#include <stdlib.h>
#include <math.h>
#include <stdio.h>
#include <string.h>
#include <time.h>
#include <pthread.h>
#include <zlib.h>

#include "world.h"
#include "raylib.h"
#include "player.h"

// Chunk loader helper declaration needed before use in world_load_or_create_chunk.
static bool load_chunk_from_file(FILE* file, Chunk* chunk);

// Human-readable error for last chunk load failure (used for diagnostics)
static char CHUNK_LOAD_ERROR[256];

// forward declare helper to write players file (defined later)
static void write_players_file(World* world, void* player_ptr, const char* world_name);

// forward declare block query function used in lighting and visible block calculations
BlockType world_get_block_or_solid(World* world, int x, int y, int z);

// forward declare visible blocks updater (defined later, used in world_set_block)
static void chunk_update_visible_blocks_immediate(Chunk* chunk, World* world);

// ============================================================================
// IMPROVED SEEDED RANDOM NUMBER GENERATION & NOISE FUNCTIONS
// ============================================================================


// Seeded random number generator (xorshift64*)
static uint64_t seed_state = 1;

uint64_t hash_seed(uint64_t x, uint64_t y, uint64_t seed)
{
    x = x ^ seed;
    y = y ^ (seed >> 32);

    uint64_t h = x ^ y;
    h ^= (h >> 33);
    h *= 0xff51afd7ed558ccdUL;
    h ^= (h >> 33);

    return h;
}

// Improved value noise that returns smoother gradients
static float noise_value(float x, float z, uint64_t seed)
{
    int xi = (int)floorf(x);
    int zi = (int)floorf(z);

    uint64_t hash = hash_seed(xi, zi, seed);
    // Map hash to range [-1, 1] for better noise
    return 2.0f * ((float)(hash % 1000) / 1000.0f) - 1.0f;
}

// Improved Perlin-like noise with interpolation
static float perlin_noise(float x, float z, uint64_t seed)
{
    int xi = (int)floorf(x);
    int zi = (int)floorf(z);
    float xf = x - xi;
    float zf = z - zi;

    // Improved smoothstep (Perlin's version for smoother curves)
    // Regular smoothstep: 3t^2 - 2t^3
    // Improved smoothstep: 6t^5 - 15t^4 + 10t^3 (makes terrain much smoother)
    float u = xf * xf * xf * (xf * (xf * 6.0f - 15.0f) + 10.0f);
    float v = zf * zf * zf * (zf * (zf * 6.0f - 15.0f) + 10.0f);

    // Get corner values
    float n00 = noise_value((float)xi, (float)zi, seed);
    float n10 = noise_value((float)(xi + 1), (float)zi, seed);
    float n01 = noise_value((float)xi, (float)(zi + 1), seed);
    float n11 = noise_value((float)(xi + 1), (float)(zi + 1), seed);

    // Interpolate
    float nx0 = n00 * (1.0f - u) + n10 * u;
    float nx1 = n01 * (1.0f - u) + n11 * u;
    float result = nx0 * (1.0f - v) + nx1 * v;

    return result;
}

// Fractional Brownian Motion - multiple octaves for complex terrain
static float fbm_noise(float x, float z, int octaves, uint64_t seed)
{
    float value = 0.0f;
    float amplitude = 1.0f;
    float frequency = 1.0f;
    float max_value = 0.0f;

    for (int i = 0; i < octaves; i++) {
        value += amplitude * perlin_noise(x * frequency, z * frequency, seed + i);
        max_value += amplitude;
        amplitude *= 0.5f;
        frequency *= 2.0f;
    }

    return value / max_value;
}

// Generate terrain height at a world position using seeded noise
static float terrain_height_seeded(float x, float z, uint64_t seed)
{
    // Base terrain with multiple scales for much smoother transitions
    float height = 0.0f;

    // Very large scale features (continental features) - creates smooth rolling hills
    height += fbm_noise(x * 0.0008f, z * 0.0008f, 5, seed) * 20.0f;

    // Large scale features (mountains/valleys) with reduced amplitude for smoother blend
    height += fbm_noise(x * 0.002f, z * 0.002f, 6, seed + 1) * 16.0f;

    // Medium scale features (hills) - smoothly blended
    height += fbm_noise(x * 0.008f, z * 0.008f, 5, seed + 2) * 10.0f;

    // Medium-small scale features for land relief
    height += fbm_noise(x * 0.025f, z * 0.025f, 4, seed + 50) * 6.0f;

    // Small scale features (terrain detail) - only subtle variation to avoid 1-block stubs
    height += fbm_noise(x * 0.06f, z * 0.06f, 3, seed + 100) * 1.0f;

    // Base level at y=100
    height += 100.0f;

    // Clamp to reasonable heights
    if (height < 85.0f) height = 85.0f;
    if (height > 120.0f) height = 120.0f;

    return height;
}

// Smooth terrain height to reduce 1-block stubs while preserving land relief
static int smooth_terrain_height(int x, int z, uint64_t seed)
{
    // Sample this position and its 8 neighbors
    float heights[9];
    int idx = 0;

    for (int dx = -1; dx <= 1; dx++) {
        for (int dz = -1; dz <= 1; dz++) {
            heights[idx] = terrain_height_seeded((float)(x + dx), (float)(z + dz), seed);
            idx++;
        }
    }

    // Center is heights[4]
    float center = heights[4];

    // Count how many neighbors are significantly different (more than 2 blocks away)
    int isolated_count = 0;
    for (int i = 0; i < 9; i++) {
        if (i != 4) {
            if (fabsf(heights[i] - center) > 2.0f) {
                isolated_count++;
            }
        }
    }

    // If this position is isolated (8 neighbors different), smooth it toward their average
    if (isolated_count == 8) {
        float neighbor_sum = 0.0f;
        for (int i = 0; i < 9; i++) {
            if (i != 4) {
                neighbor_sum += heights[i];
            }
        }
        float neighbor_avg = neighbor_sum / 8.0f;
        center = center * 0.4f + neighbor_avg * 0.6f;  // Blend 40/60 toward neighbors
    }

    return (int)(center);
}

// Ridge-based noise for cave systems (simple 2D representation)
static float ridge_noise(float x, float z, uint64_t seed)
{
    float value = fbm_noise(x * 0.02f, z * 0.02f, 2, seed + 1000);
    float ridge = 1.0f - fabsf(2.0f * value - 1.0f);
    return ridge;
}

// 3D cave noise for generating connected cave systems with proper interpolation
static float cave_noise_3d(float x, float y, float z, uint64_t seed)
{
    int xi = (int)floorf(x);
    int yi = (int)floorf(y);
    int zi = (int)floorf(z);

    float xf = x - xi;
    float yf = y - yi;
    float zf = z - zi;

    // Improved smoothstep (Perlin's version for smoother cave surfaces)
    float u = xf * xf * xf * (xf * (xf * 6.0f - 15.0f) + 10.0f);
    float v = yf * yf * yf * (yf * (yf * 6.0f - 15.0f) + 10.0f);
    float w = zf * zf * zf * (zf * (zf * 6.0f - 15.0f) + 10.0f);

    // Get 8 corner values (cube corners)
    float c000 = (float)(hash_seed(xi, yi, zi + seed) % 1000) / 1000.0f;
    float c100 = (float)(hash_seed(xi + 1, yi, zi + seed) % 1000) / 1000.0f;
    float c010 = (float)(hash_seed(xi, yi + 1, zi + seed) % 1000) / 1000.0f;
    float c110 = (float)(hash_seed(xi + 1, yi + 1, zi + seed) % 1000) / 1000.0f;
    float c001 = (float)(hash_seed(xi, yi, zi + 1 + seed) % 1000) / 1000.0f;
    float c101 = (float)(hash_seed(xi + 1, yi, zi + 1 + seed) % 1000) / 1000.0f;
    float c011 = (float)(hash_seed(xi, yi + 1, zi + 1 + seed) % 1000) / 1000.0f;
    float c111 = (float)(hash_seed(xi + 1, yi + 1, zi + 1 + seed) % 1000) / 1000.0f;

    // Interpolate along x
    float c00 = c000 * (1.0f - u) + c100 * u;
    float c10 = c010 * (1.0f - u) + c110 * u;
    float c01 = c001 * (1.0f - u) + c101 * u;
    float c11 = c011 * (1.0f - u) + c111 * u;

    // Interpolate along y
    float c0 = c00 * (1.0f - v) + c10 * v;
    float c1 = c01 * (1.0f - v) + c11 * v;

    // Interpolate along z
    float result = c0 * (1.0f - w) + c1 * w;

    return result;
}

// Create and allocate a new infinite world with chunk system
World* world_create(void)
{
    World* world = (World*)malloc(sizeof(World));

    // Preallocate large chunk cache upfront to avoid realloc during gameplay
    // This prevents pointer invalidation when worker thread is accessing chunks
    world->chunk_cache.chunk_capacity = 4096;  // Pre-allocate space for 4096 chunks (very large)
    world->chunk_cache.chunks = (Chunk*)malloc(sizeof(Chunk) * world->chunk_cache.chunk_capacity);
    world->chunk_cache.chunk_count = 0;
    world->last_loaded_chunk_x = INT32_MAX;
    world->last_loaded_chunk_y = INT32_MAX;
    world->last_loaded_chunk_z = INT32_MAX;
    world->last_chunk_update_position = (Vector3){1000000000.0f, 1000000000.0f, 1000000000.0f};
    world->last_chunk_update_forward = (Vector3){1000000000.0f, 1000000000.0f, 1000000000.0f};

    // Initialize texture cache
    world->textures.textures_loaded = false;
    world->textures.grass_texture = (Texture2D){0};
    world->textures.dirt_texture = (Texture2D){0};
    world->textures.stone_texture = (Texture2D){0};

    strncpy(world->world_name, "default", sizeof(world->world_name) - 1);
    world->world_name[sizeof(world->world_name) - 1] = '\0';

    // Initialize seed to a random value if not set later
    world->seed = (uint64_t)time(NULL);
    world->compress_chunk_files = true;

    // Initialize worker thread system
    pthread_mutex_init(&world->cache_mutex, NULL);  // Initialize cache mutex before worker starts
    worker_init(world);

    // No player attached initially
    world->current_player = NULL;
    // Default player nickname
    strncpy(world->player_nickname, "Player", sizeof(world->player_nickname)-1);
    world->player_nickname[sizeof(world->player_nickname)-1] = '\0';

    return world;
}

// Free world memory and all chunks
void world_free(World* world)
{
    if (world) {
        // Shutdown worker thread first
        worker_shutdown(world);

        // Clean up all remaining chunks
        for (int i = 0; i < world->chunk_cache.chunk_count; i++) {
            Chunk* chunk = &world->chunk_cache.chunks[i];
            chunk_free_visible_blocks(chunk);  // Free any cached mesh data
            chunk_free_merged_mesh(chunk);  // Free merged mesh buffers
            // NOTE: Don't destroy mutexes - they're part of preallocated array memory
            // They'll be reused when chunks are respawned or cleaned up
        }

        // Free chunk cache array and cache mutex
        if (world->chunk_cache.chunks) {
            free(world->chunk_cache.chunks);
        }
        pthread_mutex_destroy(&world->cache_mutex);  // Destroy cache access mutex

        // Don't unload textures - they're shared across all worlds
        // and will be unloaded when the application closes
        free(world);
    }
}

// Load textures for blocks from ./assets/textures/blocks/
void world_load_textures(World* world)
{
    if (!world || world->textures.textures_loaded) return;

