Relational Data - Data Types - Variant

From FojiSoft Docs

This type represents a union of other data types. Type Variant(T1, T2, ..., TN) means that each row of this type has a value of either type T1 or T2 or … or TN or none of them (NULL value).

The order of nested types doesn’t matter: Variant(T1, T2) = Variant(T2, T1). Nested types can be arbitrary types except Nullable(…), LowCardinality(Nullable(…)) and Variant(…) types.

It’s not recommended to use similar types as variants (for example different numeric types like Variant(UInt32, Int64) or different date types like Variant(Date, DateTime)), because working with values of such types can lead to ambiguity. By default, creating such Variant type will lead to an exception, but can be enabled using setting allow_suspicious_variant_types


The Variant data type is an experimental feature. To use it, set allow_experimental_variant_type = 1.


Creating Variant

Using Variant type in table column definition:

CREATE TABLE test (v Variant(UInt64, String, Array(UInt64))) ENGINE = Memory;
INSERT INTO test VALUES (NULL), (42), ('Hello, World!'), ([1, 2, 3]);
SELECT v FROM test;
┌─v─────────────┐
│ ᴺᵁᴸᴸ          │
│ 42            │
│ Hello, World! │
│ [1,2,3]       │
└───────────────┘

Using CAST from ordinary columns:

SELECT toTypeName(variant) as type_name, 'Hello, World!'::Variant(UInt64, String, Array(UInt64)) as variant;
┌─type_name──────────────────────────────┬─variant───────┐
│ Variant(Array(UInt64), String, UInt64) │ Hello, World! │
└────────────────────────────────────────┴───────────────┘

Using functions if/multiIf when arguments don’t have common type (setting use_variant_as_common_type should be enabled for it):

SET use_variant_as_common_type = 1;
SELECT if(number % 2, number, range(number)) as variant FROM numbers(5);
┌─variant───┐
│ []        │
│ 1         │
│ [0,1]     │
│ 3         │
│ [0,1,2,3] │
└───────────┘
SET use_variant_as_common_type = 1;
SELECT multiIf((number % 4) = 0, 42, (number % 4) = 1, [1, 2, 3], (number % 4) = 2, 'Hello, World!', NULL) AS variant FROM numbers(4);
┌─variant───────┐
│ 42            │
│ [1,2,3]       │
│ Hello, World! │
│ ᴺᵁᴸᴸ          │
└───────────────┘

Using functions ‘array/map’ if array elements/map values don’t have common type (setting use_variant_as_common_type should be enabled for it):

SET use_variant_as_common_type = 1;
SELECT array(range(number), number, 'str_' || toString(number)) as array_of_variants FROM numbers(3);
┌─array_of_variants─┐
│ [[],0,'str_0']    │
│ [[0],1,'str_1']   │
│ [[0,1],2,'str_2'] │
└───────────────────┘
SET use_variant_as_common_type = 1;
SELECT map('a', range(number), 'b', number, 'c', 'str_' || toString(number)) as map_of_variants FROM numbers(3);
┌─map_of_variants───────────────┐
│ {'a':[],'b':0,'c':'str_0'}    │
│ {'a':[0],'b':1,'c':'str_1'}   │
│ {'a':[0,1],'b':2,'c':'str_2'} │
└───────────────────────────────┘

Reading Variant nested types as subcolumns

Variant type supports reading a single nested type from a Variant column using the type name as a subcolumn. So, if you have column variant Variant(T1, T2, T3) you can read a subcolumn of type T2 using syntax variant.T2, this subcolumn will have type Nullable(T2) if T2 can be inside Nullable and T2 otherwise. This subcolumn will be the same size as original Variant column and will contain NULL values (or empty values if T2 cannot be inside Nullable) in all rows in which original Variant column doesn’t have type T2.

Variant subcolumns can be also read using function variantElement(variant_column, type_name).

Examples:

