expressions.md 21 KB

Expressions

String concatenation {#concatenation}

Executed using the binary operator ||.

As with other binary operators, if the data on either side is NULL, the result is also NULL.

Don't confuse this operator with a logical "or": in SQL, it's denoted by the OR keyword. It's also not worth doing concatenation using +.

Examples

SELECT "fo" || "o";

Matching a string by pattern {#check-match}

REGEXP and RLIKE are aliases used to call Re2::Grep. MATCH: Same for Re2::Match.

LIKE works as follows:

  • Patterns can include two special characters:

    • %: Zero or more of any characters.
    • _: Exactly one of any character.

All other characters are literals that represent themselves.

  • As opposed to REGEXP, LIKE must be matched exactly. For example, to search a substring, add % at the beginning and end of the pattern.
  • ILIKE is a case-insensitive version of LIKE.
  • If LIKE is applied to the key column of the sorted table and the pattern doesn't start with a special character, filtering by prefix drills down directly to the cluster level, which in some cases lets you avoid the full table scan. This optimization is disabled for ILIKE.
  • To escape special characters, specify the escaped character after the pattern using the ESCAPE '?' keyword. Instead of ? you can use any character except %, _ and \. For example, if you use a question mark as an escape character, the expressions ?%, ?_ and ?? will match their second character in the template: percent, underscore, and question mark, respectively. The escape character is undefined by default.

The most popular way to use the LIKE and REGEXP keywords is to filter a table using the statements with the WHERE clause. However, there are no restrictions on using templates in this context: you can use them in most of contexts involving strings, for example, with concatenation by using ||.

Examples

SELECT * FROM my_table
WHERE string_column REGEXP '\\d+';
-- the second slash is required because
-- all the standard string literals in SQL
-- can accept C-escaped strings
SELECT
    string_column LIKE '___!_!_!_!!!!!!' ESCAPE '!'
    -- searches for a string of exactly 9 characters:
    --   3 arbitrary characters
    --   followed by 3 underscores
    --  and 3 exclamation marks
FROM my_table;
SELECT * FROM my_table
WHERE key LIKE 'foo%bar';
-- if the table is sorted by key, it will only scan the keys,
-- starting with "foo", and then, among them,
-- will leave only those that end in "bar"

Operators

Arithmetic operators {#math-operators}

The operators +, -, *, /, % are defined for primitive data types that are variations of numbers.

For the Decimal data type, bankers rounding is used (to the nearest even integer).

Examples

SELECT 2 + 2;
SELECT 0.0 / 0.0;

Comparison operators {#comparison-operators}

The operators =, ==, !=, <>, >, < are defined for:

  • Primitive data types except Yson and Json.
  • Tuples and structures with the same set of fields. No order is defined for structures, but you can check for (non-)equality. Tuples are compared element-by-element left to right.

Examples

SELECT 2 > 1;

Logical operators {#logic-operators}

Use the operators AND, OR, XOR for logical operations on Boolean values (Bool).

Examples*

SELECT 3 > 0 AND false;

Bitwise operators {#bit-operators}

Bitwise operations on numbers:

  • &, |, ^: AND, OR, and XOR, respectively. Don't confuse bitwise operations with the related keywords. The keywords AND, OR, and XOR are used for Boolean values only, but not for numbers.
  • ~: A negation.
  • <, >: Left or right shifts.
  • |<, >|: Circular left or right shifts.

Examples

SELECT
    key << 10 AS key,
    ~value AS value
FROM my_table;

Precedence and associativity of operators {#operator-priority}

Operator precedence determines the order of evaluation of an expression that contains different operators. For example, the expression 1 + 2 * 3 is evaluated as 1 + (2 * 3) because the multiplication operator has a higher precedence than the addition operator.

Associativity determines the order of evaluating expressions containing operators of the same type. For example, the expression 1 + 2 + 3 is evaluated as (1 + 2) + 3 because the addition operator is left-associative. On the other hand, the expression a ?? b ?? c is evaluated as a ?? (b ?? c) because the ?? operator is right-associative

The table below shows precedence and associativity of YQL operators. The operators in the table are listed in descending order of precedence.