    // Try to load grass texture
    world->textures.grass_texture = LoadTexture("./assets/textures/blocks/grass.png");
    printf("[textures] grass texture id: %d\n", world->textures.grass_texture.id);

    // Try to load dirt texture
    world->textures.dirt_texture = LoadTexture("./assets/textures/blocks/dirt.png");
    printf("[textures] dirt texture id: %d\n", world->textures.dirt_texture.id);

    // Try to load stone texture
    world->textures.stone_texture = LoadTexture("./assets/textures/blocks/stone.png");
    printf("[textures] stone texture id: %d\n", world->textures.stone_texture.id);

    // Try to load sand texture
    world->textures.sand_texture = LoadTexture("./assets/textures/blocks/sand.png");
    printf("[textures] sand texture id: %d\n", world->textures.sand_texture.id);

    // Try to load wood texture
    world->textures.wood_texture = LoadTexture("./assets/textures/blocks/wood.png");
    printf("[textures] wood texture id: %d\n", world->textures.wood_texture.id);

    // Try to load bedrock texture
    world->textures.bedrock_texture = LoadTexture("./assets/textures/blocks/stone.png");  // Using stone texture as fallback
    printf("[textures] bedrock texture id: %d\n", world->textures.bedrock_texture.id);

    world->textures.textures_loaded = true;
    printf("[textures] Block textures loaded\n");
}

// Unload block textures
void world_unload_textures(World* world)
{
    if (!world || !world->textures.textures_loaded) return;

    UnloadTexture(world->textures.grass_texture);
    UnloadTexture(world->textures.dirt_texture);
    UnloadTexture(world->textures.stone_texture);
    UnloadTexture(world->textures.sand_texture);
    UnloadTexture(world->textures.wood_texture);
    UnloadTexture(world->textures.bedrock_texture);

    world->textures.textures_loaded = false;
}

// Get texture for a block type
Texture2D world_get_block_texture(World* world, BlockType type)
{
    if (!world || !world->textures.textures_loaded) {
        // Return invalid texture if not loaded
        return (Texture2D){0};
    }

    switch (type) {
        case BLOCK_GRASS:
            return world->textures.grass_texture;
        case BLOCK_DIRT:
            return world->textures.dirt_texture;
        case BLOCK_STONE:
            return world->textures.stone_texture;
        case BLOCK_SAND:
            return world->textures.sand_texture;
        case BLOCK_WOOD:
            return world->textures.wood_texture;
        case BLOCK_BEDROCK:
            return world->textures.bedrock_texture;
        case BLOCK_AIR:
        default:
            return (Texture2D){0};
    }
}

// Find or create a chunk in the cache
Chunk* world_get_chunk(World* world, int32_t chunk_x, int32_t chunk_y, int32_t chunk_z)
{
    if (!world) return NULL;

    // Search for existing chunk
    for (int i = 0; i < world->chunk_cache.chunk_count; i++) {
        Chunk* c = &world->chunk_cache.chunks[i];
        if (c->chunk_x == chunk_x && c->chunk_y == chunk_y && c->chunk_z == chunk_z) {
            return c;
        }
    }

    return NULL;
}

// Load or create a chunk
Chunk* world_load_or_create_chunk(World* world, int32_t chunk_x, int32_t chunk_y, int32_t chunk_z)
{
    // Try to find existing chunk
    Chunk* existing = world_get_chunk(world, chunk_x, chunk_y, chunk_z);
    if (existing) {
        return existing;
    }

    // Expand chunk cache if needed
    if (world->chunk_cache.chunk_count >= world->chunk_cache.chunk_capacity) {
        fprintf(stderr, "[ERROR] Chunk cache overflow! count=%d, capacity=%d. Preallocated buffer was too small!\n",
                world->chunk_cache.chunk_count, world->chunk_cache.chunk_capacity);
        return NULL;  // Fail instead of reallocating - we should have preallocated enough
    }

    // Create new chunk
    Chunk* new_chunk = &world->chunk_cache.chunks[world->chunk_cache.chunk_count++];
    new_chunk->chunk_x = chunk_x;
    new_chunk->chunk_y = chunk_y;
    new_chunk->chunk_z = chunk_z;
    new_chunk->loaded = false;
    new_chunk->generated = false;
    new_chunk->modified = false;  // Not modified when first created
    new_chunk->needs_relighting = true;  // New chunks need lighting calculation
    new_chunk->meshed = false;
    new_chunk->pending_save = false;
    new_chunk->pending_unload = false;
    new_chunk->in_use_count = 0;
    // Initialize double-buffered lighting (avoid tearing during worker updates)
    new_chunk->active_light_buffer = 0;
    pthread_mutex_init(&new_chunk->light_swap_mutex, NULL);

    // Initialize double-buffered visible blocks cache
    new_chunk->visible_blocks[0] = NULL;
    new_chunk->visible_blocks[1] = NULL;
    new_chunk->visible_count[0] = 0;
    new_chunk->visible_count[1] = 0;
    new_chunk->visible_capacity[0] = 0;
    new_chunk->visible_capacity[1] = 0;
    new_chunk->active_mesh = 0;  // Start with buffer 0

    // Initialize merged mesh pointers for greedy meshing
    new_chunk->merged_mesh[0] = NULL;
    new_chunk->merged_mesh[1] = NULL;
    new_chunk->active_merged_mesh = 0;

    pthread_mutex_init(&new_chunk->mesh_swap_mutex, NULL);  // Mutex for atomic mesh swaps
    pthread_mutex_init(&new_chunk->mutex, NULL);  // Initialize chunk mutex

    // Initialize blocks to air and lighting to 0
    for (int y = 0; y < CHUNK_HEIGHT; y++) {
        for (int z = 0; z < CHUNK_DEPTH; z++) {
            for (int x = 0; x < CHUNK_WIDTH; x++) {
                new_chunk->blocks[y][z][x].type = BLOCK_AIR;
                // Initialize both lighting buffers to prevent garbage data
                new_chunk->skylight[0][y][z][x] = 0;
                new_chunk->skylight[1][y][z][x] = 0;
                new_chunk->blocklight[0][y][z][x] = 0;
                new_chunk->blocklight[1][y][z][x] = 0;
            }
        }
    }

    // Try to load from disk
    char filepath[512];
    snprintf(filepath, sizeof(filepath), "./worlds/%s/chunks/chunk_%d_%d_%d.chunk",
             world->world_name, chunk_x, chunk_y, chunk_z);

    FILE* file = fopen(filepath, "rb");
    if (file) {
        bool load_success = true;

        // Detect file format and load accordingly, preserving compatibility with legacy raw format.
        if (fseek(file, 0, SEEK_SET) != 0) {
            load_success = false;
        } else {
            uint8_t magic[4];
            if (fread(magic, 1, sizeof(magic), file) == sizeof(magic) &&
                memcmp(magic, "B3DV", sizeof(magic)) == 0) {
                // Rewind to position 0 before passing to load_chunk_from_file (it expects to read from the start)
                if (fseek(file, 0, SEEK_SET) != 0) {
                    load_success = false;
                } else if (!load_chunk_from_file(file, new_chunk)) {
                    // Provide diagnostic from loader for why parsing failed
                    printf("[chunk_load] Failed to parse chunk file: %s (%s)\n", filepath, CHUNK_LOAD_ERROR);
                    load_success = false;
                }
            } else {
                // Legacy raw format: rewind and read BlockType values directly.
                if (fseek(file, 0, SEEK_SET) != 0) {
                    load_success = false;
                } else {
                    for (int y = 0; y < CHUNK_HEIGHT && load_success; y++) {
                        for (int z = 0; z < CHUNK_DEPTH && load_success; z++) {
                            for (int x = 0; x < CHUNK_WIDTH; x++) {
                                BlockType block_type;
                                if (fread(&block_type, sizeof(BlockType), 1, file) == 1) {
                                    new_chunk->blocks[y][z][x].type = block_type;
                                } else {
                                    load_success = false;
                                    break;
                                }
                            }
                        }
                    }
                }
            }
        }

        fclose(file);
        if (load_success) {
            new_chunk->loaded = true;
            new_chunk->generated = true;  // Loaded chunks are already complete
            new_chunk->modified = false;  // Not modified when loaded from disk
            new_chunk->needs_relighting = true;  // Recalculate lighting - arrays were not saved to disk
            printf("[chunk_load] Loaded chunk from %s\n", filepath);
            // Queue for lighting and meshing
            worker_queue_chunk(world, new_chunk);
        } else {
            printf("[chunk_load] Failed to parse chunk file: %s\n", filepath);
        }
    } else {
        // Chunk doesn't exist on disk - don't auto-generate, return empty chunk
        // The caller (world_load) will handle generation if needed
        printf("[chunk_load] Chunk not found: %s (will stay as air)\n", filepath);
    }

    return new_chunk;
}

// Generate a chunk procedurally with improved terrain
void world_generate_chunk(Chunk* chunk, uint64_t seed)
{
    if (!chunk) return;

    // Generate terrain with improved noise and features
    for (int x = 0; x < CHUNK_WIDTH; x++) {
        for (int z = 0; z < CHUNK_DEPTH; z++) {
            // Calculate world position
            int world_x = chunk->chunk_x * CHUNK_WIDTH + x;
            int world_z = chunk->chunk_z * CHUNK_DEPTH + z;

            // Get height at this position using improved noise function with smoothing
            // Use world seed directly to maintain terrain continuity across chunks
            int terrain_height_blocks = smooth_terrain_height(world_x, world_z, seed);

            // Generate vertical column
            for (int y = 0; y < CHUNK_HEIGHT; y++) {
                int world_y = chunk->chunk_y * CHUNK_HEIGHT + y;
                BlockType block_type = BLOCK_AIR;

                // Bedrock layer at y=0
                if (world_y == 0) {
                    block_type = BLOCK_BEDROCK;
                }
                // Depth limit: no blocks below bedrock
                else if (world_y < 0) {
                    block_type = BLOCK_AIR;
                }
                // Fill blocks below terrain height
                else if (world_y < terrain_height_blocks) {
                    // Underground cave systems - use 3D noise for connected caverns
                    // Tiered cave generation for natural progression
                    bool is_cave = false;

                    // Big caves from y=15 to y=40
                    if (world_y >= 15 && world_y < 40) {
                        float cave_val = 0.0f;
                        float cave_amp = 1.0f;

                        // Layer 1: Large cave chambers
                        cave_val += cave_amp * cave_noise_3d(
                            (float)world_x * 0.03f,
                            (float)world_y * 0.03f,
                            (float)world_z * 0.03f,
                            seed + 5000
                        );
                        cave_amp *= 0.7f;

                        // Layer 2: Medium cave corridors
                        cave_val += cave_amp * cave_noise_3d(
                            (float)world_x * 0.1f,
                            (float)world_y * 0.1f,
                            (float)world_z * 0.1f,
                            seed + 5001
                        );
                        cave_amp *= 0.6f;

                        // Layer 3: Small cave tunnels
                        cave_val += cave_amp * cave_noise_3d(
                            (float)world_x * 0.25f,
                            (float)world_y * 0.25f,
                            (float)world_z * 0.25f,
                            seed + 5002
                        );

                        cave_val /= 2.3f;

                        // Lower threshold for bigger, more connected caves
                        is_cave = cave_val < 0.42f;
                    }
                    // Smaller caves from y=40 to y=85
                    else if (world_y >= 40 && world_y < 85) {
                        float cave_val = 0.0f;
                        float cave_amp = 1.0f;