CREATE TABLE test (v Variant(UInt64, String, Array(UInt64))) ENGINE = Memory;
INSERT INTO test VALUES (NULL), (42), ('Hello, World!'), ([1, 2, 3]);
SELECT v, v.String, v.UInt64, v.`Array(UInt64)` FROM test;
┌─v─────────────┬─v.String──────┬─v.UInt64─┬─v.Array(UInt64)─┐
│ ᴺᵁᴸᴸ          │ ᴺᵁᴸᴸ          │     ᴺᵁᴸᴸ │ []              │
│ 42            │ ᴺᵁᴸᴸ          │       42 │ []              │
│ Hello, World! │ Hello, World! │     ᴺᵁᴸᴸ │ []              │
│ [1,2,3]       │ ᴺᵁᴸᴸ          │     ᴺᵁᴸᴸ │ [1,2,3]         │
└───────────────┴───────────────┴──────────┴─────────────────┘
SELECT toTypeName(v.String), toTypeName(v.UInt64), toTypeName(v.`Array(UInt64)`) FROM test LIMIT 1;
┌─toTypeName(v.String)─┬─toTypeName(v.UInt64)─┬─toTypeName(v.Array(UInt64))─┐
│ Nullable(String)     │ Nullable(UInt64)     │ Array(UInt64)               │
└──────────────────────┴──────────────────────┴─────────────────────────────┘
SELECT v, variantElement(v, 'String'), variantElement(v, 'UInt64'), variantElement(v, 'Array(UInt64)') FROM test;
┌─v─────────────┬─variantElement(v, 'String')─┬─variantElement(v, 'UInt64')─┬─variantElement(v, 'Array(UInt64)')─┐
│ ᴺᵁᴸᴸ          │ ᴺᵁᴸᴸ                        │                        ᴺᵁᴸᴸ │ []                                 │
│ 42            │ ᴺᵁᴸᴸ                        │                          42 │ []                                 │
│ Hello, World! │ Hello, World!               │                        ᴺᵁᴸᴸ │ []                                 │
│ [1,2,3]       │ ᴺᵁᴸᴸ                        │                        ᴺᵁᴸᴸ │ [1,2,3]                            │
└───────────────┴─────────────────────────────┴─────────────────────────────┴────────────────────────────────────┘

To know what variant is stored in each row function variantType(variant_column) can be used. It returns Enum with variant type name for each row (or 'None' if row is NULL).

Example:

CREATE TABLE test (v Variant(UInt64, String, Array(UInt64))) ENGINE = Memory;
INSERT INTO test VALUES (NULL), (42), ('Hello, World!'), ([1, 2, 3]);
SELECT variantType(v) from test;
┌─variantType(v)─┐
│ None           │
│ UInt64         │
│ String         │
│ Array(UInt64)  │
└────────────────┘
SELECT toTypeName(variantType(v)) FROM test LIMIT 1;
┌─toTypeName(variantType(v))──────────────────────────────────────────┐
│ Enum8('None' = -1, 'Array(UInt64)' = 0, 'String' = 1, 'UInt64' = 2) │
└─────────────────────────────────────────────────────────────────────┘

Conversion between a Variant column and other columns

There are 4 possible conversions that can be performed with a column of type Variant.

Converting a String column to a Variant column

Conversion from String to Variant is performed by parsing a value of Variant type from the string value:

SELECT '42'::Variant(String, UInt64) as variant, variantType(variant) as variant_type
┌─variant─┬─variant_type─┐
│ 42      │ UInt64       │
└─────────┴──────────────┘
SELECT '[1, 2, 3]'::Variant(String, Array(UInt64)) as variant, variantType(variant) as variant_type
┌─variant─┬─variant_type──┐
│ [1,2,3] │ Array(UInt64) │
└─────────┴───────────────┘
SELECT CAST(map('key1', '42', 'key2', 'true', 'key3', '2020-01-01'), 'Map(String, Variant(UInt64, Bool, Date))') as map_of_variants, mapApply((k, v) -> (k, variantType(v)), map_of_variants) as map_of_variant_types```
┌─map_of_variants─────────────────────────────┬─map_of_variant_types──────────────────────────┐
│ {'key1':42,'key2':true,'key3':'2020-01-01'} │ {'key1':'UInt64','key2':'Bool','key3':'Date'} │
└─────────────────────────────────────────────┴───────────────────────────────────────────────┘

Converting an ordinary column to a Variant column

It is possible to convert an ordinary column with type T to a Variant column containing this type:

SELECT toTypeName(variant) as type_name, [1,2,3]::Array(UInt64)::Variant(UInt64, String, Array(UInt64)) as variant, variantType(variant) as variant_name
┌─type_name──────────────────────────────┬─variant─┬─variant_name──┐
│ Variant(Array(UInt64), String, UInt64) │ [1,2,3] │ Array(UInt64) │
└────────────────────────────────────────┴─────────┴───────────────┘

Note: converting from String type is always performed through parsing, if you need to convert String column to String variant of a Variant without parsing, you can do the following:

SELECT '[1, 2, 3]'::Variant(String)::Variant(String, Array(UInt64), UInt64) as variant, variantType(variant) as variant_type
┌─variant───┬─variant_type─┐
│ [1, 2, 3] │ String       │
└───────────┴──────────────┘

