]> granicus.if.org Git - postgresql/blobdiff - src/backend/utils/adt/json.c
Fix initialization of fake LSN for unlogged relations
[postgresql] / src / backend / utils / adt / json.c
index d1dc87e679f696604f650568b835de5bb29f4369..d4ba3bd87db344305738250db64733f13593ce22 100644 (file)
@@ -3,7 +3,7 @@
  * json.c
  *             JSON data type support.
  *
- * Portions Copyright (c) 1996-2014, PostgreSQL Global Development Group
+ * Portions Copyright (c) 1996-2019, PostgreSQL Global Development Group
  * Portions Copyright (c) 1994, Regents of the University of California
  *
  * IDENTIFICATION
@@ -15,9 +15,9 @@
 
 #include "access/htup_details.h"
 #include "access/transam.h"
-#include "catalog/pg_cast.h"
 #include "catalog/pg_type.h"
 #include "executor/spi.h"
+#include "funcapi.h"
 #include "lib/stringinfo.h"
 #include "libpq/pqformat.h"
 #include "mb/pg_wchar.h"
@@ -25,6 +25,7 @@
 #include "parser/parse_coerce.h"
 #include "utils/array.h"
 #include "utils/builtins.h"
+#include "utils/date.h"
 #include "utils/datetime.h"
 #include "utils/lsyscache.h"
 #include "utils/json.h"
@@ -55,8 +56,9 @@ typedef enum                                  /* type categories for datum_to_json */
        JSONTYPE_NULL,                          /* null, so we didn't bother to identify */
        JSONTYPE_BOOL,                          /* boolean (built-in types only) */
        JSONTYPE_NUMERIC,                       /* numeric (ditto) */
-       JSONTYPE_TIMESTAMP,         /* we use special formatting for timestamp */
-       JSONTYPE_TIMESTAMPTZ,       /* ... and timestamptz */
+       JSONTYPE_DATE,                          /* we use special formatting for datetimes */
+       JSONTYPE_TIMESTAMP,
+       JSONTYPE_TIMESTAMPTZ,
        JSONTYPE_JSON,                          /* JSON itself (and JSONB) */
        JSONTYPE_ARRAY,                         /* array */
        JSONTYPE_COMPOSITE,                     /* composite */
@@ -64,34 +66,45 @@ typedef enum                                        /* type categories for datum_to_json */
        JSONTYPE_OTHER                          /* all else */
 } JsonTypeCategory;
 
+typedef struct JsonAggState
+{
+       StringInfo      str;
+       JsonTypeCategory key_category;
+       Oid                     key_output_func;
+       JsonTypeCategory val_category;
+       Oid                     val_output_func;
+} JsonAggState;
+
 static inline void json_lex(JsonLexContext *lex);
 static inline void json_lex_string(JsonLexContext *lex);
-static inline void json_lex_number(JsonLexContext *lex, char *s, bool *num_err);
+static inline void json_lex_number(JsonLexContext *lex, char *s,
+                                                                  bool *num_err, int *total_len);
 static inline void parse_scalar(JsonLexContext *lex, JsonSemAction *sem);
 static void parse_object_field(JsonLexContext *lex, JsonSemAction *sem);
 static void parse_object(JsonLexContext *lex, JsonSemAction *sem);
 static void parse_array_element(JsonLexContext *lex, JsonSemAction *sem);
 static void parse_array(JsonLexContext *lex, JsonSemAction *sem);
-static void report_parse_error(JsonParseContext ctx, JsonLexContext *lex);
-static void report_invalid_token(JsonLexContext *lex);
+static void report_parse_error(JsonParseContext ctx, JsonLexContext *lex) pg_attribute_noreturn();
+static void report_invalid_token(JsonLexContext *lex) pg_attribute_noreturn();
 static int     report_json_context(JsonLexContext *lex);
 static char *extract_mb_char(char *s);
 static void composite_to_json(Datum composite, StringInfo result,
-                                 bool use_line_feeds);
+                                                         bool use_line_feeds);
 static void array_dim_to_json(StringInfo result, int dim, int ndims, int *dims,
-                                 Datum *vals, bool *nulls, int *valcount,
-                                 JsonTypeCategory tcategory, Oid outfuncoid,
-                                 bool use_line_feeds);
+                                                         Datum *vals, bool *nulls, int *valcount,
+                                                         JsonTypeCategory tcategory, Oid outfuncoid,
+                                                         bool use_line_feeds);
 static void array_to_json_internal(Datum array, StringInfo result,
-                                          bool use_line_feeds);
+                                                                  bool use_line_feeds);
 static void json_categorize_type(Oid typoid,
-                                        JsonTypeCategory *tcategory,
-                                        Oid *outfuncoid);
+                                                                JsonTypeCategory *tcategory,
+                                                                Oid *outfuncoid);
 static void datum_to_json(Datum val, bool is_null, StringInfo result,
-                         JsonTypeCategory tcategory, Oid outfuncoid,
-                         bool key_scalar);
+                                                 JsonTypeCategory tcategory, Oid outfuncoid,
+                                                 bool key_scalar);
 static void add_json(Datum val, bool is_null, StringInfo result,
-                Oid val_type, bool key_scalar);
+                                        Oid val_type, bool key_scalar);
+static text *catenate_stringinfo_string(StringInfo buffer, const char *addon);
 
 /* the null action object used for pure validation */
 static JsonSemAction nullSemAction =
@@ -160,7 +173,7 @@ static inline void
 lex_expect(JsonParseContext ctx, JsonLexContext *lex, JsonTokenType token)
 {
        if (!lex_accept(lex, token, NULL))
-               report_parse_error(ctx, lex);;
+               report_parse_error(ctx, lex);
 }
 
 /* chars to consider as part of an alphanumeric token */
@@ -171,6 +184,43 @@ lex_expect(JsonParseContext ctx, JsonLexContext *lex, JsonTokenType token)
         (c) == '_' || \
         IS_HIGHBIT_SET(c))
 
+/*
+ * Utility function to check if a string is a valid JSON number.
+ *
+ * str is of length len, and need not be null-terminated.
+ */
+bool
+IsValidJsonNumber(const char *str, int len)
+{
+       bool            numeric_error;
+       int                     total_len;
+       JsonLexContext dummy_lex;
+
+       if (len <= 0)
+               return false;
+
+       /*
+        * json_lex_number expects a leading  '-' to have been eaten already.
+        *
+        * having to cast away the constness of str is ugly, but there's not much
+        * easy alternative.
+        */
+       if (*str == '-')
+       {
+               dummy_lex.input = unconstify(char *, str) +1;
+               dummy_lex.input_length = len - 1;
+       }
+       else
+       {
+               dummy_lex.input = unconstify(char *, str);
+               dummy_lex.input_length = len;
+       }
+
+       json_lex_number(&dummy_lex, dummy_lex.input, &numeric_error, &total_len);
+
+       return (!numeric_error) && (total_len == dummy_lex.input_length);
+}
+
 /*
  * Input.
  */
@@ -250,8 +300,8 @@ json_recv(PG_FUNCTION_ARGS)
 JsonLexContext *
 makeJsonLexContext(text *json, bool need_escapes)
 {
-       return makeJsonLexContextCstringLen(VARDATA(json),
-                                                                               VARSIZE(json) - VARHDRSZ,
+       return makeJsonLexContextCstringLen(VARDATA_ANY(json),
+                                                                               VARSIZE_ANY_EXHDR(json),
                                                                                need_escapes);
 }
 
