Complex type functions

boolean all_match(array<T> <a>, function<T, boolean> <predicate>)

Checks whether all the elements in Array a meet the predicate condition.

• a: required. This parameter specifies an array that T in array<T> specifies the data type of the elements in the array. The elements can be of any data type.
• predicate: required. This parameter specifies the built-in function, user-defined function, or expression that is used to determine whether all the elements in Array a meet requirements. The data type of the input parameter must be the same as the data type of the elements in Array a.

• Example 1: Check whether all the elements in array(4, 5, 6) meet the x-> x > 3 condition. This condition specifies that all elements are greater than 3. Sample statement:

  -- True is returned. 
  select all_match(array(4, 5, 6), x -> x>3);

• Example 2: The array is empty. Sample statement:

  -- True is returned. 
  select all_match(array(), x -> x>3);

• Example 3: Check whether all the elements in array(1, 2, -10, 100, -30) meet the x-> x > 3 condition. Sample statement:

  -- False is returned. 
  select all_match(array(1, 2, -10, 100, -30), x -> x>3);

• Example 4: Check whether all the elements in array(10, 100, 30, null) meet the x-> x > 3 condition. Sample statement:

  -- NULL is returned. 
  select all_match(array(10, 100, 30, null), x -> x>3);

boolean any_match(array<T> <a>, function<T, boolean> <predicate>)

Checks whether an element in Array a meets the predicate condition.

• a: required. This parameter specifies an array that T in array<T> specifies the data type of the elements in the array. The elements can be of any data type.
• predicate: required. This parameter specifies the built-in function, user-defined function, or expression that is used to determine whether all the elements in Array a meet requirements. The data type of the input parameter must be the same as the data type of the elements in Array a.

• Example 1: Check whether an element in array(1, 2, -10, 100, -30) meets the x-> x > 3 condition. Sample statement:

  -- True is returned. 
  select any_match(array(1, 2, -10, 100, -30), x-> x > 3);

• Example 2: The array is empty. Sample statement:

  -- False is returned. 
  select any_match(array(), x-> x > 3);

• Example 3: Check whether an element in array(1, 2, -10, -20, -30) meets the x-> x > 3 condition. Sample statement:

  -- False is returned. 
  select any_match(array(1, 2, -10, -20, -30), x-> x > 3);

• Example 4: Check whether an element in array(1, 2, null, -10) meets the x-> x > 3 condition. Sample statement:

  -- NULL is returned. 
  select any_match(array(1, 2, null, -10), x-> x > 3);

array array(<value>,<value>[, ...])

Creates an array based on given values.

value: required. All data types are supported. All values must be of the same data type.

A value of the ARRAY type is returned.

boolean array_contains(array<T> <a>, value <v>)

Checks whether Array a contains Element v.

• a: required. This parameter specifies an array that T in array<T> specifies the data type of the elements in the array. The elements can be of any data type.
• v: required. This parameter specifies the element that you want to check. The value of v must be of the same data type as the elements in Array a.

A value of the BOOLEAN type is returned.

array<T> array_distinct(array<T> <a>)

Removes duplicate elements from Array a.

a: required. This parameter specifies an array that T in array<T> specifies the data type of the elements in the array. The elements can be of any data type.

• Example 1: Remove duplicate elements from array(10, 20, 30, 30, 20, 10). Sample statement:

  -- [10,20,30] is returned. 
  select array_distinct(array(10, 20, 30, 30, 20, 10));

• Example 2: Remove duplicate elements from array(10, 20, 20, null, null, 30, 20, null). Sample statement:

  -- [10,20,null,30] is returned. 
  select array_distinct(array(10, 20, 20, null, null, 30, 20, null)); 

• Example 3: The array is empty. Sample statement:

  -- [] is returned. 
  select array_distinct(array());

array<T> array_except(array<T> <a>, array<T> <b>)

Finds the elements that exist in Array a but do not exist in Array b and returns the elements as a new array without duplicates.

a and b: required. These parameters specify arrays that T in array<T> specifies the data type of the elements in the arrays. The elements can be of any data type. The elements in Array a and the elements in Array b must be of the same data type.

• Example 1: Find the elements that exist in array(1, 1, 3, 3, 5, 5) but do not exist in array(1, 1, 2, 2, 3, 3) and return the elements as a new array without duplicates. Sample statement:

  -- [5] is returned. 
  select array_except(array(1, 1, 3, 3, 5, 5), array(1, 1, 2, 2, 3, 3));

• Example 2: Find the elements that exist in array(1, 1, 3, 3, 5, 5, null, null) but do not exist in array(1, 1, 2, 2, 3, 3) and return the elements as a new array without duplicates. Sample statement:

  -- [5,null] is returned. 
  select array_except(array(1, 1, 3, 3, 5, 5, null, null), array(1, 1, 2, 2, 3, 3));

• Example 3: One of the input arrays is empty. Sample statement:

  -- [2,1] is returned. 
  select array_except(array(2, 1, 1, 2), cast(array() as array<int>)); 

• Example 4: The two input arrays are empty. Sample statement:

  -- [] is returned. 
  select array_except(cast(array() as array<int>), cast(array() as array<int>));

array<T> array_intersect(array<T> <a>, array<T> <b>)  

Calculates the intersection of Array a and Array b and removes duplicate elements.

a and b: required. These parameters specify arrays that T in array<T> specifies the data type of the elements in the arrays. The elements can be of any data type. The elements in Array a and the elements in Array b must be of the same data type.

• Example 1: Calculate the intersection of array(1, 2, 3) and array(1, 3, 5), and remove duplicate elements. Sample statement:

  -- [1,3] is returned. 
  select array_intersect(array(1, 2, 3), array(1, 3, 5));

• Example 2: Calculate the intersection of array(10, 20, 20, 30, 30, null, null) and array(30, 30, 20, 20, 40, null, null), and remove duplicate elements. Sample statement:

  -- [20,30,null] is returned. 
  select array_intersect(array(10, 20, 20, 30, 30, null, null), array(30, 30, 20, 20, 40, null, null)); 

array_join(array<T> <a>, <delimiter>[, <nullreplacement>]) 

Concatenates the elements in Array a by using a delimiter. If the array contains a null element, use nullreplacement to specify the string that you want to use to replace the null element in the results. If you do not configure nullreplacement, the null element is ignored.

