Operators

GoogleSQL for Bigtable supports operators. Operators are represented by special characters or keywords; they do not use function call syntax. An operator manipulates any number of data inputs, also called operands, and returns a result.

Common conventions:

• Unless otherwise specified, all operators return NULL when one of the operands is NULL.
• All operators will throw an error if the computation result overflows.
• For all floating point operations, +/-inf and NaN may only be returned if one of the operands is +/-inf or NaN. In other cases, an error is returned.

Operator precedence

The following table lists all GoogleSQL operators from highest to lowest precedence, i.e., the order in which they will be evaluated within a statement.

Operators with the same precedence are left associative. This means that those operators are grouped together starting from the left and moving right. For example, the expression:

x AND y AND z

is interpreted as

( ( x AND y ) AND z )

The expression:

x * y / z

is interpreted as:

( ( x * y ) / z )

All comparison operators have the same priority, but comparison operators are not associative. Therefore, parentheses are required in order to resolve ambiguity. For example:

(x < y) IS FALSE

Operator list

Field access operator

expression.fieldname[. ...]

Description

Gets the value of a field. Alternatively known as the dot operator. Can be used to access nested fields. For example, expression.fieldname1.fieldname2.

Input values:

• STRUCT

Return type

• For STRUCT: SQL data type of fieldname. If a field is not found in the struct, an error is thrown.

Example

In the following example, the field access operations are .address and .country.

SELECT
  STRUCT(
    STRUCT('Yonge Street' AS street, 'Canada' AS country)
      AS address).address.country

/*---------*
 | country |
 +---------+
 | Canada  |
 *---------*/

Array subscript operator

array_expression "[" array_subscript_specifier "]"

array_subscript_specifier:
  { index | position_keyword(index) }

position_keyword:
  { OFFSET | SAFE_OFFSET | ORDINAL | SAFE_ORDINAL }

Description

Gets a value from an array at a specific position.

Input values:

• array_expression: The input array.
• position_keyword(index): Determines where the index for the array should start and how out-of-range indexes are handled. The index is an integer that represents a specific position in the array.
  • OFFSET(index): The index starts at zero. Produces an error if the index is out of range. To produce NULL instead of an error, use SAFE_OFFSET(index). This position keyword produces the same result as index by itself.
  • SAFE_OFFSET(index): The index starts at zero. Returns NULL if the index is out of range.
  • ORDINAL(index): The index starts at one. Produces an error if the index is out of range. To produce NULL instead of an error, use SAFE_ORDINAL(index).
  • SAFE_ORDINAL(index): The index starts at one. Returns NULL if the index is out of range.
• index: An integer that represents a specific position in the array. If used by itself without a position keyword, the index starts at zero and produces an error if the index is out of range. To produce NULL instead of an error, use the SAFE_OFFSET(index) or SAFE_ORDINAL(index) position keyword.

Return type

T where array_expression is ARRAY<T>.

Examples

In following query, the array subscript operator is used to return values at specific position in item_array. This query also shows what happens when you reference an index (6) in an array that is out of range. If the SAFE prefix is included, NULL is returned, otherwise an error is produced.

SELECT
  ["coffee", "tea", "milk"] AS item_array,
  ["coffee", "tea", "milk"][0] AS item_index,
  ["coffee", "tea", "milk"][OFFSET(0)] AS item_offset,
  ["coffee", "tea", "milk"][ORDINAL(1)] AS item_ordinal,
  ["coffee", "tea", "milk"][SAFE_OFFSET(6)] AS item_safe_offset

/*---------------------+------------+-------------+--------------+------------------*
 | item_array          | item_index | item_offset | item_ordinal | item_safe_offset |
 +---------------------+------------+-------------+--------------+------------------+
 | [coffee, tea, milk] | coffee     | coffee      | coffee       | NULL             |
 *----------------------------------+-------------+--------------+------------------*/

When you reference an index that is out of range in an array, and a positional keyword that begins with SAFE is not included, an error is produced. For example:

-- Error. Array index 6 is out of bounds.
SELECT ["coffee", "tea", "milk"][6] AS item_offset

-- Error. Array index 6 is out of bounds.
SELECT ["coffee", "tea", "milk"][OFFSET(6)] AS item_offset