Converting a Variant column to an ordinary column

It is possible to convert a Variant column to an ordinary column. In this case all nested variants will be converted to a destination type:

CREATE TABLE test (v Variant(UInt64, String)) ENGINE = Memory;
INSERT INTO test VALUES (NULL), (42), ('42.42');
SELECT v::Nullable(Float64) FROM test;
┌─CAST(v, 'Nullable(Float64)')─┐
│                         ᴺᵁᴸᴸ │
│                           42 │
│                        42.42 │
└──────────────────────────────┘

Converting a Variant to another Variant

It is possible to convert a Variant column to another Variant column, but only if the destination Variant column contains all nested types from the original Variant:

CREATE TABLE test (v Variant(UInt64, String)) ENGINE = Memory;
INSERT INTO test VALUES (NULL), (42), ('String');
SELECT v::Variant(UInt64, String, Array(UInt64)) FROM test;
┌─CAST(v, 'Variant(UInt64, String, Array(UInt64))')─┐
│ ᴺᵁᴸᴸ                                              │
│ 42                                                │
│ String                                            │
└───────────────────────────────────────────────────┘

Reading Variant type from the data

All text formats (TSV, CSV, CustomSeparated, Values, JSONEachRow, etc) supports reading Variant type. During data parsing ClickHouse tries to insert value into most appropriate variant type.

Example:

SELECT
    v,
    variantElement(v, 'String') AS str,
    variantElement(v, 'UInt64') AS num,
    variantElement(v, 'Float64') AS float,
    variantElement(v, 'DateTime') AS date,
    variantElement(v, 'Array(UInt64)') AS arr
FROM format(JSONEachRow, 'v Variant(String, UInt64, Float64, DateTime, Array(UInt64))', $$
{"v" : "Hello, World!"},
{"v" : 42},
{"v" : 42.42},
{"v" : "2020-01-01 00:00:00"},
{"v" : [1, 2, 3]}
$$)
┌─v───────────────────┬─str───────────┬──num─┬─float─┬────────────────date─┬─arr─────┐
│ Hello, World!       │ Hello, World! │ ᴺᵁᴸᴸ │  ᴺᵁᴸᴸ │                ᴺᵁᴸᴸ │ []      │
│ 42                  │ ᴺᵁᴸᴸ          │   42 │  ᴺᵁᴸᴸ │                ᴺᵁᴸᴸ │ []      │
│ 42.42               │ ᴺᵁᴸᴸ          │ ᴺᵁᴸᴸ │ 42.42 │                ᴺᵁᴸᴸ │ []      │
│ 2020-01-01 00:00:00 │ ᴺᵁᴸᴸ          │ ᴺᵁᴸᴸ │  ᴺᵁᴸᴸ │ 2020-01-01 00:00:00 │ []      │
│ [1,2,3]             │ ᴺᵁᴸᴸ          │ ᴺᵁᴸᴸ │  ᴺᵁᴸᴸ │                ᴺᵁᴸᴸ │ [1,2,3] │
└─────────────────────┴───────────────┴──────┴───────┴─────────────────────┴─────────┘

Comparing values of Variant type

Values of a Variant type can be compared only with values with the same Variant type.

The result of operator < for values v1 with underlying type T1 and v2 with underlying type T2 of a type Variant(..., T1, ... T2, ...) is defined as follows: - If T1 = T2 = T, the result will be v1.T < v2.T (underlying values will be compared). - If T1 != T2, the result will be T1 < T2 (type names will be compared).

Examples:

CREATE TABLE test (v1 Variant(String, UInt64, Array(UInt32)), v2 Variant(String, UInt64, Array(UInt32))) ENGINE=Memory;
INSERT INTO test VALUES (42, 42), (42, 43), (42, 'abc'), (42, [1, 2, 3]), (42, []), (42, NULL);
SELECT v2, variantType(v2) as v2_type from test order by v2;
┌─v2──────┬─v2_type───────┐
│ []      │ Array(UInt32) │
│ [1,2,3] │ Array(UInt32) │
│ abc     │ String        │
│ 42      │ UInt64        │
│ 43      │ UInt64        │
│ ᴺᵁᴸᴸ    │ None          │
└─────────┴───────────────┘
SELECT v1, variantType(v1) as v1_type, v2, variantType(v2) as v2_type, v1 = v2, v1 < v2, v1 > v2 from test;
┌─v1─┬─v1_type─┬─v2──────┬─v2_type───────┬─equals(v1, v2)─┬─less(v1, v2)─┬─greater(v1, v2)─┐
│ 42 │ UInt64  │ 42      │ UInt64        │              1 │            0 │               0 │
│ 42 │ UInt64  │ 43      │ UInt64        │              0 │            1 │               0 │
│ 42 │ UInt64  │ abc     │ String        │              0 │            0 │               1 │
│ 42 │ UInt64  │ [1,2,3] │ Array(UInt32) │              0 │            0 │               1 │
│ 42 │ UInt64  │ []      │ Array(UInt32) │              0 │            0 │               1 │
│ 42 │ UInt64  │ ᴺᵁᴸᴸ    │ None          │              0 │            1 │               0 │
└────┴─────────┴─────────┴───────────────┴────────────────┴──────────────┴─────────────────┘