• a: required. This parameter specifies an array that T in array<T> specifies the data type of the elements in the array.
  Note If the elements in the array are not of the STRING type, MaxCompute converts the data type of the elements into STRING.
• delimiter: required. A value of the STRING type. This parameter specifies the string that is used to separate the concatenated elements in Array a.
• nullreplacement: optional. This parameter specifies the string that is used to replace null elements.

A value of the STRING type is returned.

-- 10,20,20,30 is returned. 
select array_join(array(10, 20, 20, null, null, 30), ",");
-- 10##20##20##null##null##30 is returned. 
select array_join(array(10, 20, 20, null, null, 30), "##", "null");

T array_max(array<T> <a>)  

Returns the largest element in Array a.

a: required. This parameter specifies an array that T in array<T> specifies the data type of the elements in the array.

-- 20 is returned. 
select array_max(array(1, 20, null, 3));

T array_min(array<T> <a>)  

Returns the smallest element in Array a.

a: required. This parameter specifies an array that T in array<T> specifies the data type of the elements in the array.

-- 1 is returned. 
select array_min(array(1, 20, null, 3));

bigint array_position(array<T> <a>, T <element>) 

Returns the position of the first occurrence of a given element in Array a. The position numbers of elements are counted from left to right and start from 1.

• a: required. This parameter specifies an array that T in array<T> specifies the data type of the elements in the array. The following data types are supported:
  • TINYINT, SMALLINT, INT, and BIGINT
  • FLOAT and DOUBLE
  • BOOLEAN
  • DECIMAL and DECIMALVAL
  • DATE, DATETIME, TIMESTAMP, IntervalDayTime, and IntervalYearMonth
  • STRING, BINARY, VARCHAR, and CHAR
  • ARRAY, STRUCT, and MAP
• element: required. The element whose position you want to query. The data type of this parameter must be the same as the data type of the elements in Array a.

• Example 1: Return the position of the first occurrence of 1 in array(3, 2, 1). Sample statement:

  -- 3 is returned. 
  select array_position(array(3, 2, 1), 1);

• Example 2: element is null. Sample statement:

  -- NULL is returned. 
  select array_position(array(3, 1, null), null);

array<T> array_remove(array<T> <a>, T <element>) 

Removes a given element from Array a.

• a: required. This parameter specifies an array that T in array<T> specifies the data type of the elements in the array. The following data types are supported:
  • TINYINT, SMALLINT, INT, and BIGINT
  • FLOAT and DOUBLE
  • BOOLEAN
  • DECIMAL and DECIMALVAL
  • DATE, DATETIME, TIMESTAMP, IntervalDayTime, and IntervalYearMonth
  • STRING, BINARY, VARCHAR, and CHAR
  • ARRAY, STRUCT, and MAP
• element: required. The element that you want to remove. The data type of this parameter must be the same as the data type of the elements in Array a.

• Example 1: Remove 1 from array(3, 2, 1). Sample statement:

  -- [3,2] is returned. 
  select array_remove(array(3, 2, 1), 1);

• Example 2: element is null. Sample statement:

  -- NULL is returned. 
  select array_remove(array(3, 1, null), null);

• Example 3: Remove 2 from array(3, 1, null). Sample statement:

  -- [3,1,null] is returned. 
  select array_remove(array(3, 1, null), 2);

R array_reduce(array<T> <a>, buf <init>, function<buf, T, buf> <merge>, function<buf, R> <final>)

Aggregates the elements in Array a.

• a: required. This parameter specifies an array that T in array<T> specifies the data type of the elements in the array. The elements can be of any data type.
• init: required. The initial value of the intermediate result that is used to aggregate elements.
• merge: required. A built-in function, user-defined function, or expression that is used to perform an operation on each element in Array a and the intermediate result. This function or expression uses the elements of Array a and the init parameter as input parameters.
• final: required. A built-in function, user-defined function, or expression that is used to convert the intermediate result into the final result. This function or expression uses the result of merge as the input parameter. R specifies the data type of the output.

The data type of the return value is the same as the data type that is specified for final.

-- 6 is returned. 
select array_reduce(array(1, 2, 3), 0, (buf, e)->buf + e, buf->buf);
-- 2.5 is returned. 
select array_reduce(array(1, 2, 3, 4), named_struct('sum', 0, 'count', 0), (buf, e)->named_struct('sum', buf.sum + e, 'count', buf.count + 1), buf -> buf.sum / buf.count);

array<T> array_repeat(T <element>, int <count>) 

Returns a new array in which element t is repeated count times.

• t: required. This parameter specifies the element that you want to repeat. The following data types are supported:
  • TINYINT, SMALLINT, INT, and BIGINT
  • FLOAT and DOUBLE
  • BOOLEAN
  • DECIMAL and DECIMALVAL
  • DATE, DATETIME, TIMESTAMP, IntervalDayTime, and IntervalYearMonth
  • STRING, BINARY, VARCHAR, and CHAR
  • ARRAY, STRUCT, and MAP
• count: required. This parameter specifies the number of repetitions. A value of the INT type is required. The value must be greater than or equal to 0.

• Example 1: Repeat 123 twice and return the new array. Sample statement:

  -- [123, 123] is returned. 
  select array_repeat('123', 2);

• Example 2: The value of count is null. Sample statement:

  -- NULL is returned. 
  select array_repeat('123', null);

• Example 3: The value of count is less than 0. Sample statement:

  -- [] is returned. 
  select array_repeat('123', -1);

array<T> array_sort(array<T> <a>, function<T, T, bigint> <comparator>) 

Sorts the elements in Array a based on a comparator.

• a: required. This parameter specifies an array that T in array<T> specifies the data type of the elements in the array. The elements can be of any data type.
• comparator: required. A built-in function, user-defined function, or expression that is used to compare two elements in the array.

  Processing logic of comparator(a, b): If a is equal to b, 0 is returned. If a is less than b, a negative integer is returned. If a is greater than b, a positive integer is returned. If comparator(a, b) returns NULL, an error is returned.

A value of the ARRAY type is returned.

-- [{"a":1,"b":10},{"a":2,"b":12},{"a":3,"b":11}] is returned. 
select array_sort(a, (a,b)->case when a.a> b.a then 1L when a.a=b.a then 0L else -1L end)
from values (
  array(named_struct('a', 1, 'b', 10),
        named_struct('a', 3, 'b', 11),
        named_struct('a', 2, 'b', 12)))
  as t(a);

array<T> array_union(array<T> <a>,  array<T> <b>)

Calculates the union of Array a and Array b, and removes duplicate elements.

a and b: required. These parameters specify arrays that T in array<T> specifies the data type of the elements in the arrays. The elements can be of any data type. The elements in Array a and the elements in Array b must be of the same data type.