The following queries contain an array subscript operator that is applied to a column family called cell_plan in a table called [test_table][test-table]:

SELECT MAP_KEYS(cell_plan)[0] AS results FROM test_table LIMIT 1

/*----------------*
 | results        |
 +----------------+
 | data_plan_01gb |
 *----------------*/

SELECT MAP_KEYS(cell_plan)[1] AS results FROM test_table LIMIT 1

/*----------------*
 | results        |
 +----------------+
 | data_plan_05gb |
 *----------------*/

SELECT MAP_KEYS(cell_plan)[OFFSET(1)] AS results FROM test_table LIMIT 1

/*----------------*
 | results        |
 +----------------+
 | data_plan_05gb |
 *----------------*/

SELECT MAP_KEYS(cell_plan)[ORDINAL(0)] AS results FROM test_table LIMIT 1

/*----------------*
 | results        |
 +----------------+
 | data_plan_01gb |
 *----------------*/

SELECT MAP_KEYS(cell_plan)[SAFE_OFFSET(5)] AS results FROM test_table LIMIT 1

/*----------------*
 | results        |
 +----------------+
 | NULL           |
 *----------------*/

Arithmetic operators

All arithmetic operators accept input of numeric type T, and the result type has type T unless otherwise indicated in the description below:

NOTE: Divide by zero operations return an error. To return a different result, consider the IEEE_DIVIDE or SAFE_DIVIDE functions.

Result types for Addition, Subtraction and Multiplication:

Result types for Division:

Result types for Unary Plus:

Result types for Unary Minus:

Date arithmetics operators

Operators '+' and '-' can be used for arithmetic operations on dates.

date_expression + int64_expression
int64_expression + date_expression
date_expression - int64_expression

Description

Adds or subtracts int64_expression days to or from date_expression. This is equivalent to DATE_ADD or DATE_SUB functions, when interval is expressed in days.

Return Data Type

DATE

Example

SELECT DATE "2020-09-22" + 1 AS day_later, DATE "2020-09-22" - 7 AS week_ago

/*------------+------------*
 | day_later  | week_ago   |
 +------------+------------+
 | 2020-09-23 | 2020-09-15 |
 *------------+------------*/

Bitwise operators

All bitwise operators return the same type and the same length as the first operand.

Logical operators

GoogleSQL supports the AND, OR, and NOT logical operators. Logical operators allow only BOOL or NULL input and use three-valued logic to produce a result. The result can be TRUE, FALSE, or NULL:

Examples

The examples in this section reference a table called entry_table:

/*-------*
 | entry |
 +-------+
 | a     |
 | b     |
 | c     |
 | NULL  |
 *-------*/

SELECT 'a' FROM entry_table WHERE entry = 'a'

-- a => 'a' = 'a' => TRUE
-- b => 'b' = 'a' => FALSE
-- NULL => NULL = 'a' => NULL

/*-------*
 | entry |
 +-------+
 | a     |
 *-------*/

SELECT entry FROM entry_table WHERE NOT (entry = 'a')

-- a => NOT('a' = 'a') => NOT(TRUE) => FALSE
-- b => NOT('b' = 'a') => NOT(FALSE) => TRUE
-- NULL => NOT(NULL = 'a') => NOT(NULL) => NULL

/*-------*
 | entry |
 +-------+
 | b     |
 | c     |
 *-------*/

SELECT entry FROM entry_table WHERE entry IS NULL

-- a => 'a' IS NULL => FALSE
-- b => 'b' IS NULL => FALSE
-- NULL => NULL IS NULL => TRUE

/*-------*
 | entry |
 +-------+
 | NULL  |
 *-------*/

Comparison operators

Compares operands and produces the results of the comparison as a BOOL value. These comparison operators are available:

The following rules apply to operands in a comparison operator:

• The operands must be comparable.
• A comparison operator generally requires both operands to be of the same type.
• If the operands are of different types, and the values of those types can be converted to a common type without loss of precision, they are generally coerced to that common type for the comparison.
• A literal operand is generally coerced to the same data type of a non-literal operand that is part of the comparison.
• Struct operands support only these comparison operators: equal (=), not equal (!= and <>), and IN.