                        // Layer 1: Smaller chambers
                        cave_val += cave_amp * cave_noise_3d(
                            (float)world_x * 0.05f,
                            (float)world_y * 0.05f,
                            (float)world_z * 0.05f,
                            seed + 5010
                        );
                        cave_amp *= 0.7f;

                        // Layer 2: Smaller corridors
                        cave_val += cave_amp * cave_noise_3d(
                            (float)world_x * 0.15f,
                            (float)world_y * 0.15f,
                            (float)world_z * 0.15f,
                            seed + 5011
                        );
                        cave_amp *= 0.6f;

                        // Layer 3: Fine tunnels
                        cave_val += cave_amp * cave_noise_3d(
                            (float)world_x * 0.35f,
                            (float)world_y * 0.35f,
                            (float)world_z * 0.35f,
                            seed + 5012
                        );

                        cave_val /= 2.3f;

                        // Higher threshold for smaller, more fragmented caves
                        is_cave = cave_val < 0.45f;
                    }

                    if (is_cave) {
                        block_type = BLOCK_AIR;
                    }
                    // Top surface block is grass
                    else if (world_y == terrain_height_blocks - 1) {
                        block_type = BLOCK_GRASS;
                    }
                    // Few blocks below grass are dirt
                    else if (world_y > terrain_height_blocks - 4 && world_y < terrain_height_blocks - 1) {
                        block_type = BLOCK_DIRT;
                    }
                    // Deep dirt layer
                    else if (world_y > terrain_height_blocks - 7) {
                        block_type = BLOCK_DIRT;
                    }
                    // Everything else is stone
                    else {
                        block_type = BLOCK_STONE;
                    }
                } else {
                    block_type = BLOCK_AIR;
                }

                chunk->blocks[y][z][x].type = block_type;
            }
        }
    }

    chunk->generated = true;
}

// Set block at world position
void world_set_block(World* world, int x, int y, int z, BlockType type)
{
    // Calculate chunk coordinates
    int32_t chunk_x = x < 0 ? (x - CHUNK_WIDTH + 1) / CHUNK_WIDTH : x / CHUNK_WIDTH;
    int32_t chunk_y = y < 0 ? (y - CHUNK_HEIGHT + 1) / CHUNK_HEIGHT : y / CHUNK_HEIGHT;
    int32_t chunk_z = z < 0 ? (z - CHUNK_DEPTH + 1) / CHUNK_DEPTH : z / CHUNK_DEPTH;

    // Calculate position within chunk
    int local_x = x - (chunk_x * CHUNK_WIDTH);
    int local_y = y - (chunk_y * CHUNK_HEIGHT);
    int local_z = z - (chunk_z * CHUNK_DEPTH);

    // CRITICAL: Lock cache while accessing/modifying chunk cache
    pthread_mutex_lock(&world->cache_mutex);

    // Get or create chunk
    Chunk* chunk = world_load_or_create_chunk(world, chunk_x, chunk_y, chunk_z);
    if (chunk) {
        // Lock chunk while modifying blocks and invalidating cache
        pthread_mutex_lock(&chunk->mutex);

        // Get old block to check if lighting is affected
        BlockType old_block = world_chunk_get_block(chunk, local_x, local_y, local_z);
        BlockProperties old_props = get_block_properties(old_block);
        BlockProperties new_props = get_block_properties(type);

        // Check if this block change affects light propagation
        // Light is affected if: emission changed, opacity changed, or air<->solid transition
        bool affects_light = (old_props.emission != new_props.emission) ||
                            (old_props.opacity != new_props.opacity) ||
                            (old_block == BLOCK_AIR) != (type == BLOCK_AIR);

        world_chunk_set_block(chunk, local_x, local_y, local_z, type);

        // Mark chunk as needing relighting if block change affects light
        if (affects_light) {
            chunk->needs_relighting = true;
        }
        // Always mark meshed=false so worker knows to rebuild geometry
        chunk->meshed = false;

        pthread_mutex_unlock(&chunk->mutex);
        pthread_mutex_unlock(&world->cache_mutex);

        // FULL LIGHTING RECALCULATION: Calculate complete lighting for affected chunks SYNCHRONOUSLY
        // This ensures newly exposed faces have 100% correct lighting instantly (no later correction needed)
        // STRATEGY: Update geometry instantly with current lighting, defer lighting recalc to worker
        // This avoids main-thread stutter while still getting fast visual updates
        // Trade-off: Lighting is slightly stale until worker catches up (usually <100ms)
        // Benefit: Main thread stays responsive, no frame stuttering
        const int neighbor_offsets[6][3] = {
            { 1, 0, 0 }, {-1, 0, 0},
            { 0, 1, 0 }, { 0,-1, 0},
            { 0, 0, 1 }, { 0, 0,-1}
        };

        if (affects_light) {
            // Don't calculate lighting on main thread - just mark for worker to process
            pthread_mutex_lock(&chunk->mutex);
            chunk->needs_relighting = true;
            pthread_mutex_unlock(&chunk->mutex);
            
            // Mark neighbors for relighting too
            for (int ni = 0; ni < 6; ni++) {
                if ((ni == 0 && local_x != CHUNK_WIDTH - 1) || (ni == 1 && local_x != 0) ||
                    (ni == 2 && local_y != CHUNK_HEIGHT - 1) || (ni == 3 && local_y != 0) ||
                    (ni == 4 && local_z != CHUNK_DEPTH - 1) || (ni == 5 && local_z != 0)) {
                    continue;
                }

                int32_t nx = chunk_x + neighbor_offsets[ni][0];
                int32_t ny = chunk_y + neighbor_offsets[ni][1];
                int32_t nz = chunk_z + neighbor_offsets[ni][2];

                pthread_mutex_lock(&world->cache_mutex);
                Chunk* neighbor = world_get_chunk(world, nx, ny, nz);
                pthread_mutex_unlock(&world->cache_mutex);
                
                if (!neighbor || !neighbor->loaded || !neighbor->generated) continue;
                
                pthread_mutex_lock(&neighbor->mutex);
                neighbor->needs_relighting = true;
                pthread_mutex_unlock(&neighbor->mutex);
            }
        }
        
        // Update visible blocks immediately with CURRENT lighting (not waiting for recalc)
        chunk_update_visible_blocks_immediate(chunk, world);
        
        // Mark as NOT meshed - worker will re-mesh with new lighting after recalc
        pthread_mutex_lock(&chunk->mutex);
        chunk->meshed = false;  // Worker needs to re-mesh with updated lighting
        chunk->needs_relighting = true;  // And it needs to recalculate lighting
        pthread_mutex_unlock(&chunk->mutex);

        // Queue edited chunk and neighbors to worker with priority for lighting recalc + re-mesh
        worker_queue_chunk_priority(world, chunk);
        
        // Queue neighbors to worker for async mesh/lighting update
        for (int ni = 0; ni < 6; ni++) {
            if ((ni == 0 && local_x != CHUNK_WIDTH - 1) || (ni == 1 && local_x != 0) ||
                (ni == 2 && local_y != CHUNK_HEIGHT - 1) || (ni == 3 && local_y != 0) ||
                (ni == 4 && local_z != CHUNK_DEPTH - 1) || (ni == 5 && local_z != 0)) {
                continue;
            }

            int32_t nx = chunk_x + neighbor_offsets[ni][0];
            int32_t ny = chunk_y + neighbor_offsets[ni][1];
            int32_t nz = chunk_z + neighbor_offsets[ni][2];

            pthread_mutex_lock(&world->cache_mutex);
            Chunk* neighbor = world_get_chunk(world, nx, ny, nz);
            pthread_mutex_unlock(&world->cache_mutex);
            
            if (!neighbor || !neighbor->loaded || !neighbor->generated) continue;

            worker_queue_chunk_priority(world, neighbor);
        }
    } else {
        pthread_mutex_unlock(&world->cache_mutex);
    }
}

// Helper: Immediately recalculate and update visible blocks for a chunk (main thread)
// This is called on main thread for affected chunks to get instant visual updates
// Does NOT use worker thread - executes immediately on main thread
static void chunk_update_visible_blocks_immediate(Chunk* chunk, World* world)
{
    if (!chunk) return;
    
    // Build new visible blocks list in temp array
    int temp_capacity = 1024;
    int temp_count = 0;
    CachedVisibleBlock* temp_blocks = (CachedVisibleBlock*)malloc(sizeof(CachedVisibleBlock) * temp_capacity);
    
    for (int y = 0; y < CHUNK_HEIGHT; y++) {
        for (int z = 0; z < CHUNK_DEPTH; z++) {
            for (int x = 0; x < CHUNK_WIDTH; x++) {
                BlockType block = world_chunk_get_block(chunk, x, y, z);
                if (block == BLOCK_AIR) continue;
                
                uint8_t exposed_faces = 0;
                
                // Interior blocks: check locally (no locks)
                if (x + 1 < CHUNK_WIDTH) {
                    if (world_chunk_get_block(chunk, x + 1, y, z) == BLOCK_AIR) exposed_faces |= (1 << 0);
                } else {
                    if (world_get_block(world, chunk->chunk_x * CHUNK_WIDTH + x + 1, chunk->chunk_y * CHUNK_HEIGHT + y, chunk->chunk_z * CHUNK_DEPTH + z) == BLOCK_AIR) exposed_faces |= (1 << 0);
                }
                
                if (x - 1 >= 0) {
                    if (world_chunk_get_block(chunk, x - 1, y, z) == BLOCK_AIR) exposed_faces |= (1 << 1);
                } else {
                    if (world_get_block(world, chunk->chunk_x * CHUNK_WIDTH + x - 1, chunk->chunk_y * CHUNK_HEIGHT + y, chunk->chunk_z * CHUNK_DEPTH + z) == BLOCK_AIR) exposed_faces |= (1 << 1);
                }
                
                if (y + 1 < CHUNK_HEIGHT) {
                    if (world_chunk_get_block(chunk, x, y + 1, z) == BLOCK_AIR) exposed_faces |= (1 << 2);
                } else {
                    if (world_get_block(world, chunk->chunk_x * CHUNK_WIDTH + x, chunk->chunk_y * CHUNK_HEIGHT + y + 1, chunk->chunk_z * CHUNK_DEPTH + z) == BLOCK_AIR) exposed_faces |= (1 << 2);
                }
                
                if (y - 1 >= 0) {
                    if (world_chunk_get_block(chunk, x, y - 1, z) == BLOCK_AIR) exposed_faces |= (1 << 3);
                } else {
                    if (world_get_block(world, chunk->chunk_x * CHUNK_WIDTH + x, chunk->chunk_y * CHUNK_HEIGHT + y - 1, chunk->chunk_z * CHUNK_DEPTH + z) == BLOCK_AIR) exposed_faces |= (1 << 3);
                }
                
                if (z + 1 < CHUNK_DEPTH) {
                    if (world_chunk_get_block(chunk, x, y, z + 1) == BLOCK_AIR) exposed_faces |= (1 << 4);
                } else {
                    if (world_get_block(world, chunk->chunk_x * CHUNK_WIDTH + x, chunk->chunk_y * CHUNK_HEIGHT + y, chunk->chunk_z * CHUNK_DEPTH + z + 1) == BLOCK_AIR) exposed_faces |= (1 << 4);
                }
                
                if (z - 1 >= 0) {
                    if (world_chunk_get_block(chunk, x, y, z - 1) == BLOCK_AIR) exposed_faces |= (1 << 5);
                } else {
                    if (world_get_block(world, chunk->chunk_x * CHUNK_WIDTH + x, chunk->chunk_y * CHUNK_HEIGHT + y, chunk->chunk_z * CHUNK_DEPTH + z - 1) == BLOCK_AIR) exposed_faces |= (1 << 5);
                }
                
                if (exposed_faces != 0) {
                    if (temp_count >= temp_capacity) {
                        temp_capacity *= 2;
                        temp_blocks = (CachedVisibleBlock*)realloc(temp_blocks, sizeof(CachedVisibleBlock) * temp_capacity);
                    }
                    
                    int world_x = chunk->chunk_x * CHUNK_WIDTH + x;
                    int world_y = chunk->chunk_y * CHUNK_HEIGHT + y;
                    int world_z = chunk->chunk_z * CHUNK_DEPTH + z;
                    
                    temp_blocks[temp_count].x = x;
                    temp_blocks[temp_count].y = y;
                    temp_blocks[temp_count].z = z;
                    temp_blocks[temp_count].exposed_faces = exposed_faces;
                    