A value of the ARRAY type is returned. If Array a or Array b is null, NULL is returned.

• Example 1: Calculate the union of array(1, 2, 3) and array(1, 3, 5), and remove duplicate elements. Sample statement:

  -- [1,2,3,5] is returned. 
  select array_union(array(1, 2, 3), array(1, 3, 5));

• Example 2: One of the arrays is null. Sample statement:

  -- NULL is returned. 
  select array_union(array(1, 2, 3), null);

boolean arrays_overlap(array<T> <a>,  array<T> <b>)

Checks whether Array a and Array b contain the same element.

a and b: required. These parameters specify arrays that T in array<T> specifies the data type of the elements in the arrays. The elements can be of any data type. The elements in Array a and the elements in Array b must be of the same data type.

A value of the BOOLEAN type is returned. The return value varies based on the following rules:

• If Array a contains at least one element that is in Array b and is not null, True is returned.
• If Array a and Array b do not contain the same element, both of the arrays are not empty, and one or both of the arrays contain a null element, NULL is returned.
• If Array a and Array b do not contain the same element, and both of the arrays are not empty and do not contain a null element, False is returned.

• Example 1: Check whether array(1, 2, 3) and array(3, 4, 5) contain the same element. Sample statement:

  -- True is returned. 
  select arrays_overlap(array(1, 2, 3), array(3, 4, 5));

• Example 2: Check whether array(1, 2, 3) and array(6, 4, 5) contain the same element. Sample statement:

  -- False is returned. 
  select arrays_overlap(array(1, 2, 3), array(6, 4, 5));

• Example 3: One of the arrays contains a null element. Sample statement:

  -- NULL is returned. 
  select arrays_overlap(array(1, 2, 3), array(5, 4, null));

array<struct<T, U, ...>> arrays_zip(array<T> <a>, array<U> <b>[, ...])

Merges multiple given arrays and returns a struct array, in which the Nth struct contains all the Nth elements of the input arrays.

a and b: required. These parameters specify arrays that T in array<T> and U in array<U> specify the data types of the elements in the arrays. The elements can be of any data type.

• Example 1: Merge array(1, 2, 3) and array(2, 3, 4) into a struct array. Sample statement:

  -- [{0:1, 1:2}, {0:2, 1:3}, {0:3, 1:4}] is returned. 
  select arrays_zip(array(1, 2, 3), array(2, 3, 4));

• Example 2: Merge array(1, 2, 3) and array(4, 5) into a struct array. Sample statement:

  -- [{0:1, 1:4}, {0:2, 1:5}, {0:3, 1:null}] is returned. 
  select arrays_zip(array(1, 2, 3), array(4, 5));

array<T> concat(array<T> <a>, array<T> <b>[,...])
string concat(string <str1>, string <str2>[,...])

• Arrays as inputs: Concatenates all elements of multiple arrays and returns a new array.
• Strings as inputs: Concatenates multiple strings and returns a new string.

• a and b: required. These parameters specify arrays that T in array<T> specifies the data type of the elements in the arrays. The elements can be of any data type. The elements in Array a and the elements in Array b must be of the same data type. The null elements are also involved in the operation.
• str1 and str2: required. A value of the STRING type. If the input value is of the BIGINT, DOUBLE, DECIMAL, or DATETIME type, it is implicitly converted into a value of the STRING type before calculation. If the input value is of another data type, an error is returned.

• A value of the ARRAY type is returned. If one of the input arrays is null, NULL is returned.
• A value of the STRING type is returned. If no input parameters are configured or an input parameter is set to null, NULL is returned.

• Example 1: Concatenate all elements of array(10, 20) and array(20, -20). Sample statement:

  -- [10, 20, 20, -20] is returned. 
  select concat(array(10, 20), array(20, -20));

• Example 2: One of the input arrays contains a null element. Sample statement:

  -- [10, null, 20, -20] is returned. 
  select concat(array(10, null), array(20, -20));

• Example 3: One of the input arrays is null. Sample statement:

  -- NULL is returned. 
  select concat(array(10, 20), null);

• Example 4: Concatenate strings aabc and abcde. Sample statement:

  -- aabcabcde is returned. 
  select concat('aabc','abcde');

• Example 5: The input is empty. Sample statement:

  -- NULL is returned. 
  select concat();

• Example 6: One of the input strings is null. Sample statement:

  -- NULL is returned. 
  select concat('aabc', 'abcde', null);

• A SELECT statement can contain only one EXPLODE function, and no other columns of a table are allowed.
• This function cannot be used with the GROUP BY, CLUSTER BY, DISTRIBUTE BY, or SORT BY clause.

explode (<var>)

var: required. The value must be of the array<T> or map<K, V> type.

Rows after transposition are returned.

The t_table_map table contains the c1 (BIGINT) and t_map (MAP<STRING,BIGINT>) fields. Data in the table:

  +------------+-------+
| c1         | t_map |
+------------+-------+
| 1000       | {k11:86, k21:15} |
| 1001       | {k12:97, k22:2} |
| 1002       | {k13:99, k23:1} |
+------------+-------+

select explode(t_map) from t_table_map;
-- The following result is returned: 
+-----+------------+
| key | value      |
+-----+------------+
| k11 | 86         |
| k21 | 15         |
| k12 | 97         |
| k22 | 2          |
| k13 | 99         |
| k23 | 1          |
+-----+------------+

array<T> filter(array<T> <a>, function<T,boolean> <func>)

Filters the elements in Array a by using func and returns a new array.

• a: required. This parameter specifies an array that T in array<T> specifies the data type of the elements in the array. The elements can be of any data type.
• func: required. This parameter specifies the built-in function, user-defined function, or expression that is used to filter the elements in Array a. The value must be of the same data type as the elements in Array a. The output result of the function or expression is of the BOOLEAN type.

A value of the ARRAY type is returned.

-- [2, 3] is returned. 
select filter(array(1, 2, 3), x -> x > 1);

index(<var1>[<var2>])

• If var1 is of the map<K, V> type, this function obtains the element that is at position var2 in var1. The position numbers of elements are counted from left to right and start from 0.
• If var1 is of the map<K, V> type, this function obtains the value whose key is var2 in var1.

