Lexical structure and syntax in Google Standard SQL

Stay organized with collections Save and categorize content based on your preferences.

A Google Standard SQL statement comprises a series of tokens. Tokens include identifiers, quoted identifiers, literals, keywords, operators, and special characters. You can separate tokens with comments or whitespace such as spaces, backspaces, tabs, or newlines.

Identifiers

Identifiers are names that are associated with columns, tables, path expressions, and more. They can be unquoted or quoted and some are case-sensitive.

Unquoted identifiers

  • Must begin with a letter or an underscore (_) character.
  • Subsequent characters can be letters, numbers, underscores (_), and dashes, but additional rules apply for dashes (see the next rule).
  • Single dashes can be used with most unquoted identifiers, but not at the beginning or end.

    When you use dashes in an unquoted identifier, you essentially combine an unquoted identifier with one or more unquoted identifiers and numbers. When using dashes, you must generally follow this structure:

    unquoted_identifier[-{ unquoted_identifier | number }][...]
    

    For example: foo-bar, foo-22, foo-22-bar

    Dashes cannot be used in fieldnames.

Quoted identifiers

  • Must be enclosed by backtick (`) characters.
  • Can contain any characters, including spaces and symbols.
  • Cannot be empty.
  • Have the same escape sequences as string literals.
  • If an identifier is the same as a reserved keyword, the identifier must be quoted. For example, the identifier FROM must be quoted. Additional rules apply for path expressions.

Identifier examples

Path expression examples:

-- Valid. _5abc and dataField are valid identifiers.
_5abc.dataField

-- Valid. `5abc` and dataField are valid identifiers.
`5abc`.dataField

-- Invalid. 5abc is an invalid identifier because it is unquoted and starts
-- with a number rather than a letter or underscore.
5abc.dataField

-- Valid. abc5 and dataField are valid identifiers.
abc5.dataField

-- Invalid. abc5! is an invalid identifier because it is unquoted and contains
-- a character that is not a letter, number, or underscore.
abc5!.dataField

-- Valid. `GROUP` and dataField are valid identifiers.
`GROUP`.dataField

-- Invalid. GROUP is an invalid identifier because it is unquoted and is a
-- stand-alone reserved keyword.
GROUP.dataField

-- Valid. abc5 and GROUP are valid identifiers.
abc5.GROUP

Function examples:

-- Valid. dataField is a valid identifier in a function called foo().
foo().dataField

Array access operation examples:

-- Valid. dataField is a valid identifier in an array called items.
items[OFFSET(3)].dataField

Named query parameter examples:

-- Valid. param and dataField are valid identifiers.
@param.dataField

Path expressions

A path expression describes how to navigate to an object in a graph of objects and generally follows this structure:

path:
  [path_expression][. ...]

path_expression:
  [first_part]/subsequent_part[ { / | : | - } subsequent_part ][...]

first_part:
  { unquoted_identifier | quoted_identifier }

subsequent_part:
  { unquoted_identifier | quoted_identifier | number }
  • path: A graph of one or more objects.
  • path_expression: An object in a graph of objects.
  • first_part: A path expression can start with a quoted or unquoted identifier. If the path expressions starts with a reserved keyword, it must be a quoted identifier.
  • subsequent_part: Subsequent parts of a path expression can include non-identifiers, such as reserved keywords. If a subsequent part of a path expressions starts with a reserved keyword, it may be quoted or unquoted.

Examples:

foo.bar
foo.bar/25
foo/bar:25
foo/bar/25-31
/foo/bar
/25/foo/bar

Literals

A literal represents a constant value of a built-in data type. Some, but not all, data types can be expressed as literals.

String and bytes literals

Both string and bytes literals must be quoted, either with single (') or double (") quotation marks, or triple-quoted with groups of three single (''') or three double (""") quotation marks.