                    // Pre-calculate per-face lighting to prevent blinking
                    // Use current lighting buffers (fast lookup, no recalculation)
                    for (int face = 0; face < 6; face++) {
                        temp_blocks[temp_count].face_light[face] = 0;
                    }
                    
                    // +X face lighting
                    if (exposed_faces & (1 << 0)) {
                        uint8_t skyl = world_get_skylight(world, world_x + 1, world_y, world_z);
                        uint8_t blockl = world_get_blocklight(world, world_x + 1, world_y, world_z);
                        temp_blocks[temp_count].face_light[0] = (skyl > blockl) ? skyl : blockl;
                    }
                    // -X face lighting
                    if (exposed_faces & (1 << 1)) {
                        uint8_t skyl = world_get_skylight(world, world_x - 1, world_y, world_z);
                        uint8_t blockl = world_get_blocklight(world, world_x - 1, world_y, world_z);
                        temp_blocks[temp_count].face_light[1] = (skyl > blockl) ? skyl : blockl;
                    }
                    // +Y face lighting
                    if (exposed_faces & (1 << 2)) {
                        uint8_t skyl = world_get_skylight(world, world_x, world_y + 1, world_z);
                        uint8_t blockl = world_get_blocklight(world, world_x, world_y + 1, world_z);
                        temp_blocks[temp_count].face_light[2] = (skyl > blockl) ? skyl : blockl;
                    }
                    // -Y face lighting
                    if (exposed_faces & (1 << 3)) {
                        uint8_t skyl = world_get_skylight(world, world_x, world_y - 1, world_z);
                        uint8_t blockl = world_get_blocklight(world, world_x, world_y - 1, world_z);
                        temp_blocks[temp_count].face_light[3] = (skyl > blockl) ? skyl : blockl;
                    }
                    // +Z face lighting
                    if (exposed_faces & (1 << 4)) {
                        uint8_t skyl = world_get_skylight(world, world_x, world_y, world_z + 1);
                        uint8_t blockl = world_get_blocklight(world, world_x, world_y, world_z + 1);
                        temp_blocks[temp_count].face_light[4] = (skyl > blockl) ? skyl : blockl;
                    }
                    // -Z face lighting
                    if (exposed_faces & (1 << 5)) {
                        uint8_t skyl = world_get_skylight(world, world_x, world_y, world_z - 1);
                        uint8_t blockl = world_get_blocklight(world, world_x, world_y, world_z - 1);
                        temp_blocks[temp_count].face_light[5] = (skyl > blockl) ? skyl : blockl;
                    }
                    
                    temp_count++;
                }
            }
        }
    }
    
    // Swap the visible blocks buffer atomically
    pthread_mutex_lock(&chunk->mesh_swap_mutex);
    int current_mesh = __atomic_load_n(&chunk->active_mesh, __ATOMIC_ACQUIRE);
    int target_mesh = 1 - current_mesh;
    if (chunk->visible_blocks[target_mesh] != NULL) {
        free(chunk->visible_blocks[target_mesh]);
    }
    chunk->visible_blocks[target_mesh] = temp_blocks;
    chunk->visible_count[target_mesh] = temp_count;
    chunk->visible_capacity[target_mesh] = temp_capacity;
    __atomic_store_n(&chunk->active_mesh, target_mesh, __ATOMIC_RELEASE);
    pthread_mutex_unlock(&chunk->mesh_swap_mutex);
}

// Fast update: immediately recalculate visible faces for a block and its 6 neighbors
// Called right after setting a block to show face updates without waiting for worker thread
void world_update_nearby_visible_faces(World* world, int x, int y, int z)
{
    // For the edited block and its 6 neighbors, recalculate which faces are visible
    // This doesn't regenerate the mesh, just updates the cached visible face info
    // The full mesh will regenerate from the worker thread later
    
    int block_offsets[7][3] = {
        {0, 0, 0},      // The edited block itself
        {1, 0, 0}, {-1, 0, 0},  // X neighbors
        {0, 1, 0}, {0, -1, 0},  // Y neighbors
        {0, 0, 1}, {0, 0, -1}   // Z neighbors
    };
    
    for (int b = 0; b < 7; b++) {
        int nx = x + block_offsets[b][0];
        int ny = y + block_offsets[b][1];
        int nz = z + block_offsets[b][2];
        
        // Calculate chunk and local coordinates
        int32_t chunk_x = nx < 0 ? (nx - CHUNK_WIDTH + 1) / CHUNK_WIDTH : nx / CHUNK_WIDTH;
        int32_t chunk_y = ny < 0 ? (ny - CHUNK_HEIGHT + 1) / CHUNK_HEIGHT : ny / CHUNK_HEIGHT;
        int32_t chunk_z = nz < 0 ? (nz - CHUNK_DEPTH + 1) / CHUNK_DEPTH : nz / CHUNK_DEPTH;
        
        int local_x = nx - (chunk_x * CHUNK_WIDTH);
        int local_y = ny - (chunk_y * CHUNK_HEIGHT);
        int local_z = nz - (chunk_z * CHUNK_DEPTH);
        
        pthread_mutex_lock(&world->cache_mutex);
        Chunk* chunk = world_get_chunk(world, chunk_x, chunk_y, chunk_z);
        if (!chunk || !chunk->loaded) {
            pthread_mutex_unlock(&world->cache_mutex);
            continue;
        }
        pthread_mutex_unlock(&world->cache_mutex);
        
        BlockType block = world_chunk_get_block(chunk, local_x, local_y, local_z);
        if (block == BLOCK_AIR) continue;  // Skip air blocks
        
        // Check which faces are exposed
        uint8_t exposed_faces = 0;
        if (world_get_block(world, nx + 1, ny, nz) == BLOCK_AIR) exposed_faces |= (1 << 0);
        if (world_get_block(world, nx - 1, ny, nz) == BLOCK_AIR) exposed_faces |= (1 << 1);
        if (world_get_block(world, nx, ny + 1, nz) == BLOCK_AIR) exposed_faces |= (1 << 2);
        if (world_get_block(world, nx, ny - 1, nz) == BLOCK_AIR) exposed_faces |= (1 << 3);
        if (world_get_block(world, nx, ny, nz + 1) == BLOCK_AIR) exposed_faces |= (1 << 4);
        if (world_get_block(world, nx, ny, nz - 1) == BLOCK_AIR) exposed_faces |= (1 << 5);
        
        // If this block has exposed faces, find it in the visible_blocks array
        // and update its exposed_faces mask
        pthread_mutex_lock(&chunk->mutex);
        
        int current_buffer = __atomic_load_n(&chunk->active_mesh, __ATOMIC_ACQUIRE);
        CachedVisibleBlock* visible_blocks = chunk->visible_blocks[current_buffer];
        int visible_count = chunk->visible_count[current_buffer];
        
        // Search for this block in the visible list
        int block_index = -1;
        for (int i = 0; i < visible_count; i++) {
            if (visible_blocks[i].x == local_x && 
                visible_blocks[i].y == local_y && 
                visible_blocks[i].z == local_z) {
                block_index = i;
                break;
            }
        }
        
        if (exposed_faces != 0) {
            if (block_index >= 0) {
                // Block already in list, just update its exposed faces
                visible_blocks[block_index].exposed_faces = exposed_faces;
            }
            // If not in list, worker will add it during full rebuild
        } else {
            if (block_index >= 0) {
                // Block now has no exposed faces, remove it from visible list
                // Shift remaining blocks down
                for (int i = block_index; i < visible_count - 1; i++) {
                    visible_blocks[i] = visible_blocks[i + 1];
                }
                chunk->visible_count[current_buffer]--;
            }
        }
        
        pthread_mutex_unlock(&chunk->mutex);
    }
}


// Get block at world position
BlockType world_get_block(World* world, int x, int y, int z)
{
    // Calculate chunk coordinates
    int32_t chunk_x = x < 0 ? (x - CHUNK_WIDTH + 1) / CHUNK_WIDTH : x / CHUNK_WIDTH;
    int32_t chunk_y = y < 0 ? (y - CHUNK_HEIGHT + 1) / CHUNK_HEIGHT : y / CHUNK_HEIGHT;
    int32_t chunk_z = z < 0 ? (z - CHUNK_DEPTH + 1) / CHUNK_DEPTH : z / CHUNK_DEPTH;

    // Calculate position within chunk
    int local_x = x - (chunk_x * CHUNK_WIDTH);
    int local_y = y - (chunk_y * CHUNK_HEIGHT);
    int local_z = z - (chunk_z * CHUNK_DEPTH);

    // CRITICAL: Lock cache before accessing chunk to prevent concurrent unloading
    pthread_mutex_lock(&world->cache_mutex);
    Chunk* chunk = world_get_chunk(world, chunk_x, chunk_y, chunk_z);
    if (!chunk) {
        pthread_mutex_unlock(&world->cache_mutex);
        return BLOCK_AIR;  // Unloaded chunks are treated as air
    }

    BlockType result = world_chunk_get_block(chunk, local_x, local_y, local_z);
    pthread_mutex_unlock(&world->cache_mutex);
    return result;
}

// Get block, treating unloaded chunks as STONE (for lighting calculations)
// This prevents false positives where light penetrates through unloaded chunks
BlockType world_get_block_or_solid(World* world, int x, int y, int z)
{
    // Calculate chunk coordinates
    int32_t chunk_x = x < 0 ? (x - CHUNK_WIDTH + 1) / CHUNK_WIDTH : x / CHUNK_WIDTH;
    int32_t chunk_y = y < 0 ? (y - CHUNK_HEIGHT + 1) / CHUNK_HEIGHT : y / CHUNK_HEIGHT;
    int32_t chunk_z = z < 0 ? (z - CHUNK_DEPTH + 1) / CHUNK_DEPTH : z / CHUNK_DEPTH;

    // Calculate position within chunk
    int local_x = x - (chunk_x * CHUNK_WIDTH);
    int local_y = y - (chunk_y * CHUNK_HEIGHT);
    int local_z = z - (chunk_z * CHUNK_DEPTH);

    // Lock cache before accessing chunk
    pthread_mutex_lock(&world->cache_mutex);
    Chunk* chunk = world_get_chunk(world, chunk_x, chunk_y, chunk_z);
    if (!chunk || !chunk->loaded) {
        pthread_mutex_unlock(&world->cache_mutex);
        return BLOCK_STONE;  // Unloaded chunks are treated as solid (prevents false light propagation)
    }

    BlockType result = world_chunk_get_block(chunk, local_x, local_y, local_z);
    pthread_mutex_unlock(&world->cache_mutex);
    return result;
}