If you need to find the row with specific Variant value, you can do one of the following:

  • Cast value to the corresponding Variant type:
SELECT * FROM test WHERE v2 == [1,2,3]::Array(UInt32)::Variant(String, UInt64, Array(UInt32));
┌─v1─┬─v2──────┐
│ 42 │ [1,2,3] │
└────┴─────────┘
  • Compare Variant subcolumn with required type:
SELECT * FROM test WHERE v2.`Array(UInt32)` == [1,2,3] -- or using variantElement(v2, 'Array(UInt32)')
┌─v1─┬─v2──────┐
│ 42 │ [1,2,3] │
└────┴─────────┘

Sometimes it can be useful to make additional check on variant type as subcolumns with complex types like Array/Map/Tuple cannot be inside Nullable and will have default values instead of NULL on rows with different types:

SELECT v2, v2.`Array(UInt32)`, variantType(v2) FROM test WHERE v2.`Array(UInt32)` == [];
┌─v2───┬─v2.Array(UInt32)─┬─variantType(v2)─┐
│ 42   │ []               │ UInt64          │
│ 43   │ []               │ UInt64          │
│ abc  │ []               │ String          │
│ []   │ []               │ Array(UInt32)   │
│ ᴺᵁᴸᴸ │ []               │ None            │
└──────┴──────────────────┴─────────────────┘
SELECT v2, v2.`Array(UInt32)`, variantType(v2) FROM test WHERE variantType(v2) == 'Array(UInt32)' AND v2.`Array(UInt32)` == [];
┌─v2─┬─v2.Array(UInt32)─┬─variantType(v2)─┐
│ [] │ []               │ Array(UInt32)   │
└────┴──────────────────┴─────────────────┘

Note: values of variants with different numeric types are considered as different variants and not compared between each other, their type names are compared instead.

Example:

SET allow_suspicious_variant_types = 1;
CREATE TABLE test (v Variant(UInt32, Int64)) ENGINE=Memory;
INSERT INTO test VALUES (1::UInt32), (1::Int64), (100::UInt32), (100::Int64);
SELECT v, variantType(v) FROM test ORDER by v;
┌─v───┬─variantType(v)─┐
│ 1   │ Int64          │
│ 100 │ Int64          │
│ 1   │ UInt32         │
│ 100 │ UInt32         │
└─────┴────────────────┘

JSONExtract functions with Variant

All JSONExtract* functions support Variant type:

SELECT JSONExtract('{"a" : [1, 2, 3]}', 'a', 'Variant(UInt32, String, Array(UInt32))') AS variant, variantType(variant) AS variant_type;
┌─variant─┬─variant_type──┐
│ [1,2,3] │ Array(UInt32) │
└─────────┴───────────────┘
SELECT JSONExtract('{"obj" : {"a" : 42, "b" : "Hello", "c" : [1,2,3]}}', 'obj', 'Map(String, Variant(UInt32, String, Array(UInt32)))') AS map_of_variants, mapApply((k, v) -> (k, variantType(v)), map_of_variants) AS map_of_variant_types
┌─map_of_variants──────────────────┬─map_of_variant_types────────────────────────────┐
│ {'a':42,'b':'Hello','c':[1,2,3]} │ {'a':'UInt32','b':'String','c':'Array(UInt32)'} │
└──────────────────────────────────┴─────────────────────────────────────────────────┘
SELECT JSONExtractKeysAndValues('{"a" : 42, "b" : "Hello", "c" : [1,2,3]}', 'Variant(UInt32, String, Array(UInt32))') AS variants, arrayMap(x -> (x.1, variantType(x.2)), variants) AS variant_types
┌─variants───────────────────────────────┬─variant_types─────────────────────────────────────────┐
│ [('a',42),('b','Hello'),('c',[1,2,3])] │ [('a','UInt32'),('b','String'),('c','Array(UInt32)')] │
└────────────────────────────────────────┴───────────────────────────────────────────────────────┘