• var1: required. The value must be of the array<T> or map<K, V> type. T in array<T> specifies the data type of the elements in an array. The elements can be of any data type. K and V in map<K, V> specify the keys and values of a map.
• var2: required.
  • If var1 is of the array<T> type, var2 must be of the BIGINT type and greater than or equal to 0.
  • If var1 is of the map<K, V> type, var2 must be of the same data type as K.

• If var1 is of the array<T> type, a value of the data type that is specified by T is returned. The return value varies based on the following rules:
  • If the number of elements in var1 is less than var2, NULL is returned.
  • If var1 is null, NULL is returned.
• If var1 is of the map<K, V> type, a value that is of the same data type as V is returned. The return value varies based on the following rules:
  • If map<K, V> does not contain a key whose value is var2, NULL is returned.
  • If var1 is null, NULL is returned.

• Example 1: var1 is of the array<T> type. Sample statement:

  -- c is returned. 
  select array('a','b','c')[2];

• Example 2: var1 is of the map<K, V> type. Sample statement:

  -- 1 is returned. 
  select str_to_map("test1=1,test2=2")["test1"];

posexplode(array<T> <a>)

Converts Array a into a table that has two columns. The first column lists the position of each element in the array, starting from 0. The second column lists the elements.

a: required. This parameter specifies an array that T in array<T> specifies the data type of the elements in the array. The elements can be of any data type.

A table is returned.

select posexplode(array('a','c','f','b'));
-- The following result is returned: 
+------------+------------+
| pos        | val        |
+------------+------------+
| 0          | a          |
| 1          | c          |
| 2          | f          |
| 3          | b          |
+------------+------------+

int size(array<T> <a>)
int size(map<K, V> <b> )

• An array as the input: Calculates the number of elements in Array a.
• A map as the input: Calculates the number of key-value pairs in Map b.

• a: required. This parameter specifies an array that T in array<T> specifies the data type of the elements in the array. The elements can be of any data type.
• b: required. This parameter specifies a map. K and V in map<K, V> specify the keys and values of a map.

A value of the INT type is returned.

• Example 1: Calculate the number of elements in array('a','b'). Sample statement:

  -- 2 is returned. 
  select size(array('a','b'));

• Example 2: Calculate the number of key-value pairs in map('a',123,'b',456).

  -- 2 is returned. 
  select size(map('a',123,'b',456)); 

array<T> slice(array<T> <a>, <start>, <length>)

Copies the elements in an array from the start position based on a specific length and returns the elements as a new array.

• a: required. This parameter specifies an array that T in array<T> specifies the data type of the elements in the array. The elements can be of any data type.
• start: required. This parameter specifies the position at which the function starts to copy elements from left to right. The minimum positive value of this parameter is 1. You can also set the start parameter to a negative value. In this case, the start position is counted from the end of the array, but the elements are still copied from left to right.
• length: required. The number of elements in the returned array. The value must be greater than or equal to 0. If the value is greater than the length of the input array, this function returns a new array that consists of the elements from the start position to the end of the input array.

A value of the ARRAY type is returned.

• Example 1: Copy the elements in array(10, 20, 20, null, null, 30) from position 1 based on a length of 3. Sample statement:

  -- [10, 20, 20] is returned. 
  select slice(array(10, 20, 20, null, null, 30), 1, 3);

• Example 2: Copy the elements in array(10, 20, 20, null, null, 30) from position -2 based on a length of 2. Sample statement:

  -- [null, 30] is returned. 
  select slice(array(10, 20, 20, null, null, 30), -2, 2);  

• Example 3: Copy the elements in array(10, 20, 20, null, null, 30) from position 3 based on a length of 10. Sample statement:

  -- [20, null, null, 30] is returned. 
  select slice(array(10, 20, 20, null, null, 30), 3, 10); 

• Example 4: Copy the elements in array(10, 20, 20, null, null, 30) from position 3 based on a length of 0. Sample statement:

  -- [] is returned. 
  select slice(array(10, 20, 20, null, null, 30), 3, 0);

array<T> sort_array(array<T> <a>[, <isasc>])

Sorts the elements in an array.

• a: required. This parameter specifies an array that T in array<T> specifies the data type of the elements in the array. The elements can be of any data type.
• isasc: optional. This parameter specifies whether to sort elements in the array in ascending or descending order. Valid values: true and false. The value true indicates the ascending order and false indicates the descending order. If you do not specify this parameter, the elements are sorted in ascending order.

A value of the ARRAY type is returned. NULL is interpreted as the minimum value.

• Example 1: The t_array table contains the c1 ARRAY<STRING>, c2 ARRAY<INT>, and c3 ARRAY<STRING> fields. Data in the table:

  +------------+---------+--------------+
  | c1         | c2      | c3           |
  +------------+---------+--------------+
   [a, c, f, b]   [4, 5, 7, 2, 5, 8]    [You, Me, Him] 
  +------------+---------+--------------+

  Sort data in each column of the table. Sample statement:

  [a, b, c, f] [2, 4, 5, 5, 7, 8] [Him, You, Me] is returned. 
  select sort_array(c1),sort_array(c2),sort_array(c3) from t_array;

• Example 2: Sort the elements in array(10, 20, 40, 30, 30, null, 50) in descending order. Sample statement:

  -- [50, 40, 30, 30, 20, 10, null] is returned. 
  select sort_array(array(10, 20, 40, 30, 30, null, 50), false);

array<R> transform(array<T> <a>, function<T, R> <func>)

Transforms the elements in Array a by using func and returns a new array.

• a: required. This parameter specifies an array that T in array<T> specifies the data type of the elements in the array. The elements can be of any data type.
• func: required. This parameter specifies the built-in function, user-defined function, or expression that is used to transform the elements in Array a. The value must be of the same data type as the elements in Array a. R specifies the data type of the output results.

A value of the ARRAY type is returned.

-- [2, 3, 4] is returned. 
select transform(array(1, 2, 3), x -> x + 1);

array<R> array_sort(array<T> <a>, array<S> <b>, function<T, S, R> <combiner>)

Merges Array a and Array b at the element level based on element positions and a combiner and returns a new array.

• a and b: required. These parameters specify arrays that T in array<T> and S in array<S> specify the data types of the elements in the arrays. The elements can be of any data type.
• combiner: required. This parameter specifies the built-in function, user-defined function, or expression that is used to merge Array a and Array b at the element level. The combiner consists of two input parameters. One input parameter must be of the same data type as the elements in Array a. The other input parameter must be of the same data type as the elements in Array b.