// Set block within a chunk
void world_chunk_set_block(Chunk* chunk, int x, int y, int z, BlockType type)
{
    if (x >= 0 && x < CHUNK_WIDTH &&
        y >= 0 && y < CHUNK_HEIGHT &&
        z >= 0 && z < CHUNK_DEPTH) {
        chunk->blocks[y][z][x].type = type;
        chunk->modified = true;  // Mark chunk as modified when block changes
    }
}

// Get block within a chunk
BlockType world_chunk_get_block(Chunk* chunk, int x, int y, int z)
{
    if (x >= 0 && x < CHUNK_WIDTH &&
        y >= 0 && y < CHUNK_HEIGHT &&
        z >= 0 && z < CHUNK_DEPTH) {
        return chunk->blocks[y][z][x].type;
    }
    return BLOCK_AIR;
}

// Get color for block type
Color world_get_block_color(BlockType type)
{
    switch (type) {
        case BLOCK_GRASS:
            return (Color){34, 139, 34, 255};  // Forest Green
        case BLOCK_DIRT:
            return (Color){139, 69, 19, 255};  // Saddle Brown
        case BLOCK_STONE:
            return (Color){128, 128, 128, 255};  // Grey
        case BLOCK_SAND:
            return (Color){238, 214, 175, 255};  // Sandy Brown
        case BLOCK_WOOD:
            return (Color){101, 67, 33, 255};  // Dark Brown
        case BLOCK_BEDROCK:
            return (Color){64, 64, 64, 255};  // Dark Grey (Bedrock)
        case BLOCK_GLOWSTONE:
            return (Color){255, 255, 200, 255};  // Bright warm white
        case BLOCK_AIR:
        default:
            return (Color){0, 0, 0, 0};  // Transparent
    }
}

// Generate a simple prism world - now just generates starting area with chunks
void world_generate_prism(World* world)
{
    // Generate only the chunk at origin for player spawn
    Chunk* chunk = world_load_or_create_chunk(world, 0, 0, 0);
    if (chunk && !chunk->generated) {
        world_generate_chunk(chunk, world->seed);
        chunk->loaded = true;
        chunk->generated = true;
        worker_queue_chunk(world, chunk);  // Queue for worker to calculate lighting and mesh
        chunk->modified = false;  // Freshly generated chunk is not modified
    }
}

// Update loaded chunks based on player position and camera direction
void world_update_chunks(World* world, Vector3 player_pos, Vector3 camera_forward, float render_distance_blocks)
{
    if (!world) return;

    // Calculate player's chunk coordinates
    int32_t player_chunk_x = (int32_t)floorf(player_pos.x / CHUNK_WIDTH);
    int32_t player_chunk_y = (int32_t)floorf(player_pos.y / CHUNK_HEIGHT);
    int32_t player_chunk_z = (int32_t)floorf(player_pos.z / CHUNK_DEPTH);

    bool first_update = (world->last_chunk_update_position.x > 999999999.0f ||
                         world->last_chunk_update_forward.x > 999999999.0f);

    if (!first_update) {
        float dx = player_pos.x - world->last_chunk_update_position.x;
        float dy = player_pos.y - world->last_chunk_update_position.y;
        float dz = player_pos.z - world->last_chunk_update_position.z;
        float move_sq = dx*dx + dy*dy + dz*dz;

        float forward_dot = camera_forward.x * world->last_chunk_update_forward.x +
                            camera_forward.y * world->last_chunk_update_forward.y +
                            camera_forward.z * world->last_chunk_update_forward.z;

        if (player_chunk_x == world->last_loaded_chunk_x &&
            player_chunk_y == world->last_loaded_chunk_y &&
            player_chunk_z == world->last_loaded_chunk_z &&
            forward_dot > 0.999f &&
            move_sq < 1.0f) {
            return;
        }
    }

    world->last_loaded_chunk_x = player_chunk_x;
    world->last_loaded_chunk_y = player_chunk_y;
    world->last_loaded_chunk_z = player_chunk_z;
    world->last_chunk_update_position = player_pos;
    world->last_chunk_update_forward = camera_forward;

    // CRITICAL: Lock cache mutex while loading/creating chunks to prevent races with unload
    pthread_mutex_lock(&world->cache_mutex);

    // Load chunks within load distance, prioritizing forward direction
    // Compute chunk load distance from desired render distance in blocks
    int load_dist = (int)ceilf(render_distance_blocks / (float)CHUNK_WIDTH);
    if (load_dist < 1) load_dist = 1;

    float camera_forward_xz_len = sqrtf(camera_forward.x * camera_forward.x + camera_forward.z * camera_forward.z);
    bool has_horizontal_forward = camera_forward_xz_len > 1e-6f;
    float camera_forward_xz_norm_x = 0.0f;
    float camera_forward_xz_norm_z = 0.0f;
    if (has_horizontal_forward) {
        camera_forward_xz_norm_x = camera_forward.x / camera_forward_xz_len;
        camera_forward_xz_norm_z = camera_forward.z / camera_forward_xz_len;
    }

    for (int cx = player_chunk_x - load_dist; cx <= player_chunk_x + load_dist; cx++) {
        for (int cy = player_chunk_y - load_dist; cy <= player_chunk_y + load_dist; cy++) {
            for (int cz = player_chunk_z - load_dist; cz <= player_chunk_z + load_dist; cz++) {
                // Calculate chunk center relative to player
                float chunk_center_x = cx * CHUNK_WIDTH + CHUNK_WIDTH / 2.0f;
                (void)(cy * CHUNK_HEIGHT + CHUNK_HEIGHT / 2.0f);  // Y not used in load distance calculation
                float chunk_center_z = cz * CHUNK_DEPTH + CHUNK_DEPTH / 2.0f;

                // Direction from player to chunk
                float to_chunk_x = chunk_center_x - player_pos.x;
                float to_chunk_z = chunk_center_z - player_pos.z;

                // Skip chunks that are behind the player based on horizontal view only.
                // Avoid using the camera's vertical component here, because looking up or down
                // should not cause vertical chunks to be treated as behind the player.
                if (has_horizontal_forward) {
                    float dot_xz = to_chunk_x * camera_forward_xz_norm_x + to_chunk_z * camera_forward_xz_norm_z;
                    if (dot_xz < -0.3f * (load_dist + 1) * CHUNK_WIDTH) {
                        continue;
                    }
                }

                Chunk* chunk = world_load_or_create_chunk(world, cx, cy, cz);
                if (chunk && !chunk->loaded) {
                    if (!chunk->generated) {
                        // Generate this chunk procedurally
                        world_generate_chunk(chunk, world->seed);
                        chunk->generated = true;
                    }
                    // Mark as loaded again (this chunk was previously unloaded)
                    chunk->loaded = true;
                    chunk->pending_unload = false;
                    // Mark lighting/mesh dirty so the worker will rebuild.
                    chunk->needs_relighting = true;
                    chunk->meshed = false;
                    // NOTE: Don't queue yet - we'll do it after releasing the lock to avoid holding lock too long
                }
            }
        }
    }

    pthread_mutex_unlock(&world->cache_mutex);

    // Queue newly generated chunks for lighting/meshing after releasing cache_mutex
    pthread_mutex_lock(&world->cache_mutex);
    for (int cx = player_chunk_x - load_dist; cx <= player_chunk_x + load_dist; cx++) {
        for (int cy = player_chunk_y - load_dist; cy <= player_chunk_y + load_dist; cy++) {
            for (int cz = player_chunk_z - load_dist; cz <= player_chunk_z + load_dist; cz++) {
                Chunk* chunk = world_get_chunk(world, cx, cy, cz);
                if (chunk && chunk->generated && chunk->loaded && chunk->needs_relighting) {
                    // Queue for worker to calculate lighting and mesh
                    worker_queue_chunk(world, chunk);
                }
            }
        }
    }
    pthread_mutex_unlock(&world->cache_mutex);

    // Invalidate neighboring chunk meshes after loading new chunks.
    // This ensures adjacent chunks update their faces when chunk load state changes.
    {
        const int neighbor_offsets[6][3] = {
            { 1,  0,  0}, {-1,  0,  0},
            { 0,  1,  0}, { 0, -1,  0},
            { 0,  0,  1}, { 0,  0, -1}
        };

        pthread_mutex_lock(&world->cache_mutex);
        for (int cx = player_chunk_x - load_dist; cx <= player_chunk_x + load_dist; cx++) {
            for (int cy = player_chunk_y - load_dist; cy <= player_chunk_y + load_dist; cy++) {
                for (int cz = player_chunk_z - load_dist; cz <= player_chunk_z + load_dist; cz++) {
                    Chunk* chunk = world_get_chunk(world, cx, cy, cz);
                    if (!chunk || !chunk->loaded || !chunk->generated) continue;

                    for (int ni = 0; ni < 6; ni++) {
                        int nx = cx + neighbor_offsets[ni][0];
                        int ny = cy + neighbor_offsets[ni][1];
                        int nz = cz + neighbor_offsets[ni][2];
                        Chunk* neighbor = world_get_chunk(world, nx, ny, nz);
                        if (!neighbor || !neighbor->loaded || !neighbor->generated) continue;
                        if (neighbor == chunk) continue;

                        bool was_meshed;
                        pthread_mutex_lock(&neighbor->mutex);
                        was_meshed = neighbor->meshed;
                        neighbor->meshed = false;
                        pthread_mutex_unlock(&neighbor->mutex);

                        if (was_meshed) {
                            worker_queue_chunk(world, neighbor);
                        }
                    }
                }
            }
        }
        pthread_mutex_unlock(&world->cache_mutex);
    }

    // Note: We no longer flush the worker queue here to avoid stalling the main thread.
    // Chunks that are in-use by the worker (in_use_count > 0) will not be unloaded until
    // their jobs complete.

    // CRITICAL: Lock cache mutex while modifying chunk array
    pthread_mutex_lock(&world->cache_mutex);

    // Unload chunks that are too far away or behind the player
    int unload_dist = load_dist + 1;
    int i = 0;

    // Throttle unloads to avoid stuttering when crossing chunk boundaries.
    // We only unload a small number of chunks per frame, spreading work across frames.
    const int max_unloads_per_frame = 1;
    int unloads_this_frame = 0;

    while (i < world->chunk_cache.chunk_count) {
        // If we already unloaded enough chunks this frame, stop here.
        if (unloads_this_frame >= max_unloads_per_frame) {
            break;
        }

        Chunk* chunk = &world->chunk_cache.chunks[i];
        int dx = chunk->chunk_x - player_chunk_x;
        int dy = chunk->chunk_y - player_chunk_y;
        int dz = chunk->chunk_z - player_chunk_z;

        // Check if chunk is beyond unload distance or behind player
        float chunk_center_x = chunk->chunk_x * CHUNK_WIDTH + CHUNK_WIDTH / 2.0f;
        (void)(chunk->chunk_y * CHUNK_HEIGHT + CHUNK_HEIGHT / 2.0f);  // Y not used in unload distance calculation
        float chunk_center_z = chunk->chunk_z * CHUNK_DEPTH + CHUNK_DEPTH / 2.0f;

        float to_chunk_x = chunk_center_x - player_pos.x;
        float to_chunk_z = chunk_center_z - player_pos.z;

        bool behind_player = false;
        if (has_horizontal_forward) {
            float dot_xz = to_chunk_x * camera_forward_xz_norm_x + to_chunk_z * camera_forward_xz_norm_z;
            behind_player = dot_xz < -0.3f * (unload_dist + 1) * CHUNK_WIDTH;
        }
        bool too_far = dx*dx + dz*dz > unload_dist*unload_dist || dy > unload_dist || dy < -unload_dist;

        if (too_far || behind_player) {
            // Avoid unloading while a worker is still processing this chunk
            if (__atomic_load_n(&chunk->in_use_count, __ATOMIC_ACQUIRE) > 0) {
                i++;
                continue;
            }