@@ -274,7 +324,7 @@ makeJsonLexContextCstringLen(char *json, int len, bool need_escapes)
  * Publicly visible entry point for the JSON parser.
  *
  * lex is a lexing context, set up for the json to be processed by calling
- * makeJsonLexContext(). sem is a strucure of function pointers to semantic
+ * makeJsonLexContext(). sem is a structure of function pointers to semantic
  * action routines to be called at appropriate spots during parsing, and a
  * pointer to a state object to be passed to those routines.
  */
@@ -298,13 +348,52 @@ pg_parse_json(JsonLexContext *lex, JsonSemAction *sem)
                        parse_array(lex, sem);
                        break;
                default:
-                       parse_scalar(lex, sem);         /* json can be a bare scalar */
+                       parse_scalar(lex, sem); /* json can be a bare scalar */
        }
 
        lex_expect(JSON_PARSE_END, lex, JSON_TOKEN_END);
 
 }
 
+/*
+ * json_count_array_elements
+ *
+ * Returns number of array elements in lex context at start of array token
+ * until end of array token at same nesting level.
+ *
+ * Designed to be called from array_start routines.
+ */
+int
+json_count_array_elements(JsonLexContext *lex)
+{
+       JsonLexContext copylex;
+       int                     count;
+
+       /*
+        * It's safe to do this with a shallow copy because the lexical routines
+        * don't scribble on the input. They do scribble on the other pointers
+        * etc, so doing this with a copy makes that safe.
+        */
+       memcpy(&copylex, lex, sizeof(JsonLexContext));
+       copylex.strval = NULL;          /* not interested in values here */
+       copylex.lex_level++;
+
+       count = 0;
+       lex_expect(JSON_PARSE_ARRAY_START, &copylex, JSON_TOKEN_ARRAY_START);
+       if (lex_peek(&copylex) != JSON_TOKEN_ARRAY_END)
+       {
+               do
+               {
+                       count++;
+                       parse_array_element(&copylex, &nullSemAction);
+               }
+               while (lex_accept(&copylex, JSON_TOKEN_COMMA, NULL));
+       }
+       lex_expect(JSON_PARSE_ARRAY_NEXT, &copylex, JSON_TOKEN_ARRAY_END);
+
+       return count;
+}
+
 /*
  *     Recursive Descent parse routines. There is one for each structural
  *     element in a json document:
@@ -407,6 +496,8 @@ parse_object(JsonLexContext *lex, JsonSemAction *sem)
        json_struct_action oend = sem->object_end;
        JsonTokenType tok;
 
+       check_stack_depth();
+
        if (ostart != NULL)
                (*ostart) (sem->semstate);
 
@@ -418,7 +509,7 @@ parse_object(JsonLexContext *lex, JsonSemAction *sem)
         */
        lex->lex_level++;
 
-       /* we know this will succeeed, just clearing the token */
+       /* we know this will succeed, just clearing the token */
        lex_expect(JSON_PARSE_OBJECT_START, lex, JSON_TOKEN_OBJECT_START);
 
        tok = lex_peek(lex);
@@ -485,6 +576,8 @@ parse_array(JsonLexContext *lex, JsonSemAction *sem)
        json_struct_action astart = sem->array_start;
        json_struct_action aend = sem->array_end;
 
+       check_stack_depth();
+
        if (astart != NULL)
                (*astart) (sem->semstate);
 
@@ -585,7 +678,7 @@ json_lex(JsonLexContext *lex)
                                break;
                        case '-':
                                /* Negative number. */
-                               json_lex_number(lex, s + 1, NULL);
+                               json_lex_number(lex, s + 1, NULL, NULL);
                                lex->token_type = JSON_TOKEN_NUMBER;
                                break;
                        case '0':
@@ -599,7 +692,7 @@ json_lex(JsonLexContext *lex)
                        case '8':
                        case '9':
                                /* Positive number. */
-                               json_lex_number(lex, s, NULL);
+                               json_lex_number(lex, s, NULL, NULL);
                                lex->token_type = JSON_TOKEN_NUMBER;
                                break;
                        default:
@@ -690,7 +783,7 @@ json_lex_string(JsonLexContext *lex)
                        lex->token_terminator = s;
                        ereport(ERROR,
                                        (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
-                                        errmsg("invalid input syntax for type json"),
+                                        errmsg("invalid input syntax for type %s", "json"),
                                         errdetail("Character with value 0x%02x must be escaped.",
                                                           (unsigned char) *s),
                                         report_json_context(lex)));
@@ -730,7 +823,8 @@ json_lex_string(JsonLexContext *lex)
                                                lex->token_terminator = s + pg_mblen(s);
                                                ereport(ERROR,
                                                                (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
-                                                                errmsg("invalid input syntax for type json"),
+                                                                errmsg("invalid input syntax for type %s",
+                                                                               "json"),
                                                                 errdetail("\"\\u\" must be followed by four hexadecimal digits."),
                                                                 report_json_context(lex)));
                                        }
@@ -744,10 +838,11 @@ json_lex_string(JsonLexContext *lex)
                                        {
                                                if (hi_surrogate != -1)
                                                        ereport(ERROR,
-                                                          (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
-                                                               errmsg("invalid input syntax for type json"),
-                                                               errdetail("Unicode high surrogate must not follow a high surrogate."),
-                                                               report_json_context(lex)));
+                                                                       (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
+                                                                        errmsg("invalid input syntax for type %s",
+                                                                                       "json"),
+                                                                        errdetail("Unicode high surrogate must not follow a high surrogate."),
+                                                                        report_json_context(lex)));
                                                hi_surrogate = (ch & 0x3ff) << 10;
                                                continue;
                                        }
@@ -755,10 +850,10 @@ json_lex_string(JsonLexContext *lex)
                                        {
                                                if (hi_surrogate == -1)
                                                        ereport(ERROR,
-                                                          (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
-                                                               errmsg("invalid input syntax for type json"),
-                                                               errdetail("Unicode low surrogate must follow a high surrogate."),
-                                                               report_json_context(lex)));
+                                                                       (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
+                                                                        errmsg("invalid input syntax for type %s", "json"),
+                                                                        errdetail("Unicode low surrogate must follow a high surrogate."),
+                                                                        report_json_context(lex)));
                                                ch = 0x10000 + hi_surrogate + (ch & 0x3ff);
                                                hi_surrogate = -1;
                                        }
@@ -766,7 +861,7 @@ json_lex_string(JsonLexContext *lex)
                                        if (hi_surrogate != -1)
                                                ereport(ERROR,
                                                                (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
-                                                                errmsg("invalid input syntax for type json"),
+                                                                errmsg("invalid input syntax for type %s", "json"),
                                                                 errdetail("Unicode low surrogate must follow a high surrogate."),
                                                                 report_json_context(lex)));
 
@@ -774,14 +869,17 @@ json_lex_string(JsonLexContext *lex)
                                         * For UTF8, replace the escape sequence by the actual
                                         * utf8 character in lex->strval. Do this also for other
                                         * encodings if the escape designates an ASCII character,
-                                        * otherwise raise an error. We don't ever unescape a
-                                        * \u0000, since that would result in an impermissible nul
-                                        * byte.
+                                        * otherwise raise an error.
                                         */
 
                                        if (ch == 0)
                                        {
-                                               appendStringInfoString(lex->strval, "\\u0000");
+                                               /* We can't allow this, since our TEXT type doesn't */
+                                               ereport(ERROR,
+                                                               (errcode(ERRCODE_UNTRANSLATABLE_CHARACTER),
+                                                                errmsg("unsupported Unicode escape sequence"),
+                                                                errdetail("\\u0000 cannot be converted to text."),
+                                                                report_json_context(lex)));
                                        }
                                        else if (GetDatabaseEncoding() == PG_UTF8)
                                        {
@@ -801,8 +899,8 @@ json_lex_string(JsonLexContext *lex)
                                        else
                                        {
                                                ereport(ERROR,
-                                                               (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
-                                                                errmsg("invalid input syntax for type json"),
+                                                               (errcode(ERRCODE_UNTRANSLATABLE_CHARACTER),
+                                                                errmsg("unsupported Unicode escape sequence"),
                                                                 errdetail("Unicode escape values cannot be used for code point values above 007F when the server encoding is not UTF8."),
                                                                 report_json_context(lex)));
                                        }
@@ -814,7 +912,8 @@ json_lex_string(JsonLexContext *lex)
                                if (hi_surrogate != -1)
                                        ereport(ERROR,
                                                        (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
-                                                        errmsg("invalid input syntax for type json"),
+                                                        errmsg("invalid input syntax for type %s",
+                                                                       "json"),
                                                         errdetail("Unicode low surrogate must follow a high surrogate."),
                                                         report_json_context(lex)));
 