-- [2, 4, 6, null] is returned. 
select zip_with(array(1,2,3), array(1,2,3,4), (x,y) -> x + y);

map(K, V) map(K <key1>, V <value1>, K <key2>, V <value2>[, ...])

Creates a map based on given key-value pairs.

• key: required. All keys must be of the same data type after implicit conversions. Only basic data types are supported.
• value: required. All values must be of the same data type after implicit conversions. All data types are supported.

• Example 1: No duplicate keys exist. The t_table table contains the c1 (BIGINT), c2 (STRING), c3 (STRING), c4 (BIGINT), and c5 (BIGINT) fields. Data in the table:

  +------------+----+----+------------+------------+
  | c1         | c2 | c3 | c4         | c5         |
  +------------+----+----+------------+------------+
  | 1000       | k11 | k21 | 86         | 15         |
  | 1001       | k12 | k22 | 97         | 2          |
  | 1002       | k13 | k23 | 99         | 1          |
  +------------+----+----+------------+------------+

  Sample statement:

  -- Define a map based on the key-value pairs between the c2 and c4 columns, and between the c3 and c5 columns. 
  select map(c2,c4,c3,c5) from t_table;
  -- The following result is returned: 
  +------+
  | _c0  |
  +------+
  | {k11:86, k21:15} |
  | {k12:97, k22:2} |
  | {k13:99, k23:1} |
  +------+

• Example 2: Duplicate keys exist. The t_table table contains the c1 (BIGINT), c2 (STRING), c3 (STRING), c4 (BIGINT), and c5 (BIGINT) fields. Data in the table:

  1000,'k11','k11',86,15
  1001,'k12','k22',97,2
  1002,'k13','k23',99,1
  1003,'k13','k24',100,1
  1004,'k12','k25',95,1

  Sample statement:

  -- Define a map based on the key-value pairs between the c2 and c4 columns, and between the c3 and c5 columns. 
  select map(c2,c4,c3,c5) from t_table;
  -- The following result is returned: 
  +------+
  | _c0  |
  +------+
  | {'k11':15} |
  | {'k12':97, 'k22':2} |
  | {'k13':99, 'k23':1} |
  | {'k13':100, 'k24':1} |
  | {'k12':95, 'k25':1} |
  +------+

map<K, V> map_concat([string <mapDupKeyPolicy>,] map<K, V> <a>, map<K, V> <b>[,...])

Calculates the union of multiple maps.

• mapDupKeyPolicy: optional. A value of the STRING type. This parameter specifies the method that is used to process duplicate keys. Valid values:
  • exception: An error is returned.
  • last_win: The latter key overwrites the former key.

  You can also specify the odps.sql.map.key.dedup.policy parameter at the session level to configure the method that is used to process duplicate keys. For example, you can set odps.sql.map.key.dedup.policy to exception. If you do not specify this parameter, the default value last_win is used.

  Note The behavior implementation of MaxCompute is determined based on mapDupKeyPolicy. If you do not specify mapDupKeyPolicy, the value of odps.sql.map.key.dedup.policy prevails.
• a and b: required. These parameters specify maps. The keys of the maps must be of the same data type, and the values of the maps must be of the same data type. K and V in map<K, V> specify the keys and values of a map.

-- {1:a, 2:b, 3:c} is returned. 
select map_concat(map(1, 'a', 2, 'b'), map(3, 'c'));
-- {1:a, 2:d, 3:c} is returned. 
select map_concat('last_win', map(1, 'a', 2, 'b'), map(3, 'c'), map(2, 'd'));

array<struct<K, V>> map_entries(map<K, V> <a>): 

Converts key-value pairs in Map a into a struct array.

a: required. This parameter specifies a map. K and V in map<K, V> specify the keys and values of a map.

A struct array is returned. If the input is null, NULL is returned.

-- [{key:1, value:a}, {key:2, value:b}] is returned. 
select map_entries(map(1,  'a',  2,  'b'));

map<K, V> map_filter(map<K, V> <input>, function <K, V, boolean> <predicate>)

Filters the elements in Map input and retains only the elements that meet the predicate condition.

• input: required. A value of the MAP type. K and V in map<K, V> specify the keys and values of a map.
• predicate: required. This parameter specifies the built-in function, user-defined function, or expression that is used to filter the elements in the map. The predicate condition consists of two input parameters that correspond to the keys and values in input. The output result is of the BOOLEAN type.

Data of the MAP type is returned.

-- {-30:100, 20:50} is returned. 
select map_filter(map(10, -20, 20, 50, -30, 100, 21, null), (k, v) -> (k+v) > 10);

map<K, V> map_from_arrays([string <mapDupKeyPolicy>,] array<K> <a>, array<V> <b>))

Creates a map based on Array a and Array b.

• mapDupKeyPolicy: optional. A value of the STRING type. This parameter specifies the method that is used to process duplicate keys. Valid values:
  • exception: An error is returned.
  • last_win: The latter key overwrites the former key.

  You can also specify the odps.sql.map.key.dedup.policy parameter at the session level to configure the method that is used to process duplicate keys. For example, you can set odps.sql.map.key.dedup.policy to exception. If you do not specify this parameter, the default value last_win is used.

  Note The behavior implementation of MaxCompute is determined based on mapDupKeyPolicy. If you do not specify mapDupKeyPolicy, the value of odps.sql.map.key.dedup.policy prevails.
• a: required. This parameter specifies an array. This parameter corresponds to the key in the generated map. K in array<K> specifies the data type of the elements in the array. The elements can be of any data type.
• b: required. This parameter specifies an array. This parameter corresponds to the value in the generated map. V in array<V> specifies the data type of the elements in the array. The elements can be of any data type.

-- {1:2, 3:4} is returned. 
select map_from_arrays(array(1.0, 3.0), array('2', '4'));
-- {1:2, 3:6} is returned. 
select map_from_arrays('last_win', array(1.0, 3.0, 3), array('2', '4', '6'));

map<K, V> map_from_entries([string <mapDupKeyPolicy>,] array <struct<K, V> , struct<K, V>[,...]>)

Creates a map based on given struct arrays.

• mapDupKeyPolicy: optional. A value of the STRING type. This parameter specifies the method that is used to process duplicate keys. Valid values:
  • exception: An error is returned.
  • last_win: The latter key overwrites the former key.

