));
1200 case JSON_PARSE_OBJECT_NEXT:
1202 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
1203 errmsg("invalid input syntax for type json"),
1204 errdetail("Expected \",\" or \"}\", but found \"%s\".",
1206 report_json_context(lex)));
1208 case JSON_PARSE_OBJECT_COMMA:
1210 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
1211 errmsg("invalid input syntax for type json"),
1212 errdetail("Expected string, but found \"%s\".",
1214 report_json_context(lex)));
1217 elog(ERROR, "unexpected json parse state: %d", ctx);
1223 * Report an invalid input token.
1225 * lex->token_start and lex->token_terminator must identify the token.
1228 report_invalid_token(JsonLexContext *lex)
1233 /* Separate out the offending token. */
1234 toklen = lex->token_terminator - lex->token_start;
1235 token = palloc(toklen + 1);
1236 memcpy(token, lex->token_start, toklen);
1237 token[toklen] = '\0';
1240 (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
1241 errmsg("invalid input syntax for type json"),
1242 errdetail("Token \"%s\" is invalid.", token),
1243 report_json_context(lex)));
1247 * Report a CONTEXT line for bogus JSON input.
1249 * lex->token_terminator must be set to identify the spot where we detected
1250 * the error. Note that lex->token_start might be NULL, in case we recognized
1253 * The return value isn't meaningful, but we make it non-void so that this
1254 * can be invoked inside ereport().
1257 report_json_context(JsonLexContext *lex)
1259 const char *context_start;
1260 const char *context_end;
1261 const char *line_start;
1268 /* Choose boundaries for the part of the input we will display */
1269 context_start = lex->input;
1270 context_end = lex->token_terminator;
1271 line_start = context_start;
1275 /* Always advance over newlines */
1276 if (context_start < context_end && *context_start == '\n')
1279 line_start = context_start;
1283 /* Otherwise, done as soon as we are close enough to context_end */
1284 if (context_end - context_start < 50)
1286 /* Advance to next multibyte character */
1287 if (IS_HIGHBIT_SET(*context_start))
1288 context_start += pg_mblen(context_start);
1294 * We add "..." to indicate that the excerpt doesn't start at the
1295 * beginning of the line ... but if we're within 3 characters of the
1296 * beginning of the line, we might as well just show the whole line.
1298 if (context_start - line_start <= 3)
1299 context_start = line_start;
1301 /* Get a null-terminated copy of the data to present */
1302 ctxtlen = context_end - context_start;
1303 ctxt = palloc(ctxtlen + 1);
1304 memcpy(ctxt, context_start, ctxtlen);
1305 ctxt[ctxtlen] = '\0';
1308 * Show the context, prefixing "..." if not starting at start of line, and
1309 * suffixing "..." if not ending at end of line.
1311 prefix = (context_start > line_start) ? "..." : "";
1312 suffix = (lex->token_type != JSON_TOKEN_END && context_end - lex->input < lex->input_length && *context_end != '\n' && *context_end != '\r') ? "..." : "";
1314 return errcontext("JSON data, line %d: %s%s%s",
1315 line_number, prefix, ctxt, suffix);
1319 * Extract a single, possibly multi-byte char from the input string.
1322 extract_mb_char(char *s)
1328 res = palloc(len + 1);
1329 memcpy(res, s, len);
1336 * Determine how we want to print values of a given type in datum_to_json.
1338 * Given the datatype OID, return its JsonTypeCategory, as well as the type's
1339 * output function OID. If the returned category is JSONTYPE_CAST, we
1340 * return the OID of the type->JSON cast function instead.
1343 json_categorize_type(Oid typoid,
1344 JsonTypeCategory *tcategory,
1349 /* Look through any domain */
1350 typoid = getBaseType(typoid);
1352 *outfuncoid = InvalidOid;
1355 * We need to get the output function for everything except date and
1356 * timestamp types, array and composite types, booleans, and non-builtin
1357 * types where there's a cast to json.