@@ -845,9 +944,10 @@ json_lex_string(JsonLexContext *lex)
                                                lex->token_terminator = s + pg_mblen(s);
                                                ereport(ERROR,
                                                                (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
-                                                                errmsg("invalid input syntax for type json"),
-                                                       errdetail("Escape sequence \"\\%s\" is invalid.",
-                                                                         extract_mb_char(s)),
+                                                                errmsg("invalid input syntax for type %s",
+                                                                               "json"),
+                                                                errdetail("Escape sequence \"\\%s\" is invalid.",
+                                                                                  extract_mb_char(s)),
                                                                 report_json_context(lex)));
                                }
                        }
@@ -863,7 +963,7 @@ json_lex_string(JsonLexContext *lex)
                                lex->token_terminator = s + pg_mblen(s);
                                ereport(ERROR,
                                                (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
-                                                errmsg("invalid input syntax for type json"),
+                                                errmsg("invalid input syntax for type %s", "json"),
                                                 errdetail("Escape sequence \"\\%s\" is invalid.",
                                                                   extract_mb_char(s)),
                                                 report_json_context(lex)));
@@ -875,7 +975,7 @@ json_lex_string(JsonLexContext *lex)
                        if (hi_surrogate != -1)
                                ereport(ERROR,
                                                (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
-                                                errmsg("invalid input syntax for type json"),
+                                                errmsg("invalid input syntax for type %s", "json"),
                                                 errdetail("Unicode low surrogate must follow a high surrogate."),
                                                 report_json_context(lex)));
 
@@ -887,8 +987,8 @@ json_lex_string(JsonLexContext *lex)
        if (hi_surrogate != -1)
                ereport(ERROR,
                                (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
-                                errmsg("invalid input syntax for type json"),
-                       errdetail("Unicode low surrogate must follow a high surrogate."),
+                                errmsg("invalid input syntax for type %s", "json"),
+                                errdetail("Unicode low surrogate must follow a high surrogate."),
                                 report_json_context(lex)));
 
        /* Hooray, we found the end of the string! */
@@ -896,7 +996,7 @@ json_lex_string(JsonLexContext *lex)
        lex->token_terminator = s + 1;
 }
 
-/*-------------------------------------------------------------------------
+/*
  * The next token in the input stream is known to be a number; lex it.
  *
  * In JSON, a number consists of four parts:
@@ -917,29 +1017,30 @@ json_lex_string(JsonLexContext *lex)
  *        followed by at least one digit.)
  *
  * The 's' argument to this function points to the ostensible beginning
- * of part 2 - i.e. the character after any optional minus sign, and the
+ * of part 2 - i.e. the character after any optional minus sign, or the
  * first character of the string if there is none.
  *
- *-------------------------------------------------------------------------
+ * If num_err is not NULL, we return an error flag to *num_err rather than
+ * raising an error for a badly-formed number.  Also, if total_len is not NULL
+ * the distance from lex->input to the token end+1 is returned to *total_len.
  */
 static inline void
-json_lex_number(JsonLexContext *lex, char *s, bool *num_err)
+json_lex_number(JsonLexContext *lex, char *s,
+                               bool *num_err, int *total_len)
 {
        bool            error = false;
-       char       *p;
-       int                     len;
+       int                     len = s - lex->input;
 
-       len = s - lex->input;
        /* Part (1): leading sign indicator. */
        /* Caller already did this for us; so do nothing. */
 
        /* Part (2): parse main digit string. */
-       if (*s == '0')
+       if (len < lex->input_length && *s == '0')
        {
                s++;
                len++;
        }
-       else if (*s >= '1' && *s <= '9')
+       else if (len < lex->input_length && *s >= '1' && *s <= '9')
        {
                do
                {
@@ -994,18 +1095,23 @@ json_lex_number(JsonLexContext *lex, char *s, bool *num_err)
         * here should be considered part of the token for error-reporting
         * purposes.
         */
-       for (p = s; len < lex->input_length && JSON_ALPHANUMERIC_CHAR(*p); p++, len++)
+       for (; len < lex->input_length && JSON_ALPHANUMERIC_CHAR(*s); s++, len++)
                error = true;
 
+       if (total_len != NULL)
+               *total_len = len;
+
        if (num_err != NULL)
        {
-               /* let the caller handle the error */
+               /* let the caller handle any error */
                *num_err = error;
        }
        else
        {
+               /* return token endpoint */
                lex->prev_token_terminator = lex->token_terminator;
-               lex->token_terminator = p;
+               lex->token_terminator = s;
+               /* handle error if any */
                if (error)
                        report_invalid_token(lex);
        }
@@ -1026,7 +1132,7 @@ report_parse_error(JsonParseContext ctx, JsonLexContext *lex)
        if (lex->token_start == NULL || lex->token_type == JSON_TOKEN_END)
                ereport(ERROR,
                                (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
-                                errmsg("invalid input syntax for type json"),
+                                errmsg("invalid input syntax for type %s", "json"),
                                 errdetail("The input string ended unexpectedly."),
                                 report_json_context(lex)));
 