Quoted literals:

Literal Examples Description
Quoted string
  • "abc"
  • "it's"
  • 'it\'s'
  • 'Title: "Boy"'
Quoted strings enclosed by single (') quotes can contain unescaped double (") quotes, as well as the inverse.
Backslashes (\) introduce escape sequences. See the Escape Sequences table below.
Quoted strings cannot contain newlines, even when preceded by a backslash (\).
Triple-quoted string
  • """abc"""
  • '''it's'''
  • '''Title:"Boy"'''
  • '''two
    lines'''
  • '''why\?'''
Embedded newlines and quotes are allowed without escaping - see fourth example.
Backslashes (\) introduce escape sequences. See Escape Sequences table below.
A trailing unescaped backslash (\) at the end of a line is not allowed.
End the string with three unescaped quotes in a row that match the starting quotes.
Raw string
  • R"abc+"
  • r'''abc+'''
  • R"""abc+"""
  • r'f\(abc,(.*),def\)'
Quoted or triple-quoted literals that have the raw string literal prefix (r or R) are interpreted as raw/regex strings.
Backslash characters (\) do not act as escape characters. If a backslash followed by another character occurs inside the string literal, both characters are preserved.
A raw string cannot end with an odd number of backslashes.
Raw strings are useful for constructing regular expressions.

Prefix characters (r, R, b, B) are optional for quoted or triple-quoted strings, and indicate that the string is a raw/regex string or a byte sequence, respectively. For example, b'abc' and b'''abc''' are both interpreted as type bytes. Prefix characters are case-insensitive.

Quoted literals with prefixes:

Literal Example Description
Bytes
  • B"abc"
  • B'''abc'''
  • b"""abc"""
Quoted or triple-quoted literals that have the bytes literal prefix (b or B) are interpreted as bytes.
Raw bytes
  • br'abc+'
  • RB"abc+"
  • RB'''abc'''
The r and b prefixes can be combined in any order. For example, rb'abc*' is equivalent to br'abc*'.

The table below lists all valid escape sequences for representing non-alphanumeric characters in string and byte literals. Any sequence not in this table produces an error.

Escape Sequence Description
\a Bell
\b Backspace
\f Formfeed
\n Newline
\r Carriage Return
\t Tab
\v Vertical Tab
\\ Backslash (\)
\? Question Mark (?)
\" Double Quote (")
\' Single Quote (')
\` Backtick (`)
\ooo Octal escape, with exactly 3 digits (in the range 0–7). Decodes to a single Unicode character (in string literals) or byte (in bytes literals).
\xhh or \Xhh Hex escape, with exactly 2 hex digits (0–9 or A–F or a–f). Decodes to a single Unicode character (in string literals) or byte (in bytes literals). Examples:
  • '\x41' == 'A'
  • '\x41B' is 'AB'
  • '\x4' is an error
\uhhhh Unicode escape, with lowercase 'u' and exactly 4 hex digits. Valid only in string literals or identifiers.
Note that the range D800-DFFF is not allowed, as these are surrogate unicode values.
\Uhhhhhhhh Unicode escape, with uppercase 'U' and exactly 8 hex digits. Valid only in string literals or identifiers.
The range D800-DFFF is not allowed, as these values are surrogate unicode values. Also, values greater than 10FFFF are not allowed.

Integer literals

Integer literals are either a sequence of decimal digits (0–9) or a hexadecimal value that is prefixed with "0x" or "0X". Integers can be prefixed by "+" or "-" to represent positive and negative values, respectively. Examples:

123
0xABC
-123

An integer literal is interpreted as an INT64.

NUMERIC literals

You can construct NUMERIC literals using the NUMERIC keyword followed by a floating point value in quotes.

Examples:

SELECT NUMERIC '0';
SELECT NUMERIC '123456';
SELECT NUMERIC '-3.14';
SELECT NUMERIC '-0.54321';
SELECT NUMERIC '1.23456e05';
SELECT NUMERIC '-9.876e-3';

Floating point literals

Syntax options:

[+-]DIGITS.[DIGITS][e[+-]DIGITS]
[DIGITS].DIGITS[e[+-]DIGITS]
DIGITSe[+-]DIGITS

DIGITS represents one or more decimal numbers (0 through 9) and e represents the exponent marker (e or E).