            // Invalidate neighbor meshes before unloading this chunk.
            // Neighbors may now need to update faces that were previously against this chunk.
            {
                const int neighbor_offsets[6][3] = {
                    { 1,  0,  0}, {-1,  0,  0},
                    { 0,  1,  0}, { 0, -1,  0},
                    { 0,  0,  1}, { 0,  0, -1}
                };
                for (int ni = 0; ni < 6; ni++) {
                    int nx = chunk->chunk_x + neighbor_offsets[ni][0];
                    int ny = chunk->chunk_y + neighbor_offsets[ni][1];
                    int nz = chunk->chunk_z + neighbor_offsets[ni][2];
                    Chunk* neighbor = world_get_chunk(world, nx, ny, nz);
                    if (!neighbor || !neighbor->loaded || !neighbor->generated) continue;
                    if (neighbor == chunk) continue;

                    bool was_meshed;
                    pthread_mutex_lock(&neighbor->mutex);
                    was_meshed = neighbor->meshed;
                    neighbor->meshed = false;
                    pthread_mutex_unlock(&neighbor->mutex);

                    if (was_meshed) {
                        worker_queue_chunk(world, neighbor);
                    }
                }
            }

            // If modified, queue async save and mark for unload after save completes.
            // This prevents main-thread stalls due to disk I/O during unloading.
            if (chunk->modified && !chunk->pending_save) {
                chunk->pending_unload = true;
                // We can mark the chunk as unloaded for rendering purposes while we save it.
                // This keeps it in memory until save completes, but removes it from active rendering.
                chunk->loaded = false;
                worker_queue_chunk_save(world, chunk);
            }

            // If chunk is not pending save, we can unload it immediately
            if (!chunk->pending_save) {
                // Clean up chunk resources
                chunk_free_visible_blocks(chunk);  // Free mesh
                chunk_free_merged_mesh(chunk);  // Free merged mesh buffers

                // Remove chunk from cache (swap with last)
                if (i < world->chunk_cache.chunk_count - 1) {
                    world->chunk_cache.chunks[i] = world->chunk_cache.chunks[world->chunk_cache.chunk_count - 1];
                }
                world->chunk_cache.chunk_count--;
                unloads_this_frame++;
            } else {
                // Skip this chunk for now; it will be removed once the save completes
                i++;
            }
        } else {
            i++;
        }
    }

    pthread_mutex_unlock(&world->cache_mutex);
}

// Chunk file format header
static const char CHUNK_FILE_MAGIC[4] = {'B', '3', 'D', 'V'};
static const uint8_t CHUNK_FILE_VERSION = 2;

// Human-readable error for last chunk load failure (used for diagnostics)
static char CHUNK_LOAD_ERROR[256];

enum ChunkFileMethod {
    CHUNK_METHOD_RAW = 0,
    CHUNK_METHOD_RLE = 1,
    CHUNK_METHOD_RAW_COMPRESSED = 2,
    CHUNK_METHOD_RLE_COMPRESSED = 3,
};

static bool write_le32(FILE* file, uint32_t value)
{
    uint8_t bytes[4] = {
        (uint8_t)(value & 0xFF),
        (uint8_t)((value >> 8) & 0xFF),
        (uint8_t)((value >> 16) & 0xFF),
        (uint8_t)((value >> 24) & 0xFF),
    };
    return fwrite(bytes, 1, sizeof(bytes), file) == sizeof(bytes);
}

static bool read_le32(FILE* file, uint32_t* out_value)
{
    uint8_t bytes[4];
    if (fread(bytes, 1, sizeof(bytes), file) != sizeof(bytes)) return false;
    *out_value = (uint32_t)bytes[0] |
                 ((uint32_t)bytes[1] << 8) |
                 ((uint32_t)bytes[2] << 16) |
                 ((uint32_t)bytes[3] << 24);
    return true;
}

static uint8_t* serialize_chunk_raw(Chunk* chunk, size_t* out_size)
{
    const int total_blocks = CHUNK_WIDTH * CHUNK_HEIGHT * CHUNK_DEPTH;
    uint8_t* buffer = (uint8_t*)malloc(total_blocks);
    if (!buffer) return NULL;

    int offset = 0;
    for (int y = 0; y < CHUNK_HEIGHT; y++) {
        for (int z = 0; z < CHUNK_DEPTH; z++) {
            for (int x = 0; x < CHUNK_WIDTH; x++) {
                buffer[offset++] = (uint8_t)chunk->blocks[y][z][x].type;
            }
        }
    }

    *out_size = total_blocks;
    return buffer;
}

static uint8_t* serialize_chunk_rle_v1(Chunk* chunk, size_t* out_size)
{
    const int total_blocks = CHUNK_WIDTH * CHUNK_HEIGHT * CHUNK_DEPTH;
    size_t max_size = (size_t)total_blocks * 3;
    uint8_t* buffer = (uint8_t*)malloc(max_size);
    if (!buffer) return NULL;

    uint8_t current_type = (uint8_t)chunk->blocks[0][0][0].type;
    uint16_t run_length = 1;
    size_t offset = 0;

    for (int block_index = 1; block_index < total_blocks; block_index++) {
        int y = block_index / (CHUNK_WIDTH * CHUNK_DEPTH);
        int rem = block_index % (CHUNK_WIDTH * CHUNK_DEPTH);
        int z = rem / CHUNK_WIDTH;
        int x = rem % CHUNK_WIDTH;
        uint8_t next_type = (uint8_t)chunk->blocks[y][z][x].type;

        if (next_type == current_type && run_length < UINT16_MAX) {
            run_length++;
        } else {
            memcpy(buffer + offset, &run_length, sizeof(run_length));
            offset += sizeof(run_length);
            buffer[offset++] = current_type;
            current_type = next_type;
            run_length = 1;
        }
    }

    memcpy(buffer + offset, &run_length, sizeof(run_length));
    offset += sizeof(run_length);
    buffer[offset++] = current_type;

    *out_size = offset;
    return buffer;
}

static uint8_t* serialize_chunk_rle_v2(Chunk* chunk, size_t* out_size)
{
    const int total_blocks = CHUNK_WIDTH * CHUNK_HEIGHT * CHUNK_DEPTH;
    size_t max_size = (size_t)total_blocks * 5;
    uint8_t* buffer = (uint8_t*)malloc(max_size);
    if (!buffer) return NULL;

    uint8_t current_type = (uint8_t)chunk->blocks[0][0][0].type;
    uint32_t run_length = 1;
    size_t offset = 0;

    for (int block_index = 1; block_index < total_blocks; block_index++) {
        int y = block_index / (CHUNK_WIDTH * CHUNK_DEPTH);
        int rem = block_index % (CHUNK_WIDTH * CHUNK_DEPTH);
        int z = rem / CHUNK_WIDTH;
        int x = rem % CHUNK_WIDTH;
        uint8_t next_type = (uint8_t)chunk->blocks[y][z][x].type;

        if (next_type == current_type) {
            run_length++;
        } else {
            memcpy(buffer + offset, &run_length, sizeof(run_length));
            offset += sizeof(run_length);
            buffer[offset++] = current_type;
            current_type = next_type;
            run_length = 1;
        }
    }

    memcpy(buffer + offset, &run_length, sizeof(run_length));
    offset += sizeof(run_length);
    buffer[offset++] = current_type;

    *out_size = offset;
    return buffer;
}

static uint8_t* compress_buffer(const uint8_t* source, size_t source_size, size_t* out_compressed_size)
{
    uLong bound = compressBound(source_size);
    uint8_t* compressed = (uint8_t*)malloc(bound);
    if (!compressed) return NULL;

    uLongf compressed_size = bound;
    int result = compress2(compressed, &compressed_size, source, source_size, Z_BEST_SPEED);
    if (result != Z_OK) {
        free(compressed);
        return NULL;
    }

    *out_compressed_size = (size_t)compressed_size;
    return compressed;
}

static bool write_chunk_file(FILE* file, Chunk* chunk, bool allow_compression)
{
    if (!file || !chunk) return false;

    size_t raw_size = 0;
    uint8_t* raw_buffer = serialize_chunk_raw(chunk, &raw_size);
    if (!raw_buffer) return false;

    size_t rle_size = 0;
    uint8_t* rle_buffer = serialize_chunk_rle_v2(chunk, &rle_size);
    if (!rle_buffer) {
        free(raw_buffer);
        return false;
    }

    size_t raw_comp_size = 0;
    uint8_t* raw_comp = NULL;
    size_t rle_comp_size = 0;
    uint8_t* rle_comp = NULL;
    if (allow_compression) {
        raw_comp = compress_buffer(raw_buffer, raw_size, &raw_comp_size);
        rle_comp = compress_buffer(rle_buffer, rle_size, &rle_comp_size);
    }

    uint8_t method = CHUNK_METHOD_RLE;
    const uint8_t* payload = rle_buffer;
    size_t payload_size = rle_size;
    uint32_t uncompressed_size = (uint32_t)rle_size;

    if (raw_size < payload_size) {
        method = CHUNK_METHOD_RAW;
        payload = raw_buffer;
        payload_size = raw_size;
        uncompressed_size = (uint32_t)raw_size;
    }

    if (allow_compression && raw_comp && raw_comp_size < payload_size) {
        method = CHUNK_METHOD_RAW_COMPRESSED;
        payload = raw_comp;
        payload_size = raw_comp_size;
        uncompressed_size = (uint32_t)raw_size;
    }

    if (allow_compression && rle_comp && rle_comp_size < payload_size) {
        method = CHUNK_METHOD_RLE_COMPRESSED;
        payload = rle_comp;
        payload_size = rle_comp_size;
        uncompressed_size = (uint32_t)rle_size;
    }

    bool success = true;
    if (fwrite(CHUNK_FILE_MAGIC, 1, sizeof(CHUNK_FILE_MAGIC), file) != sizeof(CHUNK_FILE_MAGIC) ||
        fwrite(&CHUNK_FILE_VERSION, 1, 1, file) != 1 ||
        fwrite(&method, 1, 1, file) != 1 ||
        !write_le32(file, (uint32_t)payload_size) ||
        !write_le32(file, uncompressed_size) ||
        fwrite(payload, 1, payload_size, file) != payload_size) {
        success = false;
    }

    free(raw_buffer);
    free(rle_buffer);
    if (raw_comp) free(raw_comp);
    if (rle_comp) free(rle_comp);
    return success;
}

static bool parse_rle_payload_v1(const uint8_t* buffer, size_t buffer_size, Chunk* chunk)
{
    const int total_blocks = CHUNK_WIDTH * CHUNK_HEIGHT * CHUNK_DEPTH;
    int loaded_blocks = 0;
    size_t offset = 0;

    while (loaded_blocks < total_blocks && offset + sizeof(uint16_t) + 1 <= buffer_size) {
        uint16_t run_length;
        memcpy(&run_length, buffer + offset, sizeof(run_length));
        offset += sizeof(run_length);
        uint8_t block_type = buffer[offset++];

        if (run_length == 0 || loaded_blocks + run_length > total_blocks) {
            return false;
        }

        for (int i = 0; i < run_length; i++) {
            int flat_index = loaded_blocks + i;
            int y = flat_index / (CHUNK_WIDTH * CHUNK_DEPTH);
            int rem = flat_index % (CHUNK_WIDTH * CHUNK_DEPTH);
            int z = rem / CHUNK_WIDTH;
            int x = rem % CHUNK_WIDTH;
            chunk->blocks[y][z][x].type = (BlockType)block_type;
        }
        loaded_blocks += run_length;
    }