The following rules apply when comparing these data types:

• Floating point: All comparisons with NaN return FALSE, except for != and <>, which return TRUE.
• BOOL: FALSE is less than TRUE.
• STRING: Strings are compared codepoint-by-codepoint, which means that canonically equivalent strings are only guaranteed to compare as equal if they have been normalized first.
• NULL: Any operation with a NULL input returns NULL.

• STRUCT: When testing a struct for equality, it's possible that one or more fields are NULL. In such cases:

  • If all non-NULL field values are equal, the comparison returns NULL.
  • If any non-NULL field values are not equal, the comparison returns FALSE.

  The following table demonstrates how STRUCT data types are compared when they have fields that are NULL valued.

  Struct1	Struct2	Struct1 = Struct2
  STRUCT(1, NULL)	STRUCT(1, NULL)	NULL
  STRUCT(1, NULL)	STRUCT(2, NULL)	FALSE
  STRUCT(1,2)	STRUCT(1, NULL)	NULL

IS operators

IS operators return TRUE or FALSE for the condition they are testing. They never return NULL, even for NULL inputs, unlike the IS_INF and IS_NAN functions defined in Mathematical Functions. If NOT is present, the output BOOL value is inverted.

LIKE operator

expression_1 [NOT] LIKE expression_2

Description

LIKE returns TRUE if the string in the first operand expression_1 matches a pattern specified by the second operand expression_2, otherwise returns FALSE.

NOT LIKE returns TRUE if the string in the first operand expression_1 does not match a pattern specified by the second operand expression_2, otherwise returns FALSE.

Expressions can contain these characters:

• A percent sign (%) matches any number of characters or bytes.
• An underscore (_) matches a single character or byte.
• You can escape \, _, or % using two backslashes. For example, \\%. If you are using raw strings, only a single backslash is required. For example, r'\%'.

This operator supports collation, but caveats apply:

• Each % character in expression_2 represents an arbitrary string specifier. An arbitrary string specifier can represent any sequence of 0 or more characters.

• A character in the expression represents itself and is considered a single character specifier unless:

  • The character is a percent sign (%).

  • The character is an underscore (_) and the collator is not und:ci.

• These additional rules apply to the underscore (_) character:

  • If the collator is not und:ci, an error is produced when an underscore is not escaped in expression_2.

  • If the collator is not und:ci, the underscore is not allowed when the operands have collation specified.

  • Some compatibility composites, such as the fi-ligature (fi) and the telephone sign (℡), will produce a match if they are compared to an underscore.

  • A single underscore matches the idea of what a character is, based on an approximation known as a grapheme cluster.

• For a contiguous sequence of single character specifiers, equality depends on the collator and its language tags and tailoring.

  • By default, the und:ci collator does not fully normalize a string. Some canonically equivalent strings are considered unequal for both the = and LIKE operators.

  • The LIKE operator with collation has the same behavior as the = operator when there are no wildcards in the strings.

  • Character sequences with secondary or higher-weighted differences are considered unequal. This includes accent differences and some special cases.

    For example there are three ways to produce German sharp ß:

    • \u1E9E
    • \U00DF
    • ss

    \u1E9E and \U00DF are considered equal but differ in tertiary. They are considered equal with und:ci collation but different from ss, which has secondary differences.

  • Character sequences with tertiary or lower-weighted differences are considered equal. This includes case differences and kana subtype differences, which are considered equal.

• There are ignorable characters defined in Unicode. Ignorable characters are ignored in the pattern matching.

Return type

BOOL

Examples

The following examples illustrate how you can check to see if the string in the first operand matches a pattern specified by the second operand.

-- Returns TRUE
SELECT 'apple' LIKE 'a%';

-- Returns FALSE
SELECT '%a' LIKE 'apple';

-- Returns FALSE
SELECT 'apple' NOT LIKE 'a%';

-- Returns TRUE
SELECT '%a' NOT LIKE 'apple';

-- Produces an error
SELECT NULL LIKE 'a%';

-- Produces an error
SELECT 'apple' LIKE NULL;

The following example illustrates how to search multiple patterns in an array to find a match with the LIKE operator:

Concatenation operator

The concatenation operator combines multiple values into one.