1363 *tcategory = JSONTYPE_BOOL;
1372 getTypeOutputInfo(typoid, outfuncoid, &typisvarlena);
1373 *tcategory = JSONTYPE_NUMERIC;
1377 *tcategory = JSONTYPE_DATE;
1381 *tcategory = JSONTYPE_TIMESTAMP;
1384 case TIMESTAMPTZOID:
1385 *tcategory = JSONTYPE_TIMESTAMPTZ;
1390 getTypeOutputInfo(typoid, outfuncoid, &typisvarlena);
1391 *tcategory = JSONTYPE_JSON;
1395 /* Check for arrays and composites */
1396 if (OidIsValid(get_element_type(typoid)))
1397 *tcategory = JSONTYPE_ARRAY;
1398 else if (type_is_rowtype(typoid))
1399 *tcategory = JSONTYPE_COMPOSITE;
1402 /* It's probably the general case ... */
1403 *tcategory = JSONTYPE_OTHER;
1404 /* but let's look for a cast to json, if it's not built-in */
1405 if (typoid >= FirstNormalObjectId)
1408 CoercionPathType ctype;
1410 ctype = find_coercion_pathway(JSONOID, typoid,
1413 if (ctype == COERCION_PATH_FUNC && OidIsValid(castfunc))
1415 *tcategory = JSONTYPE_CAST;
1416 *outfuncoid = castfunc;
1420 /* non builtin type with no cast */
1421 getTypeOutputInfo(typoid, outfuncoid, &typisvarlena);
1426 /* any other builtin type */
1427 getTypeOutputInfo(typoid, outfuncoid, &typisvarlena);
1435 * Turn a Datum into JSON text, appending the string to "result".
1437 * tcategory and outfuncoid are from a previous call to json_categorize_type,
1438 * except that if is_null is true then they can be invalid.
1440 * If key_scalar is true, the value is being printed as a key, so insist
1441 * it's of an acceptable type, and force it to be quoted.
1444 datum_to_json(Datum val, bool is_null, StringInfo result,
1445 JsonTypeCategory tcategory, Oid outfuncoid,
1451 check_stack_depth();
1453 /* callers are expected to ensure that null keys are not passed in */
1454 Assert(!(key_scalar && is_null));
1458 appendStringInfoString(result, "null");
1463 (tcategory == JSONTYPE_ARRAY ||
1464 tcategory == JSONTYPE_COMPOSITE ||
1465 tcategory == JSONTYPE_JSON ||
1466 tcategory == JSONTYPE_CAST))
1468 (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1469 errmsg("key value must be scalar, not array, composite, or json")));
1473 case JSONTYPE_ARRAY:
1474 array_to_json_internal(val, result, false);
1476 case JSONTYPE_COMPOSITE:
1477 composite_to_json(val, result, false);
1480 outputstr = DatumGetBool(val) ? "true" : "false";
1482 escape_json(result, outputstr);
1484 appendStringInfoString(result, outputstr);
1486 case JSONTYPE_NUMERIC:
1487 outputstr = OidOutputFunctionCall(outfuncoid, val);
1490 * Don't call escape_json for a non-key if it's a valid JSON
1493 if (!key_scalar && IsValidJsonNumber(outputstr, strlen(outputstr)))
1494 appendStringInfoString(result, outputstr);
1496 escape_json(result, outputstr);
1503 char buf[MAXDATELEN + 1];
1505 date = DatumGetDateADT(val);
1506 /* Same as date_out(), but forcing DateStyle */
1507 if (DATE_NOT_FINITE(date))
1508 EncodeSpecialDate(date, buf);
1511 j2date(date + POSTGRES_EPOCH_JDATE,
1512 &(tm.tm_year), &(tm.tm_mon), &(tm.tm_mday));
1513 EncodeDateOnly(&tm, USE_XSD_DATES, buf);
1515 appendStringInfo(result, "\"%s\"", buf);
1518 case JSONTYPE_TIMESTAMP:
1520 Timestamp timestamp;
1523 char buf[MAXDATELEN + 1];