Priority Operator Description Associativity
1 a[], a.foo, a() Accessing a container item, calling a function Left
2 +a, -a, ~a, NOT a Unary operators: plus, minus, bitwise and logical negation Right
3 a\|\|b String concatenation Left
4 a*b, a/b, a%b Multiplication, division, remainder of division Left
5 a+b, a-b Addition/Subtraction Left
6 a ?? b Operator notation for NVL/COALESCE Right
7 a<b, a>b, a\|<b, a>\|b, a\|b, a^b, a&b Shift operators and logical bit operators Left
8 a<b, a=b, a=b, a>b Comparison Left
9 a IN b Occurrence of an element in a set Left
9 a==b, a=b, a!=b, a<>b, a is (not) distinct from b Comparison for (non-)equality Left
10 a XOR b Logical XOR Left
11 a AND b Logical AND Left
12 a OR b Logical OR Left

IS [NOT] NULL {#is-null}

Matching an empty value (NULL). Since NULL is a special value equal to nothing, the ordinary comparison operators can't be used to match it.

Examples

SELECT key FROM my_table
WHERE value IS NOT NULL;

IS [NOT] DISTINCT FROM {#is-distinct-from}

Comparing of two values. Unlike the regular comparison operators, NULLs are treated as equal to each other. More precisely, the comparison is carried out according to the following rules:

  1. The operators IS DISTINCT FROM/IS NOT DISTINCT FROM are defined for those and only for those arguments for which the operators != and = are defined.
  2. The result of IS NOT DISTINCT FROM is equal to the logical negation of the IS DISTINCT FROM result for these arguments.
  3. If the result of the == operator is not equal to zero for some arguments, then it is equal to the result of the IS NOT DISTINCT FROM operator for the same arguments.
  4. If both arguments are empty Optional or NULLs, then the value of IS NOT DISTINCT FROM is True.
  5. The result of IS NOT DISTINCT FROM for an empty Optional or NULL and filled-in Optional or non-Optional value is False.

For values of composite types, these rules are used recursively.

BETWEEN {#between}

Checking whether a value is in a range. It's equivalent to two conditions with >= and <= (range boundaries are included). Can be used with the NOT prefix to support inversion.

Examples

SELECT * FROM my_table
WHERE key BETWEEN 10 AND 20;

IN {#in}

Checking whether a value is inside of a set of values. It's logically equivalent to a chain of equality comparisons using OR but implemented more efficiently.

{% note warning "Warning" %}

Unlike a similar keyword in Python, in YQL IN DOES NOT search for a substring inside a string. To search for a substring, use the function String::Contains or LIKE/REGEXP mentioned above.

{% endnote %}

Immediately after IN, you can specify the COMPACT modifier. If COMPACT is not specified, then IN with a subquery is executed as a relevant JOIN (LEFT SEMI for IN and LEFT ONLY for NOT IN), if possible. Using the COMPACT modifier forces the in-memory execution strategy: a hash table is immediately built from the contents of the right IN part in-memory, and then the left part is filtered.

The COMPACT modifier must be used with care. Since the hash table is built in-memory, the query may fail if the right part of IN contains many large or different elements.

It is prefirable to add large lists of values to your query by URLs and use the ParseFile function.

Examples

SELECT column IN (1, 2, 3)
FROM my_table;
SELECT * FROM my_table
WHERE string_column IN ("a", "b", "c");
$foo = AsList(1, 2, 3);
SELECT 1 IN $foo;
$values = (SELECT column + 1 FROM table);
SELECT * FROM my_table WHERE
    -- filtering by an in-memory hash table for one_table
    column1 IN COMPACT $values AND
    -- followed by LEFT ONLY JOIN with other_table
    column2 NOT IN (SELECT other_column FROM other_table);

AS {#as}

Can be used in the following scenarios:

  • Adding a short name (alias) for columns or tables within the query.
  • Using named arguments in function calls.
  • To specify the target type in the case of explicit type casting, see CAST.

Examples

SELECT key AS k FROM my_table;
SELECT t.key FROM my_table AS t;
SELECT
    MyFunction(key, 123 AS my_optional_arg)
FROM my_table;

CAST {#cast}

Tries to cast the value to the specified type. The attempt may fail and return NULL. When used with numbers, it may lose precision or most significant bits. For lists and dictionaries, it can either delete or replace with NULL the elements whose conversion failed. For structures and tuples, it deletes elements that are omitted in the target type. For more information about casting rules, see here.

{% include decimal_args %}

Examples

{% include cast_examples %}

BITCAST {#bitcast}

Performs a bitwise conversion of an integer value to the specified integer type. The conversion is always successful, but may lose precision or high-order bits.

