Numeric types encoding matches memory layout of little endian CPUs like AMD64 or ARM64, allowing very efficient encoding and decoding.
| Type | Encoding |
|---|---|
| Integers (Int/UInt) | 8, 16, 32, 64, 128 or 256 bits in little endian |
| Floats (Float32/Float64) | IEEE 754 binary representation |
| String | Array of strings as (len, value) |
| FixedString(N) | Array of N-byte sequences |
| IPv4 | Alias of UInt32, represented as UInt32 |
| IPv6 | Alias of FixedString(16), represented as binary |
| Tuple | Array of columns encoded continuously. Example: Tuple(String, UInt8) = two continuous columns |
| Map | Map(K, V) = three columns: Offsets ColUInt64, Keys K, Values V. Row count in Keys/Values = last Offsets value |
| Array | Array(T) = two columns: Offsets ColUInt64, Data T. Row count in Data = last Offsets value |
| Nullable | Nullable(T) = two columns: Nulls ColUInt8, Values T with same row count. Nulls is mask: 1=null, 0=value |
| UUID | Alias of FixedString(16), represented as binary |
| Enum | Alias of Int8 or Int16, each integer mapped to a String value |
| LowCardinality | LowCardinality(T) = two columns: Index T, Keys K where K is UInt8/16/32/64. Index contains unique values, Keys contains indexes into Index |
| Bool | Alias of UInt8: 0=false, 1=true |
To encode [null, "", "hello", null, "world"]:
Values: ["", "", "hello", "", "world"] (len: 5)
Nulls: [ 1, 0, 0, 1, 0] (len: 5)
To encode ["Eko", "Eko", "Amadela", "Amadela", "Amadela", "Amadela"]:
Index: ["Eko", "Amadela"] (String)
Keys: [0, 0, 1, 1, 1, 1] (UInt8)