1525 timestamp = DatumGetTimestamp(val);
1526 /* Same as timestamp_out(), but forcing DateStyle */
1527 if (TIMESTAMP_NOT_FINITE(timestamp))
1528 EncodeSpecialTimestamp(timestamp, buf);
1529 else if (timestamp2tm(timestamp, NULL, &tm, &fsec, NULL, NULL) == 0)
1530 EncodeDateTime(&tm, fsec, false, 0, NULL, USE_XSD_DATES, buf);
1533 (errcode(ERRCODE_DATETIME_VALUE_OUT_OF_RANGE),
1534 errmsg("timestamp out of range")));
1535 appendStringInfo(result, "\"%s\"", buf);
1538 case JSONTYPE_TIMESTAMPTZ:
1540 TimestampTz timestamp;
1544 const char *tzn = NULL;
1545 char buf[MAXDATELEN + 1];
1547 timestamp = DatumGetTimestampTz(val);
1548 /* Same as timestamptz_out(), but forcing DateStyle */
1549 if (TIMESTAMP_NOT_FINITE(timestamp))
1550 EncodeSpecialTimestamp(timestamp, buf);
1551 else if (timestamp2tm(timestamp, &tz, &tm, &fsec, &tzn, NULL) == 0)
1552 EncodeDateTime(&tm, fsec, true, tz, tzn, USE_XSD_DATES, buf);
1555 (errcode(ERRCODE_DATETIME_VALUE_OUT_OF_RANGE),
1556 errmsg("timestamp out of range")));
1557 appendStringInfo(result, "\"%s\"", buf);
1561 /* JSON and JSONB output will already be escaped */
1562 outputstr = OidOutputFunctionCall(outfuncoid, val);
1563 appendStringInfoString(result, outputstr);
1567 /* outfuncoid refers to a cast function, not an output function */
1568 jsontext = DatumGetTextP(OidFunctionCall1(outfuncoid, val));
1569 outputstr = text_to_cstring(jsontext);
1570 appendStringInfoString(result, outputstr);
1575 outputstr = OidOutputFunctionCall(outfuncoid, val);
1576 escape_json(result, outputstr);
1583 * Process a single dimension of an array.
1584 * If it's the innermost dimension, output the values, otherwise call
1585 * ourselves recursively to process the next dimension.
1588 array_dim_to_json(StringInfo result, int dim, int ndims, int *dims, Datum *vals,
1589 bool *nulls, int *valcount, JsonTypeCategory tcategory,
1590 Oid outfuncoid, bool use_line_feeds)
1595 Assert(dim < ndims);
1597 sep = use_line_feeds ? ",\n " : ",";
1599 appendStringInfoChar(result, '[');
1601 for (i = 1; i <= dims[dim]; i++)
1604 appendStringInfoString(result, sep);
1606 if (dim + 1 == ndims)
1608 datum_to_json(vals[*valcount], nulls[*valcount], result, tcategory,
1615 * Do we want line feeds on inner dimensions of arrays? For now
1618 array_dim_to_json(result, dim + 1, ndims, dims, vals, nulls,
1619 valcount, tcategory, outfuncoid, false);
1623 appendStringInfoChar(result, ']');
1627 * Turn an array into JSON.
1630 array_to_json_internal(Datum array, StringInfo result, bool use_line_feeds)
1632 ArrayType *v = DatumGetArrayTypeP(array);
1633 Oid element_type = ARR_ELEMTYPE(v);
1643 JsonTypeCategory tcategory;
1648 nitems = ArrayGetNItems(ndim, dim);
1652 appendStringInfoString(result, "[]");
1656 get_typlenbyvalalign(element_type,
1657 &typlen, &typbyval, &typalign);
1659 json_categorize_type(element_type,
1660 &tcategory, &outfuncoid);
1662 deconstruct_array(v, element_type, typlen, typbyval,
1663 typalign, &elements, &nulls,
1666 array_dim_to_json(result, 0, ndim, dim, elements, nulls, &count, tcategory,
1667 outfuncoid, use_line_feeds);
1674 * Turn a composite / record into JSON.
