16 Clauses Used in SQL Query Functions and Conditions

Clauses RETURNING, wrapper, error, and empty-field are described. Each is used in one or more of the SQL functions and conditions json_value, json_query, json_table, json_serialize, json_mergepatch, is json, is not json, json_exists, and json_equal.

16.1 RETURNING Clause for SQL Query Functions

SQL functions json_value, json_query, json_serialize, and json_mergepatch accept an optional RETURNING clause, which specifies the data type of the value returned by the function. This clause and the default behavior (no RETURNING clause) are described here.

For json_value, you can use any of these predefined SQL data types in a RETURNING clause: VARCHAR2, NUMBER, DATE, TIMESTAMP, TIMESTAMP WITH TIME ZONE, SDO_GEOMETRY, and CLOB. You can also use a user-defined object type or a collection type.


An instance of Oracle SQL data type DATE includes a time component. And in your JSON data you can use a string that represents an ISO 8601 date-with-time value, that is, it can have a time component.

By default, json_value with RETURNING DATE returns a SQL DATE value that has a zero time component (zero hours, minutes, and seconds). By default, a time component in the queried JSON scalar value is truncated in the returned SQL DATE instance. But before any time truncation is done, if the value represented by an ISO 8601 date-with-time string has a time-zone component then the value is first converted to UTC, to take any time-zone information into account.

You can use RETURNING DATE PRESERVE TIME to override this default truncating behavior and preserve the time component, when present, of the queried JSON scalar value. (Using RETURNING DATE TRUNCATE TIME has the same effect as just RETURNING DATE, the default behavior.)

(The same considerations apply to item methods date(), which corresponds to TRUNCATE TIME, and dateWithTime(), which corresponds to PRESERVE TIME.)

For json_query, json_serialize, and json_mergepatch you can use VARCHAR2, CLOB, or BLOB. A BLOB result is in the AL32UTF8 character set. Whatever the data type returned by json_serialize, the returned data represents textual JSON data.

You can optionally specify a length for VARCHAR2 (default: 4000) and a precision and scale for NUMBER.

The default behavior (no RETURNING clause) is to use VARCHAR2(4000).

Data type SDO_GEOMETRY is for Oracle Spatial and Graph data. In particular, this means that you can use json_value with GeoJSON data, which is a format for encoding geographic data in JSON.

The RETURNING clause also accepts two optional keywords, PRETTY and ASCII. If both are present then PRETTY must come before ASCII. Keyword PRETTY is not allowed for json_value.

The effect of keyword PRETTY is to pretty-print the returned data, by inserting newline characters and indenting. The default behavior is not to pretty-print.

The effect of keyword ASCII is to automatically escape all non-ASCII Unicode characters in the returned data, using standard ASCII Unicode escape sequences. The default behavior is not to escape non-ASCII Unicode characters.


You can pretty-print the entire context item by using only $ as the path expression.

If VARCHAR2 is specified in a RETURNING clause then scalars in the value are represented as follows:

  • Boolean values are represented by the lowercase strings "true" and "false".

  • The null value is represented by SQL NULL.

  • A JSON number is represented in a canonical form. It can thus appear differently in the output string from its representation in textual input data. When represented in canonical form:
    • It can be subject to the precision and range limitations for a SQL NUMBER.

    • When it is not subject to the SQL NUMBER limitations:
      • The precision is limited to forty (40) digits.

      • The optional exponent is limited to nine (9) digits plus a sign (+ or -).

      • The entire text, including possible signs (-, +), decimal point (.), and exponential indicator (E), is limited to 48 characters.

    The canonical form of a JSON number:

    • Is a JSON number. (It can be parsed in JSON data as a number.)

    • Does not have a leading plus (+) sign.

    • Has a decimal point (.) only when necessary.

    • Has a single zero (0) before the decimal point if the number is a fraction (between zero and one).

    • Uses exponential notation (E) only when necessary. In particular, this can be the case if the number of output characters is too limited (by a small N for VARCHAR2(N)).

Oracle extends the SQL/JSON standard in the case when the returning data type is VARCHAR2(N), by allowing optional keyword TRUNCATE immediately after the data type. When TRUNCATE is present and the value to return is wider than N, the value is truncated — only the first N characters are returned. If TRUNCATE is absent then this case is treated as an error, handled as usual by an error clause or the default error-handling behavior.

See Also:

16.2 Wrapper Clause for SQL/JSON Query Functions JSON_QUERY and JSON_TABLE

SQL/JSON query functions json_query and json_table accept an optional wrapper clause, which specifies the form of the value returned by json_query or used for the data in a json_table column. This clause and the default behavior (no wrapper clause) are described here. Examples are provided.