  You can also specify the odps.sql.map.key.dedup.policy parameter at the session level to configure the method that is used to process duplicate keys. For example, you can set odps.sql.map.key.dedup.policy to exception. If you do not specify this parameter, the default value last_win is used.

  Note The behavior implementation of MaxCompute is determined based on mapDupKeyPolicy. If you do not specify mapDupKeyPolicy, the value of odps.sql.map.key.dedup.policy prevails.
• Data of the STRUCT type is required. K corresponds to the keys in the generated map. V corresponds to the values in the generated map. K and V in struct<K, V> specify the keys and values of a struct array.

-- {1:a, 2:b} is returned. 
select map_from_entries(array(struct(1, 'a'), struct(2, 'b')));
-- {1:a, 2:c} is returned. 
select map_from_entries(array(struct(1, 'a'), struct(2, 'b'), struct(2, 'c')));

array<K> map_keys(map<K, V> <a>)

Returns all keys in Map a as an array.

a: required. This parameter specifies a map. K and V in map<K, V> specify the keys and values of a map.

A value of the ARRAY type is returned. If the input map is null, NULL is returned.

array<V> map_values(map<K, V> <a>)

Returns all values in Map a as an array.

a: required. This parameter specifies a map. K and V in map<K, V> specify the keys and values of a map.

A value of the ARRAY type is returned. If the input map is null, NULL is returned.

<K, V1, V2, V3> map<K, V3> map_zip_wuth(map<K, V1> <input1>, map<K, V2> <input2>, function<K, V1, V2, V3> <func>)

Merges Map input1 and Map input2 into a single map. The keys of the new map are the union of the keys of the two input maps. The value of each key of the new map is calculated by using func.

• input1 and input2: required. These parameters specify maps. K and V in map<K, V> specify the keys and values of a map.
• func: required. func consists of three input parameters, which correspond to a key, the value that corresponds to the key in input1, and the value that corresponds to the key in input2. If the key does not exist in input1 or input2, null is used to replace the value that corresponds to the key in func.

Data of the data type defined by func is returned.

-- {1:[1, 1, 4], 2:[2, 2, 5], 3:[3, null, null], 4:[4, null, 7]} is returned. 
select map_zip_with(map(1, 1, 2, 2, 3, null), map(1, 4, 2, 5, 4, 7), (k, v1, v2) -> array(k, v1, v2));

map<K2, V> transform_keys([string <mapDupKeyPolicy>,] map<K1, V> <input>, function<K1, V, K2> <func>)

Transforms the keys in Map input by using func. The values in the map are not changed.

• mapDupKeyPolicy: optional. A value of the STRING type. This parameter specifies the method that is used to process duplicate keys. Valid values:
  • exception: An error is returned.
  • last_win: The latter key overwrites the former key.

  You can also specify the odps.sql.map.key.dedup.policy parameter at the session level to configure the method that is used to process duplicate keys. For example, you can set odps.sql.map.key.dedup.policy to exception. If you do not specify this parameter, the default value last_win is used.

  Note The behavior implementation of MaxCompute is determined based on mapDupKeyPolicy. If you do not specify mapDupKeyPolicy, the value of odps.sql.map.key.dedup.policy prevails.
• input: required. This parameter specifies a map. K1 and V in map<K1, V> specify the keys and values of a map.
• func: required. This parameter specifies the built-in function, user-defined function, or expression that is used to transform the keys. The function or expression consists of two input parameters that correspond to the keys and values in input. K2 specifies the data type of keys in the returned map.

Data of the MAP type is returned. If one of the new keys is null, an error is returned.

-- {-10:-20, 70:50, 71:101} is returned. 
select transform_keys(map(10, -20, 20, 50, -30, 101), (k, v) -> k + v);
-- No error is returned. The returned result depends on the order of the elements in the input map. 
select transform_keys("last_win", map(10, -20, 20, 50, -30, 100), (k, v) -> k + v);
-- An error is returned because duplicate keys exist. 
select transform_keys("exception", map(10, -20, 20, 50, -30, 100), (k, v) -> k + v);

map<K, V2> transform_values(map<K, V1> <input>, function<K, V1, V2> <func>)

Transforms the values in Map input by using func. The keys in the map are not changed.

• input: required. This parameter specifies a map. K and V1 in map<K, V1> specify the keys and values of the map.
• func: required. This parameter specifies the built-in function, user-defined function, or expression that is used to transform the keys. The function or expression consists of two input parameters that correspond to the keys and values in input. V2 specifies the data type of values in the returned map.

Data of the MAP type is returned.

-- {-30:71, 10:-10, 20:null} is returned. 
select transform_values(map(10, -20, 20, null, -30, 101), (k, v) -> k + v);

T field(struct <s>, string <fieldName>)

Obtains the value of a member variable in a struct.

• s: required. This parameter specifies a struct. The struct is in the format of {f1:T1, f2:T2[, ...]}. f1 and f2 specify member variables, T1 specifies the value of f1, and T2 specifies the value of f2.
• fieldName: required. A value of the STRING type. The value must be one of the member variables in the struct.

The value of the specific member variable in the struct is returned.

-- hello is returned. 
select field(named_struct('f1', 'hello', 'f2', 3), 'f1');

inline(array<struct<f1:T1, f2:T2[, ...]>>)

Expands a given struct array. Each array element corresponds to a row and each struct element corresponds to a column in each row.

f1:T1 and f2:T2: required. A value of any data type. f1 and f2 specify member variables, T1 specifies the value of f1, and T2 specifies the value of f2.

The expanded data of the struct array is returned.

struct struct(<value1>,<value2>[, ...])

Creates a struct based on a given value list.

value: required. Data in the array can be of any data type.

A value of the STRUCT type is returned. Column names are sequentially named as col1, col2, ...

-- {col1:a, col2:123, col3:true, col4:56.9} is returned. 
select struct('a',123,'true',56.90);

struct named_struct(string <name1>, T1 <value1>, string <name2>, T2 <value2>[, ...])

Creates a struct based on given name-value pairs.

• value: required. All data types are supported.
• name: required. The column name of the STRING type. This parameter is a constant.

A value of the STRUCT type is returned. Column names are sequentially named as name1, name2, ...

-- {user_id:10001, user_name:LiLei, married:F, weight:63.5} is returned. 
select named_struct('user_id',10001,'user_name','LiLei','married','F','weight',63.50);

from_json(<jsonStr>, <schema>)

Returns data of the ARRAY, MAP, or STRUCT type based on JSON string jsonStr and output format schema.