Examples

SELECT
    BITCAST(100000ul AS Uint32),     -- 100000
    BITCAST(100000ul AS Int16),      -- -31072
    BITCAST(100000ul AS Uint16),     -- 34464
    BITCAST(-1 AS Int16),            -- -1
    BITCAST(-1 AS Uint16);           -- 65535

CASE {#case}

Conditional expressions and branching. It's similar to if, switch and ternary operators in the imperative programming languages. If the result of the WHEN expression is true, the value of the CASE expression becomes the result following the condition, and the rest of the CASE expression isn't calculated. If the condition is not met, all the WHEN clauses that follow are checked. If none of the WHEN clauses are met, the CASE value is assigned the result from the ELSE clause. The ELSE branch is mandatory in the CASE expression. Expressions in WHEN are checked sequentially, from top to bottom.

Since its syntax is quite sophisticated, it's often more convenient to use the built-in function IF.

Examples

SELECT
  CASE
    WHEN value > 0
    THEN "positive"
    ELSE "negative"
  END
FROM my_table;
SELECT
  CASE value
    WHEN 0 THEN "zero"
    WHEN 1 THEN "one"
    ELSE "not zero or one"
  END
FROM my_table;

Named expressions {#named-nodes}

Complex queries may be sophisticated, containing lots of nested levels and/or repeating parts. In YQL, you can use named expressions to assign a name to an arbitrary expression or subquery. Named expressions can be referenced in other expressions or subqueries. In this case, the original expression/subquery is actually substituted at point of use.

A named expression is defined as follows:

<named-expr> = <expression> | <subquery>;

Here <named-expr> consists of a $ character and an arbitrary non-empty identifier (for example, $foo).

If the expression on the right is a tuple, you can automatically unpack it by specifying several named expressions separated by commas on the left:

<named-expr1>, <named-expr2>, <named-expr3> ... = <expression-returning-tuple>;

In this case, the number of expressions must match the tuple size.

Each named expression has a scope. It starts immediately after the definition of a named expression and ends at the end of the nearest enclosed namescope (for example, at the end of the query or at the end of the body of the lambda function, ACTION, SUBQUERY, or the cycle EVALUATE FOR). Redefining a named expression with the same name hides the previous expression from the current scope.

If the named expression has never been used, a warning is issued. To avoid such a warning, use the underscore as the first character in the ID (for example, $_foo). The named expression $_ is called an anonymous named expression and is processed in a special way: it works as if $_ would be automatically replaced by $_<some_uniq_name>. Anonymous named expressions are convenient when you don't need the expression value. For example, to fetch the second element from a tuple of three elements, you can write:

$_, $second, $_ = AsTuple(1, 2, 3);
select $second;

An attempt to reference an anonymous named expression results in an error:

$_ = 1;
select $_; --- error: Unable to reference anonymous name $_
export $_; --- An error: Can not export anonymous name $_

Moreover, you can't import a named expression with an anonymous alias:

import utils symbols $sqrt as $_; --- error: Can not import anonymous name $_

Anonymous argument names are also supported for lambda functions, ACTION, SUBQUERY, and in EVALUATE FOR.

{% note info %}

If named expression substitution results in completely identical subgraphs in the query execution graph, the graphs are combined to execute a subgraph only once.

{% endnote %}

Examples

$multiplier = 712;
SELECT
  a * $multiplier, -- $multiplier is 712
  b * $multiplier,
  (a + b) * $multiplier
FROM abc_table;
$multiplier = c;
SELECT
  a * $multiplier -- $multiplier is column c
FROM abc_table;
$intermediate = (
  SELECT
    value * value AS square,
    value
  FROM my_table
);
SELECT a.square * b.value
FROM $intermediate AS a
INNER JOIN $intermediate AS b
ON a.value == b.square;
$a, $_, $c = AsTuple(1, 5u, "test"); -- unpack a tuple
SELECT $a, $c;
$x, $y = AsTuple($y, $x); -- swap expression values

Table expressions {#table-contexts}

A table expression is an expression that returns a table. Table expressions in YQL are as follows:

  • Subqueries: (SELECT key, subkey FROM T)
  • Named subqueries: $foo = SELECT * FROM T; (in this case, $foo is also a table expression)
  • Subquery templates: DEFINE SUBQUERY $foo($name) AS ... END DEFINE; ($foo("InputTable") is a table expression).