    return loaded_blocks == total_blocks;
}

static bool parse_rle_payload_v2(const uint8_t* buffer, size_t buffer_size, Chunk* chunk)
{
    const int total_blocks = CHUNK_WIDTH * CHUNK_HEIGHT * CHUNK_DEPTH;
    int loaded_blocks = 0;
    size_t offset = 0;

    while (loaded_blocks < total_blocks && offset + sizeof(uint32_t) + 1 <= buffer_size) {
        uint32_t run_length;
        memcpy(&run_length, buffer + offset, sizeof(run_length));
        offset += sizeof(run_length);
        uint8_t block_type = buffer[offset++];

        if (run_length == 0 || loaded_blocks + run_length > total_blocks) {
            return false;
        }

        for (uint32_t i = 0; i < run_length; i++) {
            int flat_index = loaded_blocks + i;
            int y = flat_index / (CHUNK_WIDTH * CHUNK_DEPTH);
            int rem = flat_index % (CHUNK_WIDTH * CHUNK_DEPTH);
            int z = rem / CHUNK_WIDTH;
            int x = rem % CHUNK_WIDTH;
            chunk->blocks[y][z][x].type = (BlockType)block_type;
        }
        loaded_blocks += run_length;
    }

    return loaded_blocks == total_blocks;
}

static bool load_chunk_from_file(FILE* file, Chunk* chunk)
{
    // Clear previous error
    CHUNK_LOAD_ERROR[0] = '\0';

    uint8_t magic[4];
    if (fread(magic, 1, sizeof(magic), file) != sizeof(magic)) {
        snprintf(CHUNK_LOAD_ERROR, sizeof(CHUNK_LOAD_ERROR), "short header");
        return false;
    }
    if (memcmp(magic, CHUNK_FILE_MAGIC, sizeof(magic)) != 0) {
        snprintf(CHUNK_LOAD_ERROR, sizeof(CHUNK_LOAD_ERROR), "bad magic: %.4s", (char*)magic);
        return false;
    }

    uint8_t version;
    if (fread(&version, 1, 1, file) != 1) {
        snprintf(CHUNK_LOAD_ERROR, sizeof(CHUNK_LOAD_ERROR), "short version byte");
        return false;
    }
    if (version != 1 && version != CHUNK_FILE_VERSION) {
        snprintf(CHUNK_LOAD_ERROR, sizeof(CHUNK_LOAD_ERROR), "unsupported version %u", version);
        return false;
    }

    uint8_t method;
    if (fread(&method, 1, 1, file) != 1) {
        snprintf(CHUNK_LOAD_ERROR, sizeof(CHUNK_LOAD_ERROR), "short method byte");
        return false;
    }

    uint32_t payload_size;
    uint32_t uncompressed_size;
    if (!read_le32(file, &payload_size) || !read_le32(file, &uncompressed_size)) {
        snprintf(CHUNK_LOAD_ERROR, sizeof(CHUNK_LOAD_ERROR), "short size fields");
        return false;
    }

    uint8_t* payload = (uint8_t*)malloc(payload_size);
    if (!payload) {
        snprintf(CHUNK_LOAD_ERROR, sizeof(CHUNK_LOAD_ERROR), "malloc payload failed (%u)", payload_size);
        return false;
    }
    if (fread(payload, 1, payload_size, file) != payload_size) {
        snprintf(CHUNK_LOAD_ERROR, sizeof(CHUNK_LOAD_ERROR), "payload truncated (got %zu expected %u)", (size_t)fread(payload,1,0,file), payload_size);
        free(payload);
        return false;
    }

    uint8_t* decompressed = NULL;
    const uint8_t* parse_buffer = payload;
    size_t parse_buffer_size = payload_size;

    if (method == CHUNK_METHOD_RAW_COMPRESSED || method == CHUNK_METHOD_RLE_COMPRESSED) {
        decompressed = (uint8_t*)malloc(uncompressed_size);
        if (!decompressed) {
            snprintf(CHUNK_LOAD_ERROR, sizeof(CHUNK_LOAD_ERROR), "malloc decompressed failed (%u)", uncompressed_size);
            free(payload);
            return false;
        }
        uLongf dest_len = uncompressed_size;
        int result = uncompress(decompressed, &dest_len, payload, payload_size);
        free(payload);
        if (result != Z_OK || dest_len != uncompressed_size) {
            snprintf(CHUNK_LOAD_ERROR, sizeof(CHUNK_LOAD_ERROR), "zlib uncompress failed (%d) dest_len=%lu expected=%u", result, (unsigned long)dest_len, uncompressed_size);
            free(decompressed);
            return false;
        }
        parse_buffer = decompressed;
        parse_buffer_size = dest_len;
    }

    bool success = false;
    if (method == CHUNK_METHOD_RAW || method == CHUNK_METHOD_RAW_COMPRESSED) {
        const int total_blocks = CHUNK_WIDTH * CHUNK_HEIGHT * CHUNK_DEPTH;
        if (parse_buffer_size == (size_t)total_blocks) {
            int offset = 0;
            for (int y = 0; y < CHUNK_HEIGHT; y++) {
                for (int z = 0; z < CHUNK_DEPTH; z++) {
                    for (int x = 0; x < CHUNK_WIDTH; x++) {
                        chunk->blocks[y][z][x].type = (BlockType)parse_buffer[offset++];
                    }
                }
            }
            success = true;
        } else {
            snprintf(CHUNK_LOAD_ERROR, sizeof(CHUNK_LOAD_ERROR), "raw size mismatch parsed=%zu expected=%d", parse_buffer_size, CHUNK_WIDTH * CHUNK_HEIGHT * CHUNK_DEPTH);
        }
    } else if (method == CHUNK_METHOD_RLE || method == CHUNK_METHOD_RLE_COMPRESSED) {
        if (version == 1) {
            success = parse_rle_payload_v1(parse_buffer, parse_buffer_size, chunk);
            if (!success) snprintf(CHUNK_LOAD_ERROR, sizeof(CHUNK_LOAD_ERROR), "rle v1 parse failed (buf=%zu)", parse_buffer_size);
        } else {
            success = parse_rle_payload_v2(parse_buffer, parse_buffer_size, chunk);
            if (!success) snprintf(CHUNK_LOAD_ERROR, sizeof(CHUNK_LOAD_ERROR), "rle v2 parse failed (buf=%zu)", parse_buffer_size);
        }
    } else {
        snprintf(CHUNK_LOAD_ERROR, sizeof(CHUNK_LOAD_ERROR), "unknown method %u", method);
    }

    if (decompressed) free(decompressed);
    if (method == CHUNK_METHOD_RAW || method == CHUNK_METHOD_RLE) free(payload);
    return success;
}

// Save a single chunk to disk
bool world_save_chunk(Chunk* chunk, const char* world_name, bool allow_compression)
{
    if (!chunk || !world_name) return false;

    char filepath[512];
    snprintf(filepath, sizeof(filepath), "./worlds/%s/chunks/chunk_%d_%d_%d.chunk",
             world_name, chunk->chunk_x, chunk->chunk_y, chunk->chunk_z);

    FILE* file = fopen(filepath, "wb");
    if (!file) return false;

    bool success = write_chunk_file(file, chunk, allow_compression);
    fclose(file);
    return success;
}

// Initialize worlds folder if it doesn't exist
void world_system_init(void)
{
    #ifdef _WIN32
        system("if not exist worlds mkdir worlds");
    #else
        system("mkdir -p ./worlds");
    #endif
}

// Save world to files (chunks)
bool world_save(World* world, const char* world_name)
{
    if (!world || !world_name) return false;

    world_system_init();

    // Create world metadata file
    char world_dir[512];
    snprintf(world_dir, sizeof(world_dir), "./worlds/%s", world_name);

    #ifdef _WIN32
        char mkdir_cmd[512];
        snprintf(mkdir_cmd, sizeof(mkdir_cmd), "if not exist \"worlds\\%s\" mkdir \"worlds\\%s\"", world_name, world_name);
        system(mkdir_cmd);
        snprintf(mkdir_cmd, sizeof(mkdir_cmd), "if not exist \"worlds\\%s\\chunks\" mkdir \"worlds\\%s\\chunks\"", world_name, world_name);
        system(mkdir_cmd);
    #else
        char mkdir_cmd[512];
        snprintf(mkdir_cmd, sizeof(mkdir_cmd), "mkdir -p \"./worlds/%s/chunks\"", world_name);
        system(mkdir_cmd);
    #endif

    // Write world metadata file
    char metadata_path[512];
    snprintf(metadata_path, sizeof(metadata_path), "./worlds/%s/world.txt", world_name);

    FILE* metadata_file = fopen(metadata_path, "w");
    if (metadata_file) {
        time_t now = time(NULL);
        struct tm* timeinfo = localtime(&now);
        char time_str[64];
        strftime(time_str, sizeof(time_str), "%Y-%m-%d %H:%M:%S", timeinfo);

        fprintf(metadata_file, "name=%s\n", world_name);
        fprintf(metadata_file, "seed=%lu\n", world->seed);
        fprintf(metadata_file, "compress=%d\n", world->compress_chunk_files ? 1 : 0);
        fprintf(metadata_file, "last_saved=%s\n", time_str);
        fprintf(metadata_file, "chunk_count=%d\n", world->chunk_cache.chunk_count);
        // Player position is stored per-world in players.toml (or legacy players.txt) now
        fclose(metadata_file);
    }

    // Prefer player position from players.toml or fallback to legacy players.txt
    world_apply_players_to(world, NULL);

    // Save each loaded chunk
    for (int i = 0; i < world->chunk_cache.chunk_count; i++) {
        Chunk* chunk = &world->chunk_cache.chunks[i];
        if (world_save_chunk(chunk, world_name, world->compress_chunk_files)) {
            chunk->modified = false;  // Mark chunk as saved
        }
    }

    // Always write players.txt for this world (store last position and inventory if available)
    write_players_file(world, world->current_player, world_name);

    return true;
}

// Helper: map BlockType to string id used in players file
static const char* block_type_to_id(BlockType t)
{
    switch (t) {
        case BLOCK_STONE: return "block_stone";
        case BLOCK_DIRT: return "block_dirt";
        case BLOCK_GRASS: return "block_grass";
        case BLOCK_SAND: return "block_sand";
        case BLOCK_WOOD: return "block_wood";
        case BLOCK_BEDROCK: return "block_bedrock";
        case BLOCK_GLOWSTONE: return "block_glowstone";
        case BLOCK_AIR: default: return "block_air";
    }
}

// Helper: map id string to BlockType
static BlockType block_id_to_type(const char* id)
{
    if (!id) return BLOCK_AIR;
    if (strcmp(id, "block_stone") == 0) return BLOCK_STONE;
    if (strcmp(id, "block_dirt") == 0) return BLOCK_DIRT;
    if (strcmp(id, "block_grass") == 0) return BLOCK_GRASS;
    if (strcmp(id, "block_sand") == 0) return BLOCK_SAND;
    if (strcmp(id, "block_wood") == 0) return BLOCK_WOOD;
    if (strcmp(id, "block_bedrock") == 0) return BLOCK_BEDROCK;
    if (strcmp(id, "block_glowstone") == 0) return BLOCK_GLOWSTONE;
    return BLOCK_AIR;
}

// Write the current player's data (position + inventory) into players.toml in the world folder
static void write_players_file(World* world, void* player_ptr, const char* world_name)
{
    if (!world || !world_name) return;
    Player* player = (Player*)player_ptr;
    char path[512];
    snprintf(path, sizeof(path), "./worlds/%s/players.toml", world_name);
    FILE* f = fopen(path, "w");
    if (!f) return;