1677 composite_to_json(Datum composite, StringInfo result, bool use_line_feeds)
1683 HeapTupleData tmptup,
1686 bool needsep = false;
1689 sep = use_line_feeds ? ",\n " : ",";
1691 td = DatumGetHeapTupleHeader(composite);
1693 /* Extract rowtype info and find a tupdesc */
1694 tupType = HeapTupleHeaderGetTypeId(td);
1695 tupTypmod = HeapTupleHeaderGetTypMod(td);
1696 tupdesc = lookup_rowtype_tupdesc(tupType, tupTypmod);
1698 /* Build a temporary HeapTuple control structure */
1699 tmptup.t_len = HeapTupleHeaderGetDatumLength(td);
1703 appendStringInfoChar(result, '{');
1705 for (i = 0; i < tupdesc->natts; i++)
1710 JsonTypeCategory tcategory;
1713 if (tupdesc->attrs[i]->attisdropped)
1717 appendStringInfoString(result, sep);
1720 attname = NameStr(tupdesc->attrs[i]->attname);
1721 escape_json(result, attname);
1722 appendStringInfoChar(result, ':');
1724 val = heap_getattr(tuple, i + 1, tupdesc, &isnull);
1728 tcategory = JSONTYPE_NULL;
1729 outfuncoid = InvalidOid;
1732 json_categorize_type(tupdesc->attrs[i]->atttypid,
1733 &tcategory, &outfuncoid);
1735 datum_to_json(val, isnull, result, tcategory, outfuncoid, false);
1738 appendStringInfoChar(result, '}');
1739 ReleaseTupleDesc(tupdesc);
1743 * Append JSON text for "val" to "result".
1745 * This is just a thin wrapper around datum_to_json. If the same type will be
1746 * printed many times, avoid using this; better to do the json_categorize_type
1747 * lookups only once.
1750 add_json(Datum val, bool is_null, StringInfo result,
1751 Oid val_type, bool key_scalar)
1753 JsonTypeCategory tcategory;
1756 if (val_type == InvalidOid)
1758 (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1759 errmsg("could not determine input data type")));
1763 tcategory = JSONTYPE_NULL;
1764 outfuncoid = InvalidOid;
1767 json_categorize_type(val_type,
1768 &tcategory, &outfuncoid);
1770 datum_to_json(val, is_null, result, tcategory, outfuncoid, key_scalar);
1774 * SQL function array_to_json(row)
1777 array_to_json(PG_FUNCTION_ARGS)
1779 Datum array = PG_GETARG_DATUM(0);
1782 result = makeStringInfo();
1784 array_to_json_internal(array, result, false);
1786 PG_RETURN_TEXT_P(cstring_to_text_with_len(result->data, result->len));
1790 * SQL function array_to_json(row, prettybool)
1793 array_to_json_pretty(PG_FUNCTION_ARGS)
1795 Datum array = PG_GETARG_DATUM(0);
1796 bool use_line_feeds = PG_GETARG_BOOL(1);
1799 result = makeStringInfo();
1801 array_to_json_internal(array, result, use_line_feeds);
1803 PG_RETURN_TEXT_P(cstring_to_text_with_len(result->data, result->len));
1807 * SQL function row_to_json(row)
1810 row_to_json(PG_FUNCTION_ARGS)
1812 Datum array = PG_GETARG_DATUM(0);
1815 result = makeStringInfo();
1817 composite_to_json(array, result, false);
1819 PG_RETURN_TEXT_P(cstring_to_text_with_len(result->data, result->len));
1823 * SQL function row_to_json(row, prettybool)
1826 row_to_json_pretty(PG_FUNCTION_ARGS)
1828 Datum array = PG_GETARG_DATUM(0);
1829 bool use_line_feeds = PG_GETARG_BOOL(1);
1832 result = makeStringInfo();
1834 composite_to_json(array, result, use_line_feeds);
1836 PG_RETURN_TEXT_P(cstring_to_text_with_len(result->data, result->len));
1840 * SQL function to_json(anyvalue)
1843 to_json(PG_FUNCTION_ARGS)
1845 Datum val = PG_GETARG_DATUM(0);
1846 Oid val_type = get_fn_expr_argtype(fcinfo->flinfo, 0);
1848 JsonTypeCategory tcategory;
1851 if (val_type == InvalidOid)
1853 (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1854 errmsg("could not determine input data type")));
1856 json_categorize_type(val_type,
1857 &tcategory, &outfuncoid);
1859 result = makeStringInfo();
1861 datum_to_json(val, false, result, tcategory, outfuncoid, false);
1863 PG_RETURN_TEXT_P(cstring_to_text_with_len(result->data, result->len));
1867 * json_agg transition function
1869 * aggregate input column as a json array value.