The wrapper clause takes one of these forms:

  • WITH WRAPPER – Use a string value that represents a JSON array containing all of the JSON values that match the path expression. The order of the array elements is unspecified.

  • WITHOUT WRAPPER – Use a string value that represents the single JSON object or array that matches the path expression. Raise an error if the path expression matches either a scalar value (not an object or array) or more than one value.

  • WITH CONDITIONAL WRAPPER – Use a string value that represents all of the JSON values that match the path expression. For zero values, a single scalar value, or multiple values, WITH CONDITIONAL WRAPPER is the same as WITH WRAPPER. For a single JSON object or array value, it is the same as WITHOUT WRAPPER.

The default behavior is WITHOUT WRAPPER.

You can add the optional keyword UNCONDITIONAL immediately after keyword WITH, if you find it clearer: WITH WRAPPER and WITH UNCONDITIONAL WRAPPER mean the same thing.

You can add the optional keyword ARRAY immediately before keyword WRAPPER, if you find it clearer: WRAPPER and ARRAY WRAPPER mean the same thing.

Table 16-1 illustrates the wrapper clause possibilities. The array wrapper is shown in bold.

Table 16-1 JSON_QUERY Wrapper Clause Examples


{"id": 38327} (single object)

[{"id": 38327}]

{"id": 38327}

{"id": 38327}

[42, "a", true] (single array)

[[42, "a", true]]

[42, "a", true]

[42, "a", true]



Error (scalar)


42, "a", true

[42, "a", true]

Error (multiple values)

[42, "a", true]



Error (no values)


Consider, for example, a json_query query to retrieve a JSON object. What happens if the path expression matches a JSON scalar value instead of an object, or it matches multiple JSON values (of any kind)? You might want to retrieve the matched values instead of raising an error. For example, you might want to pick one of the values that is an object, for further processing. Using an array wrapper lets you do this.

A conditional wrapper can be convenient if the only reason you are using a wrapper is to avoid raising an error and you do not need to distinguish those error cases from non-error cases. If your application is looking for a single object or array and the data matched by a path expression is just that, then there is no need to wrap that expected value in a singleton array.

On the other hand, with an unconditional wrapper you know that the resulting array is always a wrapper — your application can count on that. If you use a conditional wrapper then your application might need extra processing to interpret a returned array. In Table 16-1, for instance, note that the same array ([42, "a", true]) is returned for the very different cases of a path expression matching that array and a path expression matching each of its elements.

16.3 Error Clause for SQL Query Functions and Conditions

Some SQL query functions and conditions accept an optional error clause, which specifies handling for a runtime error that is raised by the function or condition. This clause and the default behavior (no error clause) are summarized here.

By default, SQL functions and conditions for JSON avoid raising runtime errors. For example, when JSON data is syntactically invalid, json_exists and json_equal return false and json_value returns NULL.

But in some cases you can also specify an error clause, which overrides the default behavior. The error handling you can specify varies, but each SQL function and condition for JSON that lets you specify error handling supports at least the ERROR ON ERROR behavior of raising an error.

The optional error clause can take these forms:

  • ERROR ON ERROR – Raise the error (no special handling).

  • NULL ON ERROR – Return NULL instead of raising the error.

    Not available for json_exists.

  • FALSE ON ERROR – Return false instead of raising the error.

    Available only for json_exists and json_equal, for which it is the default.

  • TRUE ON ERROR – Return true instead of raising the error.

    Available only for json_exists and json_equal.

  • EMPTY OBJECT ON ERROR – Return an empty object ({}) instead of raising the error.

    Available only for json_query.

  • EMPTY ARRAY ON ERROR – Return an empty array ([]) instead of raising the error.

    Available only for json_query.


  • DEFAULT 'literal_return_value' ON ERROR – Return the specified value instead of raising the error. The value must be a constant at query compile time.

    Not available:

    • For json_exists, json_equal, json_serialize, json_mergepatch, or a json_table column value clause that has json_exists behavior

    • For json_query or a json_table column value clause that has json_query behavior

    • For row-level error-handing for json_table

    • When SDO_GEOMETRY is specified either as the RETURNING clause data type for json_value or as a json_table column data type

The default behavior is NULL ON ERROR, except for conditions json_exists and json_equal.


There are two levels of error handling for json_table, corresponding to its two levels of path expressions: row and column. When present, a column error handler overrides row-level error handling. The default error handler for both levels is NULL ON ERROR.