    // Use a simple fixed UID for single-player for now
    const char* uid = "00000001";
    time_t now = time(NULL);
    fprintf(f, "[player.%s]\n", uid);
    // Write nickname from world if set, otherwise default to "Player"
    const char* nick_to_write = (world->player_nickname[0] != '\0') ? world->player_nickname : "Player";
    fprintf(f, "nickname = \"%s\"\n", nick_to_write);
    fprintf(f, "last_ip = \"0.0.20-beta\"\n");
    fprintf(f, "last_login = %lu\n", (unsigned long)now);
    // Save position: prefer live player's position if available, otherwise use world's last_player_position
    Vector3 pos = player ? player->position : world->last_player_position;
    fprintf(f, "x = %.6f\n", pos.x);
    fprintf(f, "y = %.6f\n", pos.y);
    fprintf(f, "z = %.6f\n", pos.z);

    fprintf(f, "\n[slots]\n\n");
    // Total 45 slots: 0-8 hotbar, 9-44 big inventory
    if (player) {
        for (int i = 0; i < 45; i++) {
            InventorySlot slot;
            if (i < 9) slot = player->inventory[i];
            else slot = player->big_inventory[i - 9];
            if (slot.type != BLOCK_AIR && slot.count > 0) {
                fprintf(f, "[slots.%d]\n", i);
                fprintf(f, "id = \"%s\"\n", block_type_to_id(slot.type));
                fprintf(f, "count = %d\n\n", slot.count);
            }
        }
    }

    fclose(f);
}

// Read players.txt and apply found player's position into world->last_player_position
// and (optionally) apply inventory into a provided Player* when not NULL.
bool world_apply_players_to(World* world, void* player_ptr)
{
    if (!world) return false;
    Player* player = (Player*)player_ptr;
    char path[512];
    snprintf(path, sizeof(path), "./worlds/%s/players.toml", world->world_name);
    FILE* f = fopen(path, "r");
    if (!f) {
        snprintf(path, sizeof(path), "./worlds/%s/players.txt", world->world_name);
        f = fopen(path, "r");
        if (!f) return false;
    }

    char line[512];
    int current_slot = -1;
    bool in_player_section = false;
    while (fgets(line, sizeof(line), f)) {
        // Trim leading whitespace
        char* p = line;
        while (*p == ' ' || *p == '\t' || *p == '\r' || *p == '\n') p++;
        if (*p == '\0' || *p == '\n' || *p == '#') continue;

        if (p[0] == '[') {
            // Section header
            if (strncmp(p, "[player.", 8) == 0) {
                in_player_section = true;
                current_slot = -1;
                continue;
            }
            // entering any other section clears player section flag
            in_player_section = false;
            // Slots header like [slots] or [slots.N]
            if (strncmp(p, "[slots.", 7) == 0) {
                // parse number between '.' and ']'
                char* dot = strchr(p, '.');
                char* endb = strchr(p, ']');
                if (dot && endb && endb > dot+1) {
                    char numbuf[16] = {0};
                    int len = endb - (dot+1);
                    if (len > 0 && len < (int)sizeof(numbuf)) {
                        strncpy(numbuf, dot+1, len);
                        numbuf[len] = '\0';
                        int idx = atoi(numbuf);
                        current_slot = idx;
                    } else current_slot = -1;
                } else current_slot = -1;
                continue;
            }
            // other sections - ignore
            current_slot = -1;
            continue;
        }

        // Parse key = value
        char val[256];
        // If in player section, look for nickname
        if (in_player_section) {
            // look for nickname = "..."
            const char* nick_key = "nickname";
            if (strncmp(p, nick_key, strlen(nick_key)) == 0) {
                char* q1 = strchr(p, '"');
                if (q1) {
                    char* q2 = strchr(q1 + 1, '"');
                    if (q2) {
                        size_t len = q2 - (q1 + 1);
                        if (len >= sizeof(world->player_nickname)) len = sizeof(world->player_nickname)-1;
                        memcpy(world->player_nickname, q1 + 1, len);
                        world->player_nickname[len] = '\0';
                    }
                }
                continue;
            }
        }
        if (sscanf(p, "x = %f", &world->last_player_position.x) == 1) continue;
        if (sscanf(p, "y = %f", &world->last_player_position.y) == 1) continue;
        if (sscanf(p, "z = %f", &world->last_player_position.z) == 1) continue;

        if (current_slot >= 0) {
            // parse id (quoted string) or count
            char* q1 = strchr(p, '"');
            if (q1) {
                char* q2 = strchr(q1 + 1, '"');
                if (q2) {
                    size_t len = q2 - (q1 + 1);
                    if (len < sizeof(val)) {
                        memcpy(val, q1 + 1, len);
                        val[len] = '\0';
                        BlockType t = block_id_to_type(val);
                        if (player) {
                            if (current_slot < 9) {
                                player->inventory[current_slot].type = t;
                            } else if (current_slot < 45) {
                                player->big_inventory[current_slot - 9].type = t;
                            }
                        }
                    }
                }
            } else {
                // try parse count
                int count_val;
                if (sscanf(p, "count = %d", &count_val) == 1) {
                    if (player) {
                        if (current_slot < 9) player->inventory[current_slot].count = count_val;
                        else if (current_slot < 45) player->big_inventory[current_slot - 9].count = count_val;
                    }
                }
            }
        }
    }

    // Reparse to read slot counts properly (two-pass: first set types, second set counts)
    rewind(f);
    current_slot = -1;
    while (fgets(line, sizeof(line), f)) {
        char* p = line;
        while (*p == ' ' || *p == '\t' || *p == '\r' || *p == '\n') p++;
        if (*p == '\0' || *p == '\n' || *p == '#') continue;
        if (p[0] == '[') {
            if (strncmp(p, "[player.", 8) == 0) { current_slot = -1; continue; }
            if (strncmp(p, "[slots.", 7) == 0) {
                char* dot = strchr(p, '.');
                char* endb = strchr(p, ']');
                if (dot && endb && endb > dot+1) {
                    char numbuf[16] = {0};
                    int len = endb - (dot+1);
                    if (len > 0 && len < (int)sizeof(numbuf)) {
                        strncpy(numbuf, dot+1, len);
                        numbuf[len] = '\0';
                        current_slot = atoi(numbuf);
                    } else current_slot = -1;
                } else current_slot = -1;
                continue;
            }
            current_slot = -1; continue;
        }

        if (current_slot >= 0) {
            char idbuf[256];
            // parse quoted id first
            char* q1 = strchr(p, '"');
            if (q1) {
                char* q2 = strchr(q1 + 1, '"');
                if (q2) {
                    int len = (int)(q2 - (q1 + 1));
                    if (len < (int)sizeof(idbuf)) {
                        memcpy(idbuf, q1 + 1, len);
                        idbuf[len] = '\0';
                        BlockType t = block_id_to_type(idbuf);
                        if (current_slot < 9) {
                            if (player) player->inventory[current_slot].type = t;
                        } else if (current_slot < 45) {
                            if (player) player->big_inventory[current_slot - 9].type = t;
                        }
                    }
                }
            } else {
                int countv;
                if (sscanf(p, "count = %d", &countv) == 1) {
                    if (current_slot < 9) {
                        if (player) player->inventory[current_slot].count = countv;
                    } else if (current_slot < 45) {
                        if (player) player->big_inventory[current_slot - 9].count = countv;
                    }
                }
            }
        } else {
            // parse top-level x y z again
            float fx, fy, fz;
            if (sscanf(p, "x = %f", &fx) == 1) world->last_player_position.x = fx;
            if (sscanf(p, "y = %f", &fy) == 1) world->last_player_position.y = fy;
            if (sscanf(p, "z = %f", &fz) == 1) world->last_player_position.z = fz;
        }
    }

    fclose(f);

    // Ensure player arrays have defaults for empty slots
    if (player) {
        for (int i = 0; i < INVENTORY_SIZE; i++) {
            if (player->inventory[i].count <= 0) player->inventory[i].type = BLOCK_AIR;
        }
        for (int i = 0; i < BIG_INVENTORY_SIZE; i++) {
            if (player->big_inventory[i].count <= 0) player->big_inventory[i].type = BLOCK_AIR;
        }
    }

    return true;
}

// Load world from files (chunks)
bool world_load(World* world, const char* world_name)
{
    if (!world || !world_name) return false;

    // CRITICAL: Flush the worker queue before clearing chunks
    // This prevents the worker thread from accessing chunks we're about to reset
    worker_flush_queue(world);

    // Clear existing chunks first
    // NOTE: Don't destroy mutexes - worker thread is still running and might use them
    // Just clear out the data
    if (world->chunk_cache.chunks) {
        for (int i = 0; i < world->chunk_cache.chunk_count; i++) {
            chunk_free_visible_blocks(&world->chunk_cache.chunks[i]);
            chunk_free_merged_mesh(&world->chunk_cache.chunks[i]);  // Free merged mesh buffers
            // Don't destroy mutexes - they'll be reused when new chunks are loaded
        }
    }

    // Reset the chunk count (keeps pre-allocated memory)
    world->chunk_cache.chunk_count = 0;

    // Set the world name so chunk loading uses the correct directory
    strncpy(world->world_name, world_name, sizeof(world->world_name) - 1);
    world->world_name[sizeof(world->world_name) - 1] = '\0';

    // Reset chunk loading tracker
    world->last_loaded_chunk_x = INT32_MAX;
    world->last_loaded_chunk_y = INT32_MAX;
    world->last_loaded_chunk_z = INT32_MAX;

    // Default player position (used only if players.txt missing)
    world->last_player_position = (Vector3){8.0f, 20.0f, 8.0f};  // Default position
    char metadata_path[512];
    snprintf(metadata_path, sizeof(metadata_path), "./worlds/%s/world.txt", world_name);
    FILE* metadata_file = fopen(metadata_path, "r");
    if (metadata_file) {
        char line[256];
        while (fgets(line, sizeof(line), metadata_file)) {
            uint64_t seed_val;
            if (sscanf(line, "seed=%lu", &seed_val) == 1) {
                world->seed = seed_val;
            } else if (strncmp(line, "compress=", 9) == 0) {
                world->compress_chunk_files = (line[9] == '1');
            }
        }
        fclose(metadata_file);
    }

    // If players.txt exists, prefer player position from there (players file supersedes world.txt)
    world_apply_players_to(world, NULL);

    // Try to load initial chunks from disk
    // Only generate minimal spawn area to avoid startup lag
    int spawn_dist = 1;  // Only load immediate area around spawn

    for (int cx = -spawn_dist; cx <= spawn_dist; cx++) {
        for (int cy = -1; cy <= 1; cy++) {
            for (int cz = -spawn_dist; cz <= spawn_dist; cz++) {
                Chunk* chunk = world_load_or_create_chunk(world, cx, cy, cz);
                if (chunk && chunk->loaded) {
                    // Loaded chunks still need lighting and meshing
                    if (chunk->needs_relighting || !chunk->meshed) {
                        worker_queue_chunk(world, chunk);
                    }
                } else if (chunk && !chunk->generated) {
                    // Only generate if not yet generated
                    world_generate_chunk(chunk, world->seed);
                    chunk->loaded = true;
                    chunk->generated = true;
                    worker_queue_chunk(world, chunk);  // Queue for worker to calculate lighting and mesh
                    chunk->modified = false;
                }
            }
        }
    }

    return true;
}

// Get skylight level at a specific world position