Examples:

123.456e-67
.1E4
58.
4e2

Numeric literals that contain either a decimal point or an exponent marker are presumed to be type double.

Implicit coercion of floating point literals to float type is possible if the value is within the valid float range.

There is no literal representation of NaN or infinity, but the following case-insensitive strings can be explicitly cast to float:

  • "NaN"
  • "inf" or "+inf"
  • "-inf"

Array literals

Array literals are comma-separated lists of elements enclosed in square brackets. The ARRAY keyword is optional, and an explicit element type T is also optional.

Examples:

[1, 2, 3]
['x', 'y', 'xy']
ARRAY[1, 2, 3]
ARRAY<string>['x', 'y', 'xy']
ARRAY<int64>[]

Struct literals

A struct literal is a struct whose fields are all literals. Struct literals can be written using any of the syntaxes for constructing a struct (tuple syntax, typeless struct syntax, or typed struct syntax).

Note that tuple syntax requires at least two fields, in order to distinguish it from an ordinary parenthesized expression. To write a struct literal with a single field, use typeless struct syntax or typed struct syntax.

Example Output Type
(1, 2, 3) STRUCT<INT64, INT64, INT64>
(1, 'abc') STRUCT<INT64, STRING>
STRUCT(1 AS foo, 'abc' AS bar) STRUCT<foo INT64, bar STRING>
STRUCT<INT32, INT64>(1, 2) STRUCT<INT32, INT64>
STRUCT(1) STRUCT<INT64>
STRUCT<INT64>(1) STRUCT<INT64>

Date literals

Syntax:

DATE 'YYYY-M[M]-D[D]'

Date literals contain the DATE keyword followed by a string literal that conforms to the canonical date format, enclosed in single quotation marks. Date literals support a range between the years 1 and 9999, inclusive. Dates outside of this range are invalid.

For example, the following date literal represents September 27, 2014:

DATE '2014-09-27'

String literals in canonical date format also implicitly coerce to DATE type when used where a DATE-type expression is expected. For example, in the query

SELECT * FROM foo WHERE date_col = "2014-09-27"

the string literal "2014-09-27" will be coerced to a date literal.

Timestamp literals

Syntax:

TIMESTAMP 'YYYY-[M]M-[D]D[( |T)[H]H:[M]M:[S]S[.F]] [time_zone]'
TIMESTAMP 'YYYY-[M]M-[D]D[( |T)[H]H:[M]M:[S]S[.F]][time_zone_offset]'

Timestamp literals contain the TIMESTAMP keyword and a string literal that conforms to the canonical timestamp format, enclosed in single quotation marks.

Timestamp literals support a range between the years 1 and 9999, inclusive. Timestamps outside of this range are invalid.

A timestamp literal can include a numerical suffix to indicate the time zone:

TIMESTAMP '2014-09-27 12:30:00.45-08'

If this suffix is absent, the default time zone, America/Los_Angeles, is used.

For example, the following timestamp represents 12:30 p.m. on September 27, 2014 in the default time zone, America/Los_Angeles:

TIMESTAMP '2014-09-27 12:30:00.45'

For more information about time zones, see Time zone.

String literals with the canonical timestamp format, including those with time zone names, implicitly coerce to a timestamp literal when used where a timestamp expression is expected. For example, in the following query, the string literal "2014-09-27 12:30:00.45 America/Los_Angeles" is coerced to a timestamp literal.

SELECT * FROM foo
WHERE timestamp_col = "2014-09-27 12:30:00.45 America/Los_Angeles"

A timestamp literal can include these optional characters:

  • T or t: A flag for time. Use as a separator between the date and time.
  • Z or z: A flag for the default timezone. This cannot be used with [timezone].