@@ -1040,7 +1146,7 @@ report_parse_error(JsonParseContext ctx, JsonLexContext *lex)
        if (ctx == JSON_PARSE_END)
                ereport(ERROR,
                                (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
-                                errmsg("invalid input syntax for type json"),
+                                errmsg("invalid input syntax for type %s", "json"),
                                 errdetail("Expected end of input, but found \"%s\".",
                                                   token),
                                 report_json_context(lex)));
@@ -1051,7 +1157,7 @@ report_parse_error(JsonParseContext ctx, JsonLexContext *lex)
                        case JSON_PARSE_VALUE:
                                ereport(ERROR,
                                                (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
-                                                errmsg("invalid input syntax for type json"),
+                                                errmsg("invalid input syntax for type %s", "json"),
                                                 errdetail("Expected JSON value, but found \"%s\".",
                                                                   token),
                                                 report_json_context(lex)));
@@ -1059,7 +1165,7 @@ report_parse_error(JsonParseContext ctx, JsonLexContext *lex)
                        case JSON_PARSE_STRING:
                                ereport(ERROR,
                                                (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
-                                                errmsg("invalid input syntax for type json"),
+                                                errmsg("invalid input syntax for type %s", "json"),
                                                 errdetail("Expected string, but found \"%s\".",
                                                                   token),
                                                 report_json_context(lex)));
@@ -1067,7 +1173,7 @@ report_parse_error(JsonParseContext ctx, JsonLexContext *lex)
                        case JSON_PARSE_ARRAY_START:
                                ereport(ERROR,
                                                (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
-                                                errmsg("invalid input syntax for type json"),
+                                                errmsg("invalid input syntax for type %s", "json"),
                                                 errdetail("Expected array element or \"]\", but found \"%s\".",
                                                                   token),
                                                 report_json_context(lex)));
@@ -1075,23 +1181,23 @@ report_parse_error(JsonParseContext ctx, JsonLexContext *lex)
                        case JSON_PARSE_ARRAY_NEXT:
                                ereport(ERROR,
                                                (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
-                                                errmsg("invalid input syntax for type json"),
-                                         errdetail("Expected \",\" or \"]\", but found \"%s\".",
-                                                               token),
+                                                errmsg("invalid input syntax for type %s", "json"),
+                                                errdetail("Expected \",\" or \"]\", but found \"%s\".",
+                                                                  token),
                                                 report_json_context(lex)));
                                break;
                        case JSON_PARSE_OBJECT_START:
                                ereport(ERROR,
                                                (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
-                                                errmsg("invalid input syntax for type json"),
-                                        errdetail("Expected string or \"}\", but found \"%s\".",
-                                                          token),
+                                                errmsg("invalid input syntax for type %s", "json"),
+                                                errdetail("Expected string or \"}\", but found \"%s\".",
+                                                                  token),
                                                 report_json_context(lex)));
                                break;
                        case JSON_PARSE_OBJECT_LABEL:
                                ereport(ERROR,
                                                (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
-                                                errmsg("invalid input syntax for type json"),
+                                                errmsg("invalid input syntax for type %s", "json"),
                                                 errdetail("Expected \":\", but found \"%s\".",
                                                                   token),
                                                 report_json_context(lex)));
@@ -1099,15 +1205,15 @@ report_parse_error(JsonParseContext ctx, JsonLexContext *lex)
                        case JSON_PARSE_OBJECT_NEXT:
                                ereport(ERROR,
                                                (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
-                                                errmsg("invalid input syntax for type json"),
-                                         errdetail("Expected \",\" or \"}\", but found \"%s\".",
-                                                               token),
+                                                errmsg("invalid input syntax for type %s", "json"),
+                                                errdetail("Expected \",\" or \"}\", but found \"%s\".",
+                                                                  token),
                                                 report_json_context(lex)));
                                break;
                        case JSON_PARSE_OBJECT_COMMA:
                                ereport(ERROR,
                                                (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
-                                                errmsg("invalid input syntax for type json"),
+                                                errmsg("invalid input syntax for type %s", "json"),
                                                 errdetail("Expected string, but found \"%s\".",
                                                                   token),
                                                 report_json_context(lex)));
@@ -1137,7 +1243,7 @@ report_invalid_token(JsonLexContext *lex)
 
        ereport(ERROR,
                        (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
-                        errmsg("invalid input syntax for type json"),
+                        errmsg("invalid input syntax for type %s", "json"),
                         errdetail("Token \"%s\" is invalid.", token),
                         report_json_context(lex)));
 }
@@ -1248,10 +1354,14 @@ json_categorize_type(Oid typoid,
        /* Look through any domain */
        typoid = getBaseType(typoid);
 
-       /* We'll usually need to return the type output function */
-       getTypeOutputInfo(typoid, outfuncoid, &typisvarlena);
+       *outfuncoid = InvalidOid;
+
+       /*
+        * We need to get the output function for everything except date and
+        * timestamp types, array and composite types, booleans, and non-builtin
+        * types where there's a cast to json.
+        */
 
-       /* Check for known types */
        switch (typoid)
        {
                case BOOLOID:
@@ -1264,9 +1374,14 @@ json_categorize_type(Oid typoid,
                case FLOAT4OID:
                case FLOAT8OID:
                case NUMERICOID:
+                       getTypeOutputInfo(typoid, outfuncoid, &typisvarlena);
                        *tcategory = JSONTYPE_NUMERIC;
                        break;
 
+               case DATEOID:
+                       *tcategory = JSONTYPE_DATE;
+                       break;
+
                case TIMESTAMPOID:
                        *tcategory = JSONTYPE_TIMESTAMP;
                        break;
@@ -1277,14 +1392,16 @@ json_categorize_type(Oid typoid,
 
                case JSONOID:
                case JSONBOID:
+                       getTypeOutputInfo(typoid, outfuncoid, &typisvarlena);
                        *tcategory = JSONTYPE_JSON;
                        break;
 
                default:
                        /* Check for arrays and composites */
-                       if (OidIsValid(get_element_type(typoid)))
+                       if (OidIsValid(get_element_type(typoid)) || typoid == ANYARRAYOID
+                               || typoid == RECORDARRAYOID)
                                *tcategory = JSONTYPE_ARRAY;
-                       else if (type_is_rowtype(typoid))
+                       else if (type_is_rowtype(typoid))       /* includes RECORDOID */
                                *tcategory = JSONTYPE_COMPOSITE;
                        else
                        {
@@ -1293,24 +1410,28 @@ json_categorize_type(Oid typoid,
                                /* but let's look for a cast to json, if it's not built-in */
                                if (typoid >= FirstNormalObjectId)
                                {
-                                       HeapTuple       tuple;
+                                       Oid                     castfunc;
+                                       CoercionPathType ctype;
 
-                                       tuple = SearchSysCache2(CASTSOURCETARGET,
-                                                                                       ObjectIdGetDatum(typoid),
-                                                                                       ObjectIdGetDatum(JSONOID));
-                                       if (HeapTupleIsValid(tuple))
+                                       ctype = find_coercion_pathway(JSONOID, typoid,
+                                                                                                 COERCION_EXPLICIT,
+                                                                                                 &castfunc);
+                                       if (ctype == COERCION_PATH_FUNC && OidIsValid(castfunc))
                                        {
-                                               Form_pg_cast castForm = (Form_pg_cast) GETSTRUCT(tuple);
-
-                                               if (castForm->castmethod == COERCION_METHOD_FUNCTION)
-                                               {
-                                                       *tcategory = JSONTYPE_CAST;
-                                                       *outfuncoid = castForm->castfunc;
-                                               }
-
-                                               ReleaseSysCache(tuple);
+                                               *tcategory = JSONTYPE_CAST;
+                                               *outfuncoid = castfunc;
+                                       }
+                                       else
+                                       {
+                                               /* non builtin type with no cast */
+                                               getTypeOutputInfo(typoid, outfuncoid, &typisvarlena);
                                        }
                                }
+                               else
+                               {
+                                       /* any other builtin type */
+                                       getTypeOutputInfo(typoid, outfuncoid, &typisvarlena);
+                               }
                        }
                        break;
        }
@@ -1332,8 +1453,11 @@ datum_to_json(Datum val, bool is_null, StringInfo result,
 {
        char       *outputstr;
        text       *jsontext;
-       bool            numeric_error;
-       JsonLexContext dummy_lex;
+
+       check_stack_depth();
+
+       /* callers are expected to ensure that null keys are not passed in */
+       Assert(!(key_scalar && is_null));
 
        if (is_null)
        {
@@ -1348,7 +1472,7 @@ datum_to_json(Datum val, bool is_null, StringInfo result,
                 tcategory == JSONTYPE_CAST))
                ereport(ERROR,
                                (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
-                 errmsg("key value must be scalar, not array, composite, or json")));
+                                errmsg("key value must be scalar, not array, composite, or json")));
 