• jsonStr: required. The JSON string that you entered.
• schema: required. The schema of the JSON string. The value of this parameter must be in the same format as that in the statement for creating a table, such as array<bigint>, map<string, array<string>>, or struct<a:int, b:double, `C`:map<string,string>>.
  Note Keys in a struct are case-sensitive. You can also specify a struct in the format of a BIGINT, b DOUBLE, which is equivalent to STRUCT<a:BIGINT, b:DOUBLE>.

  The following table describes the mappings between JSON data types and MaxCompute data types.

  JSON data type	MaxCompute data type
  OBJECT	STRUCT, MAP, and STRING
  ARRAY	ARRAY and STRING
  NUMBER	TINYINT, SMALLINT, INT, BIGINT, FLOAT, DOUBLE, DECIMAL, and STRING
  BOOLEAN	BOOLEAN and STRING
  STRING	STRING, CHAR, VARCHAR, BINARY, DATE, and DATETIME
  NULL	All types

  Note The JSON string of the OBJECT and ARRAY types are parsed as much as possible. If the data type of the JSON string is not mapped to any MaxCompute data type, the JSON string is omitted. For ease of use, all JSON data types can be converted into the STRING data type supported by MaxCompute. When you convert a JSON string of the NUMBER type to a value of the FLOAT, DOUBLE, or DECIMAL type, the precision of the value cannot be ensured. We recommend you convert the JSON string to a value of the STRING type and then convert the obtained value to a value of the FLOAT, DOUBLE, or DECIMAL type.

A value of the ARRAY, MAP, or STRUCT type is returned.

• Example 1: Convert a specific JSON string into a value of a specific data type. Sample statements:

  -- {"a":1,"b":0.8} is returned. 
  select from_json('{"a":1, "b":0.8}', 'a int, b double');
  -- {"time":"26/08/2015"} is returned. 
  select from_json('{"time":"26/08/2015"}', 'time string');
  -- {"a":1,"b":0.8} is returned. 
  select from_json('{"a":1, "b":0.8}', 'a int, b double, c string');
  -- [1,2,3] is returned. 
  select from_json('[1, 2, 3, "a"]', 'array<bigint>');
  -- {"d":"v","a":"1","b":"[1,2,3]","c":"{}"} is returned. 
  select from_json('{"a":1,"b":[1,2,3],"c":{},"d":"v"}', 'map<string, string>');

• Example 2: Use the map_keys and from_json functions to obtain all keys in the JSON string. You can also use JSON_KEYS for the same purpose. Sample statement:

  -- ["a","b"] is returned. 
  select map_keys(from_json('{"a":1,"b":2}','map<string,string>'));

string get_json_object(string <json>, string <path>)

Extracts a single string from a standard JSON string based on path. The original data is read each time this function is called. Therefore, repeated calls may affect system performance and increase costs. To prevent repeated calls, you can use the GET_JSON_OBJECT function with UDTFs. For more information, see Convert JSON log data by using MaxCompute built-in functions and UDTFs.

• json: required. A value of the STRING type. This parameter specifies a standard JSON object in the format of {Key:Value, Key:Value,...}. If the string contains a double quotation mark ("), use two backslashes (\\) to escape the double quotation mark (") before extraction. If the string contains a single quotation mark ('), use a single backslash (\) to escape the single quotation mark (') before extraction.
• path: required. A value of the STRING type. This parameter specifies the path in the value of the json parameter and starts with $. For more information about the path parameter, see LanguageManual UDF. For more information about best practices, see Migrate JSON data from OSS to MaxCompute. Different characters have the following meanings:
  • $: indicates the root node.
  • . or ['']: indicates a child node. MaxCompute parses JSON objects by using . or ['']. If a key in a JSON object contains a period (.), [''] can be used.
  • [] ([number]): indicates an array subscript, which starts from 0.
  • *: indicates the wildcard for [], which returns an entire array. An asterisk (*) cannot be escaped.

• If the json parameter is empty or invalid, NULL is returned.
• If the format of json is valid and path exists, the related string is returned.
• To determine how this function returns a value, you can specify odps.sql.udf.getjsonobj.new for a session.
  • If you run the set odps.sql.udf.getjsonobj.new=true; command, this function retains the original strings when it returns a value.
    We recommend that you use this configuration because it results in more standard function return behavior. This facilitates data processing and improves data processing performance. If a MaxCompute project has jobs in which this function escapes JSON reserved characters, we recommend that you retain the original escaping operation to prevent errors caused by lack of verification. The function complies with the following rules when it returns a value:

    • The return value is still a JSON string, which can be parsed as JSON data. You do not need to use the REPLACE or REGEXP_REPLACE function to replace backslashes (\).
    • Duplicate keys are allowed in a JSON object. If duplicate keys exist, the data can be parsed.

      -- 1 is returned. 
      select get_json_object('{"a":"1","a":"2"}', '$.a');

    • The encoded strings that correspond to emojis are supported. However, DataWorks does not allow you to enter emojis. DataWorks allows you to enter only the encoded strings that correspond to emojis to MaxCompute by using a tool, such as Data Integration. DataWorks uses the GET_JSON_OBJECT function to process the data.

      -- An emoji is returned. 
      select get_json_object('{"a":"<Emoji>"}', '$.a');

    • The output results are displayed in alphabetical order.

      -- {"b":"1","a":"2"} is returned. 
      select get_json_object('{"b":"1","a":"2"}', '$');

  • If you run the set odps.sql.udf.getjsonobj.new=false; command, this function escapes JSON reserved characters when it returns a value. The function complies with the following rules when it returns a value:
    • JSON reserved characters, such as line feeds (\n) and double quotation marks (") are displayed as '\n' and '\"'.
    • Each key in a JSON object must be unique. If duplicate keys exist, the data may fail to be parsed.

      -- NULL is returned. 
      select get_json_object('{"a":"1","a":"2"}', '$.a');

    • The encoded strings that correspond to emojis cannot be parsed.

      -- NULL is returned. 
      select get_json_object('{"a":"<Emoji>"}', '$.a');

    • The output results are displayed in alphabetical order.

