Storage

Column-oriented storage with typed arrays, null bitmaps, and a single-file format.

Column Types

Four concrete types, each backed by a native Go slice:

Type	Go backing	File
`Int64`	`[]int64`	int64.go
`Float64`	`[]float64`	float64.go
`String`	`[]string`	string.go
`Bool`	`[]bool`	bool.go

ColumnChunk

The fundamental unit of storage. Each chunk holds:

A typed array of values (one of the four types above)
A NullBitmap — bit-packed, 0 = not null, 1 = null
Stats — min, max, null count (used for predicate pushdown)

NullBitmap

Bit-packed representation where each bit tracks one row's null status. Uses uint64 words internally — 64 rows per word. Operations:

bitmap.Set(i)       // mark row i as null
bitmap.IsNull(i)    // check if row i is null
bitmap.NullCount()  // count nulls

RowGroup

A collection of ColumnChunks that share the same row count. The RowGroup is the unit of I/O — the execution engine reads one RowGroup at a time.

RowGroup (N rows) ├── ColumnChunk "name" [string × N] + bitmap ├── ColumnChunk "age" [int64 × N] + bitmap ├── ColumnChunk "city" [string × N] + bitmap └── ColumnChunk "salary" [float64× N] + bitmap

File Format

Single-file columnar format, similar to Parquet:

┌──────────────┐ │ Magic (8B) │ ├──────────────┤ │ RowGroup 0 │ │ RowGroup 1 │ │ ... │ ├──────────────┤ │ Footer │ ← column names, types, offsets ├──────────────┤ │ Magic (8B) │ └──────────────┘

The footer stores metadata (column names, types, byte offsets) so the reader can seek directly to any column without scanning the whole file.

Files

File	Role
`types.go`	Column type enum and interface
`chunk.go`	ColumnChunk struct and methods
`rowgroup.go`	RowGroup — collection of chunks
`bitmap.go`	Bit-packed null bitmap
`format.go`	File header, footer, metadata structs
`writer.go`	Serialize row groups to disk
`reader.go`	Deserialize row groups from disk