        switch (tcategory)
        {
@@ -1367,103 +1491,182 @@ datum_to_json(Datum val, bool is_null, StringInfo result,
                        break;
                case JSONTYPE_NUMERIC:
                        outputstr = OidOutputFunctionCall(outfuncoid, val);
-                       if (key_scalar)
-                       {
-                               /* always quote keys */
+
+                       /*
+                        * Don't call escape_json for a non-key if it's a valid JSON
+                        * number.
+                        */
+                       if (!key_scalar && IsValidJsonNumber(outputstr, strlen(outputstr)))
+                               appendStringInfoString(result, outputstr);
+                       else
                                escape_json(result, outputstr);
+                       pfree(outputstr);
+                       break;
+               case JSONTYPE_DATE:
+                       {
+                               char            buf[MAXDATELEN + 1];
+
+                               JsonEncodeDateTime(buf, val, DATEOID, NULL);
+                               appendStringInfo(result, "\"%s\"", buf);
                        }
-                       else
+                       break;
+               case JSONTYPE_TIMESTAMP:
                        {
-                               /*
-                                * Don't call escape_json for a non-key if it's a valid JSON
-                                * number.
-                                */
-                               dummy_lex.input = *outputstr == '-' ? outputstr + 1 : outputstr;
-                               dummy_lex.input_length = strlen(dummy_lex.input);
-                               json_lex_number(&dummy_lex, dummy_lex.input, &numeric_error);
-                               if (!numeric_error)
-                                       appendStringInfoString(result, outputstr);
-                               else
-                                       escape_json(result, outputstr);
+                               char            buf[MAXDATELEN + 1];
+
+                               JsonEncodeDateTime(buf, val, TIMESTAMPOID, NULL);
+                               appendStringInfo(result, "\"%s\"", buf);
                        }
+                       break;
+               case JSONTYPE_TIMESTAMPTZ:
+                       {
+                               char            buf[MAXDATELEN + 1];
+
+                               JsonEncodeDateTime(buf, val, TIMESTAMPTZOID, NULL);
+                               appendStringInfo(result, "\"%s\"", buf);
+                       }
+                       break;
+               case JSONTYPE_JSON:
+                       /* JSON and JSONB output will already be escaped */
+                       outputstr = OidOutputFunctionCall(outfuncoid, val);
+                       appendStringInfoString(result, outputstr);
                        pfree(outputstr);
                        break;
-               case JSONTYPE_TIMESTAMP:
+               case JSONTYPE_CAST:
+                       /* outfuncoid refers to a cast function, not an output function */
+                       jsontext = DatumGetTextPP(OidFunctionCall1(outfuncoid, val));
+                       outputstr = text_to_cstring(jsontext);
+                       appendStringInfoString(result, outputstr);
+                       pfree(outputstr);
+                       pfree(jsontext);
+                       break;
+               default:
+                       outputstr = OidOutputFunctionCall(outfuncoid, val);
+                       escape_json(result, outputstr);
+                       pfree(outputstr);
+                       break;
+       }
+}
+
+/*
+ * Encode 'value' of datetime type 'typid' into JSON string in ISO format using
+ * optionally preallocated buffer 'buf'.  Optional 'tzp' determines time-zone
+ * offset (in seconds) in which we want to show timestamptz.
+ */
+char *
+JsonEncodeDateTime(char *buf, Datum value, Oid typid, const int *tzp)
+{
+       if (!buf)
+               buf = palloc(MAXDATELEN + 1);
+
+       switch (typid)
+       {
+               case DATEOID:
                        {
-                               Timestamp       timestamp;
+                               DateADT         date;
                                struct pg_tm tm;
+
+                               date = DatumGetDateADT(value);
+
+                               /* Same as date_out(), but forcing DateStyle */
+                               if (DATE_NOT_FINITE(date))
+                                       EncodeSpecialDate(date, buf);
+                               else
+                               {
+                                       j2date(date + POSTGRES_EPOCH_JDATE,
+                                                  &(tm.tm_year), &(tm.tm_mon), &(tm.tm_mday));
+                                       EncodeDateOnly(&tm, USE_XSD_DATES, buf);
+                               }
+                       }
+                       break;
+               case TIMEOID:
+                       {
+                               TimeADT         time = DatumGetTimeADT(value);
+                               struct pg_tm tt,
+                                                  *tm = &tt;
                                fsec_t          fsec;
-                               char            buf[MAXDATELEN + 1];
 
-                               timestamp = DatumGetTimestamp(val);
+                               /* Same as time_out(), but forcing DateStyle */
+                               time2tm(time, tm, &fsec);
+                               EncodeTimeOnly(tm, fsec, false, 0, USE_XSD_DATES, buf);
+                       }
+                       break;
+               case TIMETZOID:
+                       {
+                               TimeTzADT  *time = DatumGetTimeTzADTP(value);
+                               struct pg_tm tt,
+                                                  *tm = &tt;
+                               fsec_t          fsec;
+                               int                     tz;
+
+                               /* Same as timetz_out(), but forcing DateStyle */
+                               timetz2tm(time, tm, &fsec, &tz);
+                               EncodeTimeOnly(tm, fsec, true, tz, USE_XSD_DATES, buf);
+                       }
+                       break;
+               case TIMESTAMPOID:
+                       {
+                               Timestamp       timestamp;
+                               struct pg_tm tm;
+                               fsec_t          fsec;
 
-                               /* XSD doesn't support infinite values */
+                               timestamp = DatumGetTimestamp(value);
+                               /* Same as timestamp_out(), but forcing DateStyle */
                                if (TIMESTAMP_NOT_FINITE(timestamp))
-                                       ereport(ERROR,
-                                                       (errcode(ERRCODE_DATETIME_VALUE_OUT_OF_RANGE),
-                                                        errmsg("timestamp out of range"),
-                                                        errdetail("JSON does not support infinite timestamp values.")));
+                                       EncodeSpecialTimestamp(timestamp, buf);
                                else if (timestamp2tm(timestamp, NULL, &tm, &fsec, NULL, NULL) == 0)
                                        EncodeDateTime(&tm, fsec, false, 0, NULL, USE_XSD_DATES, buf);
                                else
                                        ereport(ERROR,
                                                        (errcode(ERRCODE_DATETIME_VALUE_OUT_OF_RANGE),
                                                         errmsg("timestamp out of range")));
-
-                               appendStringInfo(result,"\"%s\"",buf);
                        }
                        break;
-               case JSONTYPE_TIMESTAMPTZ:
+               case TIMESTAMPTZOID:
                        {
                                TimestampTz timestamp;
                                struct pg_tm tm;
                                int                     tz;
                                fsec_t          fsec;
                                const char *tzn = NULL;
-                               char            buf[MAXDATELEN + 1];
 
-                               timestamp = DatumGetTimestamp(val);
+                               timestamp = DatumGetTimestampTz(value);
+
+                               /*
+                                * If a time zone is specified, we apply the time-zone shift,
+                                * convert timestamptz to pg_tm as if it were without a time
+                                * zone, and then use the specified time zone for converting
+                                * the timestamp into a string.
+                                */
+                               if (tzp)
+                               {
+                                       tz = *tzp;
+                                       timestamp -= (TimestampTz) tz * USECS_PER_SEC;
+                               }
 
-                               /* XSD doesn't support infinite values */
+                               /* Same as timestamptz_out(), but forcing DateStyle */
                                if (TIMESTAMP_NOT_FINITE(timestamp))
-                                       ereport(ERROR,
-                                                       (errcode(ERRCODE_DATETIME_VALUE_OUT_OF_RANGE),
-                                                        errmsg("timestamp out of range"),
-                                                        errdetail("JSON does not support infinite timestamp values.")));
-                               else if (timestamp2tm(timestamp, &tz, &tm, &fsec, &tzn, NULL) == 0)
+                                       EncodeSpecialTimestamp(timestamp, buf);
+                               else if (timestamp2tm(timestamp, tzp ? NULL : &tz, &tm, &fsec,
+                                                                         tzp ? NULL : &tzn, NULL) == 0)
+                               {
+                                       if (tzp)
+                                               tm.tm_isdst = 1;        /* set time-zone presence flag */
+
                                        EncodeDateTime(&tm, fsec, true, tz, tzn, USE_XSD_DATES, buf);
+                               }
                                else
                                        ereport(ERROR,
                                                        (errcode(ERRCODE_DATETIME_VALUE_OUT_OF_RANGE),
                                                         errmsg("timestamp out of range")));
-
-                               appendStringInfo(result,"\"%s\"",buf);
                        }
                        break;
-               case JSONTYPE_JSON:
-                       /* JSON and JSONB output will already be escaped */
-                       outputstr = OidOutputFunctionCall(outfuncoid, val);
-                       appendStringInfoString(result, outputstr);
-                       pfree(outputstr);
-                       break;
-               case JSONTYPE_CAST:
-                       /* outfuncoid refers to a cast function, not an output function */
-                       jsontext = DatumGetTextP(OidFunctionCall1(outfuncoid, val));
-                       outputstr = text_to_cstring(jsontext);
-                       appendStringInfoString(result, outputstr);
-                       pfree(outputstr);
-                       pfree(jsontext);
-                       break;
                default:
-                       outputstr = OidOutputFunctionCall(outfuncoid, val);
-                       if (key_scalar && *outputstr == '\0')
-                               ereport(ERROR,
-                                               (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
-                                                errmsg("key value must not be empty")));
-                       escape_json(result, outputstr);
-                       pfree(outputstr);
-                       break;
+                       elog(ERROR, "unknown jsonb value datetime type oid %d", typid);
+                       return NULL;
        }
+
+       return buf;
 }
 