Semantics of a table expression depends on the context where it is used. In YQL, table expressions can be used in the following contexts:

  • Table context: after FROM. In this case, table expressions work as expected: for example, $input = SELECT a, b, c FROM T; SELECT * FROM $input returns a table with three columns. The table context also occurs after UNION ALL, JOIN, PROCESS, REDUCE;
  • Vector context: after IN. In this context, the table expression must contain exactly one column (the name of this column doesn't affect the expression result in any way). A table expression in a vector context is typed as a list (the type of the list element is the same as the column type in this case). Example: SELECT * FROM T WHERE key IN (SELECT k FROM T1);
  • A scalar context arises in all the other cases. As in a vector context, a table expression must contain exactly one column, but the value of the table expression is a scalar, that is, an arbitrarily selected value of this column (if no rows are returned, the result is NULL). Example: $count = SELECT COUNT(*) FROM T; SELECT * FROM T ORDER BY key LIMIT $count / 2;

The order of rows in a table context, the order of elements in a vector context, and the rule for selecting a value from a scalar context (if multiple values are returned), aren't defined. This order also cannot be affected by ORDER BY: ORDER BY without LIMIT is ignored in table expressions with a warning, and ORDER BY with LIMIT defines a set of elements rather than the order within that set.

There is an exception to this rule. Named expression with PROCESS, if used in a scalar context, behaves as in a table context:

$input = SELECT 1 AS key, 2 AS value;
$process = PROCESS $input;

SELECT FormatType(TypeOf($process)); -- $process is used in a scalar context,
                                     -- but the SELECT result in this case is List<Struct'key':Int32,'value':Int32>

SELECT $process[0].key; -- that returns 1

SELECT FormatType(TypeOf($input)); -- throws an error: $input in a scalar context must contain one column

{% note warning %}

A common error is to use an expression in a scalar context rather than a table context or vector context. For example:

$dict = SELECT key, value FROM T1;

DEFINE SUBQUERY $merge_dict($table, $dict) AS
SELECT * FROM $table LEFT JOIN $dict USING(key);
END DEFINE;

SELECT * FROM $merge_dict("Input", $dict); -- $dict is used in a scalar context in this case.
                                           -- an error: exactly one column is expected in a scalar context

A correct notation in this case is:

DEFINE SUBQUERY $dict() AS
SELECT key, value FROM T1;
END DEFINE;

DEFINE SUBQUERY $merge_dict($table, $dict) AS
SELECT * FROM $table LEFT JOIN $dict() USING(key); -- Using the table expression $dict()
                                                   -- (Calling a subquery template) in a table context
END DEFINE;

SELECT * FROM $merge_dict("Input", $dict); -- $dict - is a subquery template (rather than a table expression)
                                           -- that is passed as an argument of a table expression

{% endnote %}

Lambda functions {#lambda}

Let you combine multiple expressions into a single callable value.

List arguments in round brackets, following them by the arrow and lambda function body. The lambda function body includes either an expression in round brackets or curly brackets around an optional chain of named expressions assignments and the call result after the RETURN keyword in the last expression.

The scope for the lambda body: first the local named expressions, then arguments, then named expressions defined above by the lambda function at the top level of the query.

Only use pure expressions inside the lambda body (those might also be other lambdas, possibly passed through arguments). However, you can't use SELECT, INSERT INTO, or other top-level expressions.

One or more of the last lambda parameters can be marked with a question mark as optional: if they haven't been specified when calling lambda, they are assigned the NULL value.

Examples

$f = ($y) -> {
    $prefix = "x";
    RETURN $prefix || $y;
};

$g = ($y) -> ("x" || $y);

$h = ($x, $y?) -> ($x + ($y ?? 0));

SELECT $f("y"), $g("z"), $h(1), $h(2, 3); -- "xy", "xz", 1, 5
-- if the lambda result is calculated by a single expression, then you can use a more compact syntax:
$f = ($x, $_) -> ($x || "suffix"); -- the second argument is not used
SELECT $f("prefix_", "whatever");

Accessing containers {#items-access}

For accessing the values inside containers:

  • Struct<>, Tuple<> and Variant<>, use a dot. The set of keys (for the tuple and the corresponding variant — indexes) is known at the query compilation time. The key is validated before beginning the query execution.
  • List<> and Dict<>, use square brackets. The set of keys (set of indexes for keys) is known only at the query execution time. The key is not validated before beginning the query execution. If no value is found, an empty value (NULL) is returned.

Description and list of available containers.

When using this syntax to access containers within table columns, be sure to specify the full column name, including the table name or table alias separated by a dot (see the first example below).

Examples

SELECT
  t.struct.member,
  t.tuple.7,
  t.dict["key"],
  t.list[7]
FROM my_table AS t;
SELECT
  Sample::ReturnsStruct().member;