If you use one of these characters, a space can't be included before or after it. These are valid:

TIMESTAMP '2017-01-18T12:34:56.123456Z'
TIMESTAMP '2017-01-18t12:34:56.123456'
TIMESTAMP '2017-01-18 12:34:56.123456z'
TIMESTAMP '2017-01-18 12:34:56.123456Z'

Time zone

Since timestamp literals must be mapped to a specific point in time, a time zone is necessary to correctly interpret a literal. If a time zone is not specified as part of the literal itself, then Google Standard SQL uses the default time zone value, which the Google Standard SQL implementation sets.

Google Standard SQL represents time zones using strings in the following canonical format, which represents the offset from Coordinated Universal Time (UTC).

Format:

(+|-)H[H][:M[M]]

Examples:

'-08:00'
'-8:15'
'+3:00'
'+07:30'
'-7'

Time zones can also be expressed using string time zone names from the tz database. For a less comprehensive but simpler reference, see the List of tz database time zones on Wikipedia. Canonical time zone names have the format <continent/[region/]city>, such as America/Los_Angeles.

Example:

TIMESTAMP '2014-09-27 12:30:00 America/Los_Angeles'
TIMESTAMP '2014-09-27 12:30:00 America/Argentina/Buenos_Aires'

JSON literals

Syntax:

JSON 'json_formatted_data'

A JSON literal represents JSON-formatted data.

Example:

JSON '
{
  "id": 10,
  "type": "fruit",
  "name": "apple",
  "on_menu": true,
  "recipes":
    {
      "salads":
      [
        { "id": 2001, "type": "Walnut Apple Salad" },
        { "id": 2002, "type": "Apple Spinach Salad" }
      ],
      "desserts":
      [
        { "id": 3001, "type": "Apple Pie" },
        { "id": 3002, "type": "Apple Scones" },
        { "id": 3003, "type": "Apple Crumble" }
      ]
    }
}
'

Case sensitivity

Google Standard SQL follows these rules for case sensitivity:

Category Case-sensitive? Notes
Keywords No  
Function names No  
Table names See Notes Table names are usually case-insensitive, but they might be case-sensitive when querying a database that uses case-sensitive table names.
Column names No  
String values Yes
String comparisons Yes  
Aliases within a query No  
Regular expression matching See Notes Regular expression matching is case-sensitive by default, unless the expression itself specifies that it should be case-insensitive.
LIKE matching Yes  

Reserved keywords

Keywords are a group of tokens that have special meaning in the Google Standard SQL language, and have the following characteristics:

  • Keywords cannot be used as identifiers unless enclosed by backtick (`) characters.
  • Keywords are case-insensitive.

Google Standard SQL has the following reserved keywords.

ALL
AND
ANY
ARRAY
AS
ASC
ASSERT_ROWS_MODIFIED
AT
BETWEEN
BY
CASE
CAST
COLLATE
CONTAINS
CREATE
CROSS
CUBE
CURRENT
DEFAULT
DEFINE
DESC
DISTINCT
ELSE
END
ENUM
ESCAPE
EXCEPT
EXCLUDE
EXISTS
EXTRACT
FALSE
FETCH
FOLLOWING
FOR
FROM
FULL
GROUP
GROUPING
GROUPS
HASH
HAVING
IF
IGNORE
IN
INNER
INTERSECT
INTERVAL
INTO
IS
JOIN
LATERAL
LEFT
LIKE
LIMIT
LOOKUP
MERGE
NATURAL
NEW
NO
NOT
NULL
NULLS
OF
ON
OR
ORDER
OUTER
OVER
PARTITION
PRECEDING
PROTO
RANGE
RECURSIVE
RESPECT
RIGHT
ROLLUP
ROWS
SELECT
SET
SOME
STRUCT
TABLESAMPLE
THEN
TO
TREAT
TRUE
UNBOUNDED
UNION
UNNEST
USING
WHEN
WHERE
WINDOW
WITH
WITHIN

Terminating semicolons

You can optionally use a terminating semicolon (;) when you submit a query string statement through an Application Programming Interface (API).