 /*
@@ -1596,15 +1799,16 @@ composite_to_json(Datum composite, StringInfo result, bool use_line_feeds)
                char       *attname;
                JsonTypeCategory tcategory;
                Oid                     outfuncoid;
+               Form_pg_attribute att = TupleDescAttr(tupdesc, i);
 
-               if (tupdesc->attrs[i]->attisdropped)
+               if (att->attisdropped)
                        continue;
 
                if (needsep)
                        appendStringInfoString(result, sep);
                needsep = true;
 
-               attname = NameStr(tupdesc->attrs[i]->attname);
+               attname = NameStr(att->attname);
                escape_json(result, attname);
                appendStringInfoChar(result, ':');
 
@@ -1616,8 +1820,7 @@ composite_to_json(Datum composite, StringInfo result, bool use_line_feeds)
                        outfuncoid = InvalidOid;
                }
                else
-                       json_categorize_type(tupdesc->attrs[i]->atttypid,
-                                                                &tcategory, &outfuncoid);
+                       json_categorize_type(att->atttypid, &tcategory, &outfuncoid);
 
                datum_to_json(val, isnull, result, tcategory, outfuncoid, false);
        }
@@ -1660,7 +1863,7 @@ add_json(Datum val, bool is_null, StringInfo result,
 /*
  * SQL function array_to_json(row)
  */
-extern Datum
+Datum
 array_to_json(PG_FUNCTION_ARGS)
 {
        Datum           array = PG_GETARG_DATUM(0);
@@ -1676,7 +1879,7 @@ array_to_json(PG_FUNCTION_ARGS)
 /*
  * SQL function array_to_json(row, prettybool)
  */
-extern Datum
+Datum
 array_to_json_pretty(PG_FUNCTION_ARGS)
 {
        Datum           array = PG_GETARG_DATUM(0);
@@ -1693,7 +1896,7 @@ array_to_json_pretty(PG_FUNCTION_ARGS)
 /*
  * SQL function row_to_json(row)
  */
-extern Datum
+Datum
 row_to_json(PG_FUNCTION_ARGS)
 {
        Datum           array = PG_GETARG_DATUM(0);
@@ -1709,7 +1912,7 @@ row_to_json(PG_FUNCTION_ARGS)
 /*
  * SQL function row_to_json(row, prettybool)
  */
-extern Datum
+Datum
 row_to_json_pretty(PG_FUNCTION_ARGS)
 {
        Datum           array = PG_GETARG_DATUM(0);
@@ -1752,22 +1955,16 @@ to_json(PG_FUNCTION_ARGS)
 
 /*
  * json_agg transition function
+ *
+ * aggregate input column as a json array value.
  */
 Datum
 json_agg_transfn(PG_FUNCTION_ARGS)
 {
-       Oid                     val_type = get_fn_expr_argtype(fcinfo->flinfo, 1);
        MemoryContext aggcontext,
                                oldcontext;
-       StringInfo      state;
+       JsonAggState *state;
        Datum           val;
-       JsonTypeCategory tcategory;
-       Oid                     outfuncoid;
-
-       if (val_type == InvalidOid)
-               ereport(ERROR,
-                               (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
-                                errmsg("could not determine input data type")));
 
        if (!AggCheckCallContext(fcinfo, &aggcontext))
        {
@@ -1777,50 +1974,59 @@ json_agg_transfn(PG_FUNCTION_ARGS)
 
        if (PG_ARGISNULL(0))
        {
+               Oid                     arg_type = get_fn_expr_argtype(fcinfo->flinfo, 1);
+
+               if (arg_type == InvalidOid)
+                       ereport(ERROR,
+                                       (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
+                                        errmsg("could not determine input data type")));
+
                /*
-                * Make this StringInfo in a context where it will persist for the
+                * Make this state object in a context where it will persist for the
                 * duration of the aggregate call.  MemoryContextSwitchTo is only
                 * needed the first time, as the StringInfo routines make sure they
                 * use the right context to enlarge the object if necessary.
                 */
                oldcontext = MemoryContextSwitchTo(aggcontext);
-               state = makeStringInfo();
+               state = (JsonAggState *) palloc(sizeof(JsonAggState));
+               state->str = makeStringInfo();
                MemoryContextSwitchTo(oldcontext);
 
-               appendStringInfoChar(state, '[');
+               appendStringInfoChar(state->str, '[');
+               json_categorize_type(arg_type, &state->val_category,
+                                                        &state->val_output_func);
        }
        else
        {
-               state = (StringInfo) PG_GETARG_POINTER(0);
-               appendStringInfoString(state, ", ");
+               state = (JsonAggState *) PG_GETARG_POINTER(0);
+               appendStringInfoString(state->str, ", ");
        }
 
        /* fast path for NULLs */
        if (PG_ARGISNULL(1))
        {
-               datum_to_json((Datum) 0, true, state, JSONTYPE_NULL, InvalidOid, false);
+               datum_to_json((Datum) 0, true, state->str, JSONTYPE_NULL,
+                                         InvalidOid, false);
                PG_RETURN_POINTER(state);
        }
 
        val = PG_GETARG_DATUM(1);
 
-       /* XXX we do this every time?? */
-       json_categorize_type(val_type,
-                                                &tcategory, &outfuncoid);
-
        /* add some whitespace if structured type and not first item */
        if (!PG_ARGISNULL(0) &&
-               (tcategory == JSONTYPE_ARRAY || tcategory == JSONTYPE_COMPOSITE))
+               (state->val_category == JSONTYPE_ARRAY ||
+                state->val_category == JSONTYPE_COMPOSITE))
        {
-               appendStringInfoString(state, "\n ");
+               appendStringInfoString(state->str, "\n ");
        }
 
-       datum_to_json(val, false, state, tcategory, outfuncoid, false);
+       datum_to_json(val, false, state->str, state->val_category,
+                                 state->val_output_func, false);
 
        /*
-        * The transition type for array_agg() is declared to be "internal", which
+        * The transition type for json_agg() is declared to be "internal", which
         * is a pass-by-value type the same size as a pointer.  So we can safely
-        * pass the ArrayBuildState pointer through nodeAgg.c's machinations.
+        * pass the JsonAggState pointer through nodeAgg.c's machinations.
         */
        PG_RETURN_POINTER(state);
 }
@@ -1831,136 +2037,155 @@ json_agg_transfn(PG_FUNCTION_ARGS)
 Datum
 json_agg_finalfn(PG_FUNCTION_ARGS)
 {
-       StringInfo      state;
+       JsonAggState *state;
 
        /* cannot be called directly because of internal-type argument */
        Assert(AggCheckCallContext(fcinfo, NULL));
 
