Instance block format

This page describes the binary format used by the module to save instances to files or databases.


Compressed container

A block is usually serialized as compressed data.

Compressed data starts with one byte. Depending on its value, what follows is different.

  • 0: no compression. Following bytes can be read as as block format directly. This is rarely used and could be for debugging.
  • 1: LZ4 compression. The next big-endian 32-bit unsigned integer is the size of the decompressed data, and following bytes are compressed data using LZ4 default parameters. This mode is used by default.

Binary data

This data uses big-endian.

In pseudo-code:

// Root structure
struct InstanceBlockData {
    // Version tag in case more stuff is added in the future
    uint8_t version = 0;
    // There can be up to 256 different layers in one block
    uint8_t layer_count;
    // To compress positions we need to know their range.
    // It's local to the block so we know it starts from zero.
    float position_range;
    LayerData layers[layer_count];
    // Magic number to signal the end of the data block
    uint32_t control_end = 0x900df00d;

struct LayerData {
    uint16_t id; // Identifies the type of instances (rocks, grass, pebbles, bushes etc)
    uint16_t count;
    // To be able to compress scale we must know its range
    float scale_min;
    float scale_max;
    // This tells which format instances of this layer use. For now I always use the same format,
    // But maybe some types of instances will need more, or less data?
    uint8_t format = 0;
    InstanceData data[count];

struct InstanceData {
    // Position is lossy-compressed based on the size of the block
    uint16_t x;
    uint16_t y;
    uint16_t z;
    // Scale is uniform and is lossy-compressed to 256 values
    uint8_t scale;
    // Rotation is a compressed quaternion
    uint8_t x;
    uint8_t y;
    uint8_t z;
    uint8_t w;