In a request containing multiple statements, you must separate statements with semicolons, but the semicolon is generally optional after the final statement. Some interactive tools require statements to have a terminating semicolon.

Trailing commas

You can optionally use a trailing comma (,) at the end of a column list in a SELECT statement. You might have a trailing comma as the result of programmatically creating a column list.

Example

SELECT name, release_date, FROM Books

Query parameters

You can use query parameters to substitute arbitrary expressions. However, query parameters cannot be used to substitute identifiers, column names, table names, or other parts of the query itself. Query parameters are defined outside of the query statement.

Client APIs allow the binding of parameter names to values; the query engine substitutes a bound value for a parameter at execution time.

Named query parameters

Syntax:

@parameter_name

A named query parameter is denoted using an identifier preceded by the @ character.

A named query parameter can start with an identifier or a reserved keyword. An identifier can be unquoted or quoted.

Example:

This example returns all rows where LastName is equal to the value of the named query parameter myparam.

SELECT * FROM Roster WHERE LastName = @myparam

Hints

@{ hint [, ...] }

hint:
  [engine_name.]hint_name = value

The purpose of a hint is to modify the execution strategy for a query without changing the result of the query. Hints generally do not affect query semantics, but may have performance implications. These hint types are available:

Hint syntax requires the @ character followed by curly braces. You can create one hint or a group of hints. The optional engine_name. prefix allows for multiple engines to define hints with the same hint_name. This is important if you need to suggest different engine-specific execution strategies or different engines support different hints.

You can assign identifiers and literals to hints.

  • Identifiers are useful for hints that are meant to act like enums. You can use an identifier to avoid using a quoted string. In the resolved AST, identifier hints are represented as string literals, so @{hint="abc"} is the same as @{hint=abc}. Identifier hints can also be used for hints that take a table name or column name as a single identifier.
  • NULL literals are allowed and are inferred as integers.

Hints are meant to apply only to the node they are attached to, and not to a larger scope. For example, a hint on a JOIN in the middle of the FROM clause is meant to apply to that JOIN only, and not other JOINs in the FROM clause. Statement-level hints can be used for hints that modify execution of an entire statement, for example an overall memory budget or deadline.

Examples

In this example, a literal is assigned to a hint. This hint is only used with two database engines called database_engine_a and database_engine_b. The value for the hint is different for each database engine.

@{ database_engine_a.file_count=23, database_engine_b.file_count=10 }

In this example, an identifier is assigned to a hint. There are unique identifiers for each hint type. You can view a list of hint types at the beginning of this topic.

@{ JOIN_METHOD=HASH_JOIN }

Comments

Comments are sequences of characters that the parser ignores. Google Standard SQL supports the following types of comments.

Single-line comments

Use a single-line comment if you want the comment to appear on a line by itself.

Examples

# this is a single-line comment
SELECT book FROM library;
-- this is a single-line comment
SELECT book FROM library;
/* this is a single-line comment */
SELECT book FROM library;
SELECT book FROM library
/* this is a single-line comment */
WHERE book = "Ulysses";

Inline comments

Use an inline comment if you want the comment to appear on the same line as a statement. A comment that is prepended with # or -- must appear to the right of a statement.

Examples

SELECT book FROM library; # this is an inline comment
SELECT book FROM library; -- this is an inline comment
SELECT book FROM library; /* this is an inline comment */
SELECT book FROM library /* this is an inline comment */ WHERE book = "Ulysses";

Multiline comments

Use a multiline comment if you need the comment to span multiple lines. Nested multiline comments are not supported.

Examples

SELECT book FROM library
/*
  This is a multiline comment
  on multiple lines
*/
WHERE book = "Ulysses";
SELECT book FROM library
/* this is a multiline comment
on two lines */
WHERE book = "Ulysses";