-       state = PG_ARGISNULL(0) ? NULL : (StringInfo) PG_GETARG_POINTER(0);
+       state = PG_ARGISNULL(0) ?
+               NULL :
+               (JsonAggState *) PG_GETARG_POINTER(0);
 
+       /* NULL result for no rows in, as is standard with aggregates */
        if (state == NULL)
                PG_RETURN_NULL();
 
-       appendStringInfoChar(state, ']');
-
-       PG_RETURN_TEXT_P(cstring_to_text_with_len(state->data, state->len));
+       /* Else return state with appropriate array terminator added */
+       PG_RETURN_TEXT_P(catenate_stringinfo_string(state->str, "]"));
 }
 
 /*
  * json_object_agg transition function.
  *
- * aggregate two input columns as a single json value.
+ * aggregate two input columns as a single json object value.
  */
 Datum
 json_object_agg_transfn(PG_FUNCTION_ARGS)
 {
-       Oid                     val_type;
        MemoryContext aggcontext,
                                oldcontext;
-       StringInfo      state;
+       JsonAggState *state;
        Datum           arg;
 
        if (!AggCheckCallContext(fcinfo, &aggcontext))
        {
                /* cannot be called directly because of internal-type argument */
-               elog(ERROR, "json_agg_transfn called in non-aggregate context");
+               elog(ERROR, "json_object_agg_transfn called in non-aggregate context");
        }
 
        if (PG_ARGISNULL(0))
        {
+               Oid                     arg_type;
+
                /*
-                * Make this StringInfo in a context where it will persist for the
-                * duration off the aggregate call. It's only needed for this initial
-                * piece, as the StringInfo routines make sure they use the right
-                * context to enlarge the object if necessary.
+                * Make the StringInfo in a context where it will persist for the
+                * duration of the aggregate call. Switching context is only needed
+                * for this initial step, as the StringInfo routines make sure they
+                * use the right context to enlarge the object if necessary.
                 */
                oldcontext = MemoryContextSwitchTo(aggcontext);
-               state = makeStringInfo();
+               state = (JsonAggState *) palloc(sizeof(JsonAggState));
+               state->str = makeStringInfo();
                MemoryContextSwitchTo(oldcontext);
 
-               appendStringInfoString(state, "{ ");
-       }
-       else
-       {
-               state = (StringInfo) PG_GETARG_POINTER(0);
-               appendStringInfoString(state, ", ");
-       }
+               arg_type = get_fn_expr_argtype(fcinfo->flinfo, 1);
 
-       if (PG_ARGISNULL(1))
-               ereport(ERROR,
-                               (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
-                                errmsg("field name must not be null")));
+               if (arg_type == InvalidOid)
+                       ereport(ERROR,
+                                       (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
+                                        errmsg("could not determine data type for argument %d", 1)));
 
-       val_type = get_fn_expr_argtype(fcinfo->flinfo, 1);
+               json_categorize_type(arg_type, &state->key_category,
+                                                        &state->key_output_func);
 
-       /*
-        * turn a constant (more or less literal) value that's of unknown type
-        * into text. Unknowns come in as a cstring pointer.
-        */
-       if (val_type == UNKNOWNOID && get_fn_expr_arg_stable(fcinfo->flinfo, 1))
-       {
-               val_type = TEXTOID;
-               arg = CStringGetTextDatum(PG_GETARG_POINTER(1));
+               arg_type = get_fn_expr_argtype(fcinfo->flinfo, 2);
+
+               if (arg_type == InvalidOid)
+                       ereport(ERROR,
+                                       (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
+                                        errmsg("could not determine data type for argument %d", 2)));
+
+               json_categorize_type(arg_type, &state->val_category,
+                                                        &state->val_output_func);
+
+               appendStringInfoString(state->str, "{ ");
        }
        else
        {
-               arg = PG_GETARG_DATUM(1);
+               state = (JsonAggState *) PG_GETARG_POINTER(0);
+               appendStringInfoString(state->str, ", ");
        }
 
-       if (val_type == InvalidOid || val_type == UNKNOWNOID)
+       /*
+        * Note: since json_object_agg() is declared as taking type "any", the
+        * parser will not do any type conversion on unknown-type literals (that
+        * is, undecorated strings or NULLs).  Such values will arrive here as
+        * type UNKNOWN, which fortunately does not matter to us, since
+        * unknownout() works fine.
+        */
+
+       if (PG_ARGISNULL(1))
                ereport(ERROR,
                                (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
-                                errmsg("could not determine data type for argument %d", 1)));
+                                errmsg("field name must not be null")));
 
-       add_json(arg, false, state, val_type, true);
+       arg = PG_GETARG_DATUM(1);
 
-       appendStringInfoString(state, " : ");
+       datum_to_json(arg, false, state->str, state->key_category,
+                                 state->key_output_func, true);
 
-       val_type = get_fn_expr_argtype(fcinfo->flinfo, 2);
-       /* see comments above */
-       if (val_type == UNKNOWNOID && get_fn_expr_arg_stable(fcinfo->flinfo, 2))
-       {
-               val_type = TEXTOID;
-               if (PG_ARGISNULL(2))
-                       arg = (Datum) 0;
-               else
-                       arg = CStringGetTextDatum(PG_GETARG_POINTER(2));
-       }
+       appendStringInfoString(state->str, " : ");
+
+       if (PG_ARGISNULL(2))
+               arg = (Datum) 0;
        else
-       {
                arg = PG_GETARG_DATUM(2);
-       }
 
-       if (val_type == InvalidOid || val_type == UNKNOWNOID)
-               ereport(ERROR,
-                               (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
-                                errmsg("could not determine data type for argument %d", 2)));
-
-       add_json(arg, PG_ARGISNULL(2), state, val_type, false);
+       datum_to_json(arg, PG_ARGISNULL(2), state->str, state->val_category,
+                                 state->val_output_func, false);
 
        PG_RETURN_POINTER(state);
 }
 
 /*
  * json_object_agg final function.
- *
  */
 Datum
 json_object_agg_finalfn(PG_FUNCTION_ARGS)
 {
-       StringInfo      state;
+       JsonAggState *state;
 
        /* cannot be called directly because of internal-type argument */
        Assert(AggCheckCallContext(fcinfo, NULL));
 
-       state = PG_ARGISNULL(0) ? NULL : (StringInfo) PG_GETARG_POINTER(0);
+       state = PG_ARGISNULL(0) ? NULL : (JsonAggState *) PG_GETARG_POINTER(0);
 
+       /* NULL result for no rows in, as is standard with aggregates */
        if (state == NULL)
-               PG_RETURN_TEXT_P(cstring_to_text("{}"));
+               PG_RETURN_NULL();
 
-       appendStringInfoString(state, " }");
+       /* Else return state with appropriate object terminator added */
+       PG_RETURN_TEXT_P(catenate_stringinfo_string(state->str, " }"));
+}
 
-       PG_RETURN_TEXT_P(cstring_to_text_with_len(state->data, state->len));
+/*
+ * Helper function for aggregates: return given StringInfo's contents plus
+ * specified trailing string, as a text datum.  We need this because aggregate
+ * final functions are not allowed to modify the aggregate state.
+ */
+static text *
+catenate_stringinfo_string(StringInfo buffer, const char *addon)
+{
+       /* custom version of cstring_to_text_with_len */
+       int                     buflen = buffer->len;
+       int                     addlen = strlen(addon);
+       text       *result = (text *) palloc(buflen + addlen + VARHDRSZ);
+
+       SET_VARSIZE(result, buflen + addlen + VARHDRSZ);
+       memcpy(VARDATA(result), buffer->data, buflen);
+       memcpy(VARDATA(result) + buflen, addon, addlen);
+
+       return result;
 }
 