1872 json_agg_transfn(PG_FUNCTION_ARGS)
1874 MemoryContext aggcontext,
1876 JsonAggState *state;
1879 if (!AggCheckCallContext(fcinfo, &aggcontext))
1881 /* cannot be called directly because of internal-type argument */
1882 elog(ERROR, "json_agg_transfn called in non-aggregate context");
1885 if (PG_ARGISNULL(0))
1887 Oid arg_type = get_fn_expr_argtype(fcinfo->flinfo, 1);
1889 if (arg_type == InvalidOid)
1891 (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1892 errmsg("could not determine input data type")));
1895 * Make this state object in a context where it will persist for the
1896 * duration of the aggregate call. MemoryContextSwitchTo is only
1897 * needed the first time, as the StringInfo routines make sure they
1898 * use the right context to enlarge the object if necessary.
1900 oldcontext = MemoryContextSwitchTo(aggcontext);
1901 state = (JsonAggState *) palloc(sizeof(JsonAggState));
1902 state->str = makeStringInfo();
1903 MemoryContextSwitchTo(oldcontext);
1905 appendStringInfoChar(state->str, '[');
1906 json_categorize_type(arg_type, &state->val_category,
1907 &state->val_output_func);
1911 state = (JsonAggState *) PG_GETARG_POINTER(0);
1912 appendStringInfoString(state->str, ", ");
1915 /* fast path for NULLs */
1916 if (PG_ARGISNULL(1))
1918 datum_to_json((Datum) 0, true, state->str, JSONTYPE_NULL,
1920 PG_RETURN_POINTER(state);
1923 val = PG_GETARG_DATUM(1);
1925 /* add some whitespace if structured type and not first item */
1926 if (!PG_ARGISNULL(0) &&
1927 (state->val_category == JSONTYPE_ARRAY ||
1928 state->val_category == JSONTYPE_COMPOSITE))
1930 appendStringInfoString(state->str, "\n ");
1933 datum_to_json(val, false, state->str, state->val_category,
1934 state->val_output_func, false);
1937 * The transition type for array_agg() is declared to be "internal", which
1938 * is a pass-by-value type the same size as a pointer. So we can safely
1939 * pass the JsonAggState pointer through nodeAgg.c's machinations.
1941 PG_RETURN_POINTER(state);
1945 * json_agg final function
1948 json_agg_finalfn(PG_FUNCTION_ARGS)
1950 JsonAggState *state;
1952 /* cannot be called directly because of internal-type argument */
1953 Assert(AggCheckCallContext(fcinfo, NULL));
1955 state = PG_ARGISNULL(0) ?
1957 (JsonAggState *) PG_GETARG_POINTER(0);
1959 /* NULL result for no rows in, as is standard with aggregates */
1963 /* Else return state with appropriate array terminator added */
1964 PG_RETURN_TEXT_P(catenate_stringinfo_string(state->str, "]"));
1968 * json_object_agg transition function.
1970 * aggregate two input columns as a single json object value.
1973 json_object_agg_transfn(PG_FUNCTION_ARGS)
1975 MemoryContext aggcontext,
1977 JsonAggState *state;
1980 if (!AggCheckCallContext(fcinfo, &aggcontext))
1982 /* cannot be called directly because of internal-type argument */
1983 elog(ERROR, "json_object_agg_transfn called in non-aggregate context");
1986 if (PG_ARGISNULL(0))
1991 * Make the StringInfo in a context where it will persist for the
1992 * duration of the aggregate call. Switching context is only needed
1993 * for this initial step, as the StringInfo routines make sure they
1994 * use the right context to enlarge the object if necessary.