An ON EMPTY clause overrides the behavior specified by ON ERROR for the error of trying to match a missing field.


The ON ERROR clause takes effect only for runtime errors that arise when a syntactically correct SQL/JSON path expression is matched against JSON data. A path expression that is syntactically incorrect results in a compile-time syntax error; it is not handled by the ON ERROR clause.

See Also:

16.4 Empty-Field Clause for SQL/JSON Query Functions

SQL/JSON query functions json_value, json_query, and json_table accept an optional ON EMPTY clause, which specifies the handling to use when a targeted JSON field is absent from the data queried. This clause and the default behavior (no ON EMPTY clause) are described here.

You generally handle errors for SQL/JSON functions and conditions using an error clause (ON ERROR). However, there is a special case where you might want different handling from this general error handling: when querying to match given JSON fields that are missing from the data. Sometimes you do not want to raise an error just because a field to be matched is absent. (A missing field is normally treated as an error.)

You typically use a NULL ON EMPTY clause in conjunction with an accompanying ON ERROR clause. This combination specifies that other errors are handled according to the ON ERROR clause, but the error of trying to match a missing field is handled by just returning NULL. If no ON EMPTY clause is present then an ON ERROR clause handles also the missing-field case.

In addition to NULL ON EMPTY there are ERROR ON EMPTY and DEFAULT ... ON EMPTY, which are analogous to the similarly named ON ERROR clauses.

If only an ON EMPTY clause is present (no ON ERROR clause) then missing-field behavior is specified by the ON EMPTY clause, and other errors are handled the same as if NULL ON ERROR were present (it is the ON ERROR default). If both clauses are absent then only NULL ON ERROR is used.

Use NULL ON EMPTY for an Index Created on JSON_VALUE

NULL ON EMPTY is especially useful for the case of a functional index created on a json_value expression. The clause has no effect on whether or when the index is picked up, but it is effective in allowing some data to be indexed that would otherwise not be because it is missing a field targeted by the json_value expression.

You generally want to use ERROR ON ERROR for the queries that populate the index, so that a query path expression that results in multiple values or complex values raises an error. But you sometimes do not want to raise an error just because the field targeted by a path expression is missing — you want that data to be indexed. Example 28-5 illustrates this use of NULL ON EMPTY when creating an index on a json_value expression.


When the RETURNING clause specifies a user-defined object-type or collection-type instance, function json_value accepts an optional ON MISMATCH clause, which specifies handling to use when a targeted JSON value does not match the specified SQL return value. This clause and its default behavior (no ON MISMATCH clause) are described here.


Clauses ON ERROR and ON EMPTY apply only when no JSON data matches the path expression. Clause ON MISMATCH applies when the JSON data that matches the path expression does not match the specified object or collection return type. Clause ON MISMATCH is allowed only when the return type is an instance of a user-defined object type or collection type. If you use it with another return type then a query compile-time error is raised.

When you return an instance of a SQL object or collection type that reflects the JSON data targeted by function json_value, the definitions of that targeted data and the object or collection to be returned must match, or else a query compile-time error applies.

The default handling of such an error is just to ignore it. But you can instead handle such an error in various ways, by providing one or more ON MISMATCH clauses, as follows:

  • IGNORE ON MISMATCH — Explicitly specify the default behavior: ignore the mismatch. The object or collection instance returned can contain one or more SQL NULL values because of mismatches against the targeted JSON data.

  • NULL ON MISMATCH — Return SQL NULL as the object-type or collection-type value.

  • ERROR ON MISMATCH — Raise a query compile-time error for the mismatch.

Each of these ON MISMATCH clause types can also be followed, in parentheses (()), by one or more clauses that each indicates a kind of mismatch to handle, separated by commas (,). These are the possible mismatch kinds:

  • MISSING DATA — Some JSON data was needed to match the object-type or collection-type data, but it was missing.

  • EXTRA DATA — One or more JSON fields have no corresponding object-type or collection-type data. For example, for JSON field address there is no object-type attribute with the same name (matching case-insensitively, by default).

  • TYPE ERROR — A JSON scalar value has a data type that is incompatible with the corresponding return SQL scalar data type. This can be because of type incompatibility, as put forth in Table 18-1, or because the SQL data type is too constraining (e.g., VARCHAR(2) is two short for JSON string "hello").

If no such kind-of-mismatch clause (e.g. EXTRA DATA) is present for a given handler (e.g. NULL ON MISMATCH) then that handler applies to all kinds of mismatch.

You can have any number of ON MISMATCH clauses of different kinds, but if two or more such contradict each other then a query compile-time error is raised.