      -- {"a":"2","b":"1"} is returned. 
      select get_json_object('{"b":"1","a":"2"}', '$');

  Note For MaxCompute projects that are created on or after January 21, 2021, the GET_JSON_OBJECT function retains the original strings when it returns a value. For MaxCompute projects that are created before January 21, 2021, the GET_JSON_OBJECT function escapes JSON reserved characters when it returns a value. The following example helps you determine how the GET_JSON_OBJECT function returns a value in a MaxCompute project.

  select get_json_object('{"a":"[\\"1\\"]"}', '$.a');
  -- Return JSON reserved characters by using escape characters.
  [\"1\"]
  
  -- Return the original strings.
  ["1"]

  You can submit a ticket to request the MaxCompute technical support team to enable the GET_JSON_OBJECT function. This way, you do not need to frequently specify flagodps.sql.udf.getjsonobj.new for a session. This function retains the original strings when it returns a value.

• Example 1: Extract information from the JSON object src_json.json. Sample statement:

  -- The JSON string src_json.json contains the following content: 
  +----+
  json
  +----+
  {"store":
  {"fruit":[{"weight":8,"type":"apple"},{"weight":9,"type":"pear"}],
  "bicycle":{"price":19.95,"color":"red"}
  },
  "email":"amy@only_for_json_udf_test.net",
  "owner":"amy"
  }
  -- Extract the information of the owner field and return amy. 
  select get_json_object(src_json.json, '$.owner') from src_json;
  -- Extract the information of the first array in the store.fruit field and return {"weight":8,"type":"apple"}. 
  select get_json_object(src_json.json, '$.store.fruit[0]') from src_json;
  -- Extract the information of the non-existent field and return NULL. 
  select get_json_object(src_json.json, '$.non_exist_key') from src_json;

• Example 2: Extract information from a JSON object of the ARRAY type. Sample statement:

  -- 2222 is returned. 
  select get_json_object('{"array":[["aaaa",1111],["bbbb",2222],["cccc",3333]]}','$.array[1][1]');
  -- ["h0","h1","h2"] is returned. 
  set odps.sql.udf.getjsonobj.new=true;
  select get_json_object('{"aaa":"bbb","ccc":{"ddd":"eee","fff":"ggg","hhh":["h0","h1","h2"]},"iii":"jjj"}','$.ccc.hhh[*]');
  -- ["h0","h1","h2"] is returned. 
  set odps.sql.udf.getjsonobj.new=false;
  select get_json_object('{"aaa":"bbb","ccc":{"ddd":"eee","fff":"ggg","hhh":["h0","h1","h2"]},"iii":"jjj"}','$.ccc.hhh');
  -- h1 is returned. 
  select get_json_object('{"aaa":"bbb","ccc":{"ddd":"eee","fff":"ggg","hhh":["h0","h1","h2"]},"iii":"jjj"}','$.ccc.hhh[1]');

• Example 3: Extract information from a JSON object that contains a period (.). Sample statement:

  -- Create a table. 
  create table mf_json (id string, json string);
  -- Insert data into the table. The key in the data contains a period (.). 
  insert into table mf_json (id, json) values ("1", "{
  \"China.beijing\":{\"school\":{\"id\":0,\"book\":[{\"title\": \"A\",
  \"price\": 8.95},{\"title\": \"B\",\"price\": 10.2}]}}}");
  -- Insert data into the table. The key in the data does not contain a period (.). 
  insert into table mf_json (id, json) values ("2", "{
  \"China_beijing\":{\"school\":{\"id\":0,\"book\":[{\"title\": \"A\",
  \"price\": 8.95},{\"title\": \"B\",\"price\": 10.2}]}}}");
  -- Query the value of id in the JSON object whose key is China.beijing. 0 is returned. Only [''] can be used to specify the key because the key contains a period (.). This way, MaxCompute can parse the key. 
  select get_json_object(json, "$['China.beijing'].school['id']") from mf_json where id =1;
  -- -- Query the value of id in the JSON object whose key is China_beijing. 0 is returned. You can use one of the following statements: 
  select get_json_object(json, "$['China_beijing'].school['id']") from mf_json where id =2;
  select get_json_object(json, "$.China_beijing.school['id']") from mf_json where id =2;

• Example 4: The json parameter is empty or invalid. Sample statement:

  -- NULL is returned. 
  select get_json_object('','$.array[1][1]');
  -- NULL is returned. 
  select get_json_object('"array":["aaaa",1111],"bbbb":["cccc",3333]','$.array[1][1]');

• Example 5: Escape a JSON string. Sample statement:

  set odps.sql.udf.getjsonobj.new=true;
  -- "1" is returned. 
  select get_json_object('{"a":"\\"1\\"","b":"2"}', '$.a'); 
  -- '1' is returned. 
  select get_json_object('{"a":"\'1\'","b":"2"}', '$.a'); 

string json_tuple(string <json>, string <key1>, string <key2>,...)

Extracts strings from a standard JSON string based on a set of input keys, such as (key1,key2,...).

• json: required. A value of the STRING type. This parameter specifies a standard JSON string.
• key: required. A value of the STRING type. This parameter is used to describe the path of a JSON object in the JSON string. The value cannot start with a dollar sign ($). You can enter multiple keys at a time. MaxCompute parses JSON objects by using . or ['']. If a key in a JSON object contains a period (.), [''] can be used.

A value of the STRING type is returned.

to_json(<expr>)

Converts an expression of a given complex data type into a JSON string.

• Example 1: Convert an expression of a given data type into a JSON string. Sample statement:

  -- {"a":1,"b":2} is returned. 
  select to_json(named_struct('a', 1, 'b', 2));
  -- {"time":"26/08/2015"} is returned. 
  select to_json(named_struct('time', "26/08/2015"));
  -- [{"a":1,"b":2}] is returned. 
  select to_json(array(named_struct('a', 1, 'b', 2)));
  -- {"a":{"b":1}} is returned. 
  select to_json(map('a', named_struct('b', 1)));
  -- {"a":1} is returned. 
  select to_json(map('a', 1));
  -- [{"a":1}] is returned. 
  select to_json(array((map('a', 1))));

• Example 2: The input expression is of the STRUCT type. Sample statement:

  -- {"a":"B"} is returned. If the expression of the STRUCT type is converted into a JSON string, all keys are converted into lowercase letters. 
  select to_json(named_struct("A", "B"));
  -- {"k2":"v2"} is returned. The key-value pair to which null belongs is not included in the JSON string that is returned. 
  select to_json(named_struct("k1", cast(null as string), "k2", "v2"));