1996 oldcontext = MemoryContextSwitchTo(aggcontext);
1997 state = (JsonAggState *) palloc(sizeof(JsonAggState));
1998 state->str = makeStringInfo();
1999 MemoryContextSwitchTo(oldcontext);
2001 arg_type = get_fn_expr_argtype(fcinfo->flinfo, 1);
2003 if (arg_type == InvalidOid)
2005 (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
2006 errmsg("could not determine data type for argument 1")));
2008 json_categorize_type(arg_type, &state->key_category,
2009 &state->key_output_func);
2011 arg_type = get_fn_expr_argtype(fcinfo->flinfo, 2);
2013 if (arg_type == InvalidOid)
2015 (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
2016 errmsg("could not determine data type for argument 2")));
2018 json_categorize_type(arg_type, &state->val_category,
2019 &state->val_output_func);
2021 appendStringInfoString(state->str, "{ ");
2025 state = (JsonAggState *) PG_GETARG_POINTER(0);
2026 appendStringInfoString(state->str, ", ");
2030 * Note: since json_object_agg() is declared as taking type "any", the
2031 * parser will not do any type conversion on unknown-type literals (that
2032 * is, undecorated strings or NULLs). Such values will arrive here as
2033 * type UNKNOWN, which fortunately does not matter to us, since
2034 * unknownout() works fine.
2037 if (PG_ARGISNULL(1))
2039 (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
2040 errmsg("field name must not be null")));
2042 arg = PG_GETARG_DATUM(1);
2044 datum_to_json(arg, false, state->str, state->key_category,
2045 state->key_output_func, true);
2047 appendStringInfoString(state->str, " : ");
2049 if (PG_ARGISNULL(2))
2052 arg = PG_GETARG_DATUM(2);
2054 datum_to_json(arg, PG_ARGISNULL(2), state->str, state->val_category,
2055 state->val_output_func, false);
2057 PG_RETURN_POINTER(state);
2061 * json_object_agg final function.
2064 json_object_agg_finalfn(PG_FUNCTION_ARGS)
2066 JsonAggState *state;
2068 /* cannot be called directly because of internal-type argument */
2069 Assert(AggCheckCallContext(fcinfo, NULL));
2071 state = PG_ARGISNULL(0) ? NULL : (JsonAggState *) PG_GETARG_POINTER(0);
2073 /* NULL result for no rows in, as is standard with aggregates */
2077 /* Else return state with appropriate object terminator added */
2078 PG_RETURN_TEXT_P(catenate_stringinfo_string(state->str, " }"));
2082 * Helper function for aggregates: return given StringInfo's contents plus
2083 * specified trailing string, as a text datum. We need this because aggregate
2084 * final functions are not allowed to modify the aggregate state.
2087 catenate_stringinfo_string(StringInfo buffer, const char *addon)
2089 /* custom version of cstring_to_text_with_len */
2090 int buflen = buffer->len;
2091 int addlen = strlen(addon);
2092 text *result = (text *) palloc(buflen + addlen + VARHDRSZ);
2094 SET_VARSIZE(result, buflen + addlen + VARHDRSZ);
2095 memcpy(VARDATA(result), buffer->data, buflen);
2096 memcpy(VARDATA(result) + buflen, addon, addlen);
2102 * SQL function json_build_object(variadic "any")
2105 json_build_object(PG_FUNCTION_ARGS)
2107 int nargs = PG_NARGS();
2110 const char *sep = "";
2116 (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
2117 errmsg("argument list must have even number of elements"),
2118 errhint("The arguments of json_build_object() must consist of alternating keys and values.")));
2120 result = makeStringInfo();
2122 appendStringInfoChar(result, '{');
2124 for (i = 0; i < nargs; i += 2)
2127 * Note: since json_build_object() is declared as taking type "any",
2128 * the parser will not do any type conversion on unknown-type literals
2129 * (that is, undecorated strings or NULLs). Such values will arrive
2130 * here as type UNKNOWN, which fortunately does not matter to us,
2131 * since unknownout() works fine.