 /*
@@ -1971,16 +2196,25 @@ json_build_object(PG_FUNCTION_ARGS)
 {
        int                     nargs = PG_NARGS();
        int                     i;
-       Datum           arg;
-       char       *sep = "";
+       const char *sep = "";
        StringInfo      result;
-       Oid                     val_type;
+       Datum      *args;
+       bool       *nulls;
+       Oid                *types;
+
+       /* fetch argument values to build the object */
+       nargs = extract_variadic_args(fcinfo, 0, false, &args, &types, &nulls);
+
+       if (nargs < 0)
+               PG_RETURN_NULL();
 
        if (nargs % 2 != 0)
                ereport(ERROR,
                                (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
                                 errmsg("argument list must have even number of elements"),
-                                errhint("The arguments of json_build_object() must consist of alternating keys and values.")));
+               /* translator: %s is a SQL function name */
+                                errhint("The arguments of %s must consist of alternating keys and values.",
+                                                "json_build_object()")));
 
        result = makeStringInfo();
 
@@ -1988,61 +2222,22 @@ json_build_object(PG_FUNCTION_ARGS)
 
        for (i = 0; i < nargs; i += 2)
        {
-               /* process key */
+               appendStringInfoString(result, sep);
+               sep = ", ";
 
-               if (PG_ARGISNULL(i))
+               /* process key */
+               if (nulls[i])
                        ereport(ERROR,
                                        (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
                                         errmsg("argument %d cannot be null", i + 1),
                                         errhint("Object keys should be text.")));
-               val_type = get_fn_expr_argtype(fcinfo->flinfo, i);
 
-               /*
-                * turn a constant (more or less literal) value that's of unknown type
-                * into text. Unknowns come in as a cstring pointer.
-                */
-               if (val_type == UNKNOWNOID && get_fn_expr_arg_stable(fcinfo->flinfo, i))
-               {
-                       val_type = TEXTOID;
-                       arg = CStringGetTextDatum(PG_GETARG_POINTER(i));
-               }
-               else
-               {
-                       arg = PG_GETARG_DATUM(i);
-               }
-               if (val_type == InvalidOid || val_type == UNKNOWNOID)
-                       ereport(ERROR,
-                                       (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
-                                        errmsg("could not determine data type for argument %d",
-                                                       i + 1)));
-               appendStringInfoString(result, sep);
-               sep = ", ";
-               add_json(arg, false, result, val_type, true);
+               add_json(args[i], false, result, types[i], true);
 
                appendStringInfoString(result, " : ");
 
                /* process value */
-
-               val_type = get_fn_expr_argtype(fcinfo->flinfo, i + 1);
-               /* see comments above */
-               if (val_type == UNKNOWNOID && get_fn_expr_arg_stable(fcinfo->flinfo, i + 1))
-               {
-                       val_type = TEXTOID;
-                       if (PG_ARGISNULL(i + 1))
-                               arg = (Datum) 0;
-                       else
-                               arg = CStringGetTextDatum(PG_GETARG_POINTER(i + 1));
-               }
-               else
-               {
-                       arg = PG_GETARG_DATUM(i + 1);
-               }
-               if (val_type == InvalidOid || val_type == UNKNOWNOID)
-                       ereport(ERROR,
-                                       (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
-                                        errmsg("could not determine data type for argument %d",
-                                                       i + 2)));
-               add_json(arg, PG_ARGISNULL(i + 1), result, val_type, false);
+               add_json(args[i + 1], nulls[i + 1], result, types[i + 1], false);
        }
 
        appendStringInfoChar(result, '}');
@@ -2065,13 +2260,19 @@ json_build_object_noargs(PG_FUNCTION_ARGS)
 Datum
 json_build_array(PG_FUNCTION_ARGS)
 {
-       int                     nargs = PG_NARGS();
+       int                     nargs;
        int                     i;
-       Datum           arg;
-       char       *sep = "";
+       const char *sep = "";
        StringInfo      result;
-       Oid                     val_type;
+       Datum      *args;
+       bool       *nulls;
+       Oid                *types;
 
+       /* fetch argument values to build the array */
+       nargs = extract_variadic_args(fcinfo, 0, false, &args, &types, &nulls);
+
+       if (nargs < 0)
+               PG_RETURN_NULL();
 
        result = makeStringInfo();
 
@@ -2079,34 +2280,14 @@ json_build_array(PG_FUNCTION_ARGS)
 
        for (i = 0; i < nargs; i++)
        {
-               val_type = get_fn_expr_argtype(fcinfo->flinfo, i);
-               arg = PG_GETARG_DATUM(i + 1);
-               /* see comments in json_build_object above */
-               if (val_type == UNKNOWNOID && get_fn_expr_arg_stable(fcinfo->flinfo, i))
-               {
-                       val_type = TEXTOID;
-                       if (PG_ARGISNULL(i))
-                               arg = (Datum) 0;
-                       else
-                               arg = CStringGetTextDatum(PG_GETARG_POINTER(i));
-               }
-               else
-               {
-                       arg = PG_GETARG_DATUM(i);
-               }
-               if (val_type == InvalidOid || val_type == UNKNOWNOID)
-                       ereport(ERROR,
-                                       (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
-                                        errmsg("could not determine data type for argument %d",
-                                                       i + 1)));
                appendStringInfoString(result, sep);
                sep = ", ";
-               add_json(arg, PG_ARGISNULL(i), result, val_type, false);
+               add_json(args[i], nulls[i], result, types[i], false);
        }
+
        appendStringInfoChar(result, ']');
 
        PG_RETURN_TEXT_P(cstring_to_text_with_len(result->data, result->len));
-
 }
 
 /*
@@ -2303,7 +2484,7 @@ escape_json(StringInfo buf, const char *str)
 {
        const char *p;
 
-       appendStringInfoCharMacro(buf, '\"');
+       appendStringInfoCharMacro(buf, '"');
        for (p = str; *p; p++)
        {
                switch (*p)
@@ -2327,29 +2508,7 @@ escape_json(StringInfo buf, const char *str)
                                appendStringInfoString(buf, "\\\"");
                                break;
                        case '\\':
-                               /*
-                                * Unicode escapes are passed through as is. There is no
-                                * requirement that they denote a valid character in the
-                                * server encoding - indeed that is a big part of their
-                                * usefulness.
-                                *
-                                * All we require is that they consist of \uXXXX where
-                                * the Xs are hexadecimal digits. It is the responsibility
-                                * of the caller of, say, to_json() to make sure that the
-                                * unicode escape is valid.
-                                *
-                                * In the case of a jsonb string value being escaped, the
-                                * only unicode escape that should be present is \u0000,
-                                * all the other unicode escapes will have been resolved.
-                                */
-                               if (p[1] == 'u' &&
-                                       isxdigit((unsigned char) p[2]) &&
-                                       isxdigit((unsigned char) p[3]) &&
-                                       isxdigit((unsigned char) p[4]) &&
-                                       isxdigit((unsigned char) p[5]))
-                                       appendStringInfoCharMacro(buf, *p);
-                               else
-                                       appendStringInfoString(buf, "\\\\");
+                               appendStringInfoString(buf, "\\\\");
                                break;
                        default:
                                if ((unsigned char) *p < ' ')
@@ -2359,7 +2518,7 @@ escape_json(StringInfo buf, const char *str)
                                break;
                }
        }
-       appendStringInfoCharMacro(buf, '\"');
+       appendStringInfoCharMacro(buf, '"');
 }
 
 /*
@@ -2383,7 +2542,7 @@ json_typeof(PG_FUNCTION_ARGS)
        JsonTokenType tok;
        char       *type;
 
-       json = PG_GETARG_TEXT_P(0);
+       json = PG_GETARG_TEXT_PP(0);
        lex = makeJsonLexContext(json, false);
 
        /* Lex exactly one token from the input and check its type. */