2133 appendStringInfoString(result, sep);
2137 val_type = get_fn_expr_argtype(fcinfo->flinfo, i);
2139 if (val_type == InvalidOid)
2141 (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
2142 errmsg("could not determine data type for argument %d",
2145 if (PG_ARGISNULL(i))
2147 (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
2148 errmsg("argument %d cannot be null", i + 1),
2149 errhint("Object keys should be text.")));
2151 arg = PG_GETARG_DATUM(i);
2153 add_json(arg, false, result, val_type, true);
2155 appendStringInfoString(result, " : ");
2158 val_type = get_fn_expr_argtype(fcinfo->flinfo, i + 1);
2160 if (val_type == InvalidOid)
2162 (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
2163 errmsg("could not determine data type for argument %d",
2166 if (PG_ARGISNULL(i + 1))
2169 arg = PG_GETARG_DATUM(i + 1);
2171 add_json(arg, PG_ARGISNULL(i + 1), result, val_type, false);
2174 appendStringInfoChar(result, '}');
2176 PG_RETURN_TEXT_P(cstring_to_text_with_len(result->data, result->len));
2180 * degenerate case of json_build_object where it gets 0 arguments.
2183 json_build_object_noargs(PG_FUNCTION_ARGS)
2185 PG_RETURN_TEXT_P(cstring_to_text_with_len("{}", 2));
2189 * SQL function json_build_array(variadic "any")
2192 json_build_array(PG_FUNCTION_ARGS)
2194 int nargs = PG_NARGS();
2197 const char *sep = "";
2201 result = makeStringInfo();
2203 appendStringInfoChar(result, '[');
2205 for (i = 0; i < nargs; i++)
2208 * Note: since json_build_array() is declared as taking type "any",
2209 * the parser will not do any type conversion on unknown-type literals
2210 * (that is, undecorated strings or NULLs). Such values will arrive
2211 * here as type UNKNOWN, which fortunately does not matter to us,
2212 * since unknownout() works fine.
2214 appendStringInfoString(result, sep);
2217 val_type = get_fn_expr_argtype(fcinfo->flinfo, i);
2219 if (val_type == InvalidOid)
2221 (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
2222 errmsg("could not determine data type for argument %d",
2225 if (PG_ARGISNULL(i))
2228 arg = PG_GETARG_DATUM(i);
2230 add_json(arg, PG_ARGISNULL(i), result, val_type, false);
2233 appendStringInfoChar(result, ']');
2235 PG_RETURN_TEXT_P(cstring_to_text_with_len(result->data, result->len));
2239 * degenerate case of json_build_array where it gets 0 arguments.
2242 json_build_array_noargs(PG_FUNCTION_ARGS)
2244 PG_RETURN_TEXT_P(cstring_to_text_with_len("[]", 2));
2248 * SQL function json_object(text[])
2250 * take a one or two dimensional array of text as key/value pairs
2251 * for a json object.
2254 json_object(PG_FUNCTION_ARGS)
2256 ArrayType *in_array = PG_GETARG_ARRAYTYPE_P(0);
2257 int ndims = ARR_NDIM(in_array);
2258 StringInfoData result;
2270 PG_RETURN_DATUM(CStringGetTextDatum("{}"));
2274 if ((ARR_DIMS(in_array)[0]) % 2)
2276 (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
2277 errmsg("array must have even number of elements")));
2281 if ((ARR_DIMS(in_array)[1]) != 2)
2283 (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
2284 errmsg("array must have two columns")));
2289 (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
2290 errmsg("wrong number of array subscripts")));
2293 deconstruct_array(in_array,
2294 TEXTOID, -1, false, 'i',
2295 &in_datums, &in_nulls, &in_count);
2297 count = in_count / 2;
2299 initStringInfo(&result);
2301 appendStringInfoChar(&result, '{');
2303 for (i = 0; i < count; ++i)
2305 if (in_nulls[i * 2])
2307 (errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED),
2308 errmsg("null value not allowed for object key")));
2310 v = TextDatumGetCString(in_datums[i * 2]);
2312 appendStringInfoString(&result, ", ");
2313 escape_json(&result, v);
2314 appendStringInfoString(&result, " : ");
2316 if (in_nulls[i * 2 + 1])
2317 appendStringInfoString(&result, "null");
2320 v = TextDatumGetCString(in_datums[i * 2 + 1]);
2321 escape_json(&result, v);
2326 appendStringInfoChar(&result, '}');
2331 rval = cstring_to_text_with_len(result.data, result.len);
2334 PG_RETURN_TEXT_P(rval);
2339 * SQL function json_object(text[], text[])
2341 * take separate key and value arrays of text to construct a json object
2345 json_object_two_arg(PG_FUNCTION_ARGS)
2347 ArrayType *key_array = PG_GETARG_ARRAYTYPE_P(0);
2348 ArrayType *val_array = PG_GETARG_ARRAYTYPE_P(1);
2349 int nkdims = ARR_NDIM(key_array);
2350 int nvdims = ARR_NDIM(val_array);
2351 StringInfoData result;
2362 if (nkdims > 1 || nkdims != nvdims)
2364 (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
2365 errmsg("wrong number of array subscripts")));
2368 PG_RETURN_DATUM(CStringGetTextDatum("{}"));
2370 deconstruct_array(key_array,
2371 TEXTOID, -1, false, 'i',
2372 &key_datums, &key_nulls, &key_count);
2374 deconstruct_array(val_array,
2375 TEXTOID, -1, false, 'i',
2376 &val_datums, &val_nulls, &val_count);
2378 if (key_count != val_count)
2380 (errcode(ERRCODE_ARRAY_SUBSCRIPT_ERROR),
2381 errmsg("mismatched array dimensions")));
2383 initStringInfo(&result);
2385 appendStringInfoChar(&result, '{');
2387 for (i = 0; i < key_count; ++i)
2391 (errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED),
2392 errmsg("null value not allowed for object key")));
2394 v = TextDatumGetCString(key_datums[i]);
2396 appendStringInfoString(&result, ", ");
2397 escape_json(&result, v);
2398 appendStringInfoString(&result, " : ");
2401 appendStringInfoString(&result, "null");
2404 v = TextDatumGetCString(val_datums[i]);
2405 escape_json(&result, v);
2410 appendStringInfoChar(&result, '}');
2417 rval = cstring_to_text_with_len(result.data, result.len);
2420 PG_RETURN_TEXT_P(rval);
2425 * Produce a JSON string literal, properly escaping characters in the text.
2428 escape_json(StringInfo buf, const char *str)
2432 appendStringInfoCharMacro(buf, '"');
2433 for (p = str; *p; p++)
2438 appendStringInfoString(buf, "\\b");
2441 appendStringInfoString(buf, "\\f");
2444 appendStringInfoString(buf, "\\n");
2447 appendStringInfoString(buf, "\\r");
2450 appendStringInfoString(buf, "\\t");
2453 appendStringInfoString(buf, "\\\"");
2456 appendStringInfoString(buf, "\\\\");
2459 if ((unsigned char) *p < ' ')
2460 appendStringInfo(buf, "\\u%04x", (int) *p);
2462 appendStringInfoCharMacro(buf, *p);
2466 appendStringInfoCharMacro(buf, '"');
2470 * SQL function json_typeof(json) -> text
2472 * Returns the type of the outermost JSON value as TEXT. Possible types are
2473 * "object", "array", "string", "number", "boolean", and "null".
2475 * Performs a single call to json_lex() to get the first token of the supplied
2476 * value. This initial token uniquely determines the value's type. As our
2477 * input must already have been validated by json_in() or json_recv(), the
2478 * initial token should never be JSON_TOKEN_OBJECT_END, JSON_TOKEN_ARRAY_END,
2479 * JSON_TOKEN_COLON, JSON_TOKEN_COMMA, or JSON_TOKEN_END.
2482 json_typeof(PG_FUNCTION_ARGS)
2486 JsonLexContext *lex;
2490 json = PG_GETARG_TEXT_P(0);
2491 lex = makeJsonLexContext(json, false);
2493 /* Lex exactly one token from the input and check its type. */
2495 tok = lex_peek(lex);
2498 case JSON_TOKEN_OBJECT_START:
2501 case JSON_TOKEN_ARRAY_START:
2504 case JSON_TOKEN_STRING:
2507 case JSON_TOKEN_NUMBER:
2510 case JSON_TOKEN_TRUE:
2511 case JSON_TOKEN_FALSE:
2514 case JSON_TOKEN_NULL:
2518 elog(ERROR, "unexpected json token: %d", tok);
2521 PG_RETURN_TEXT_P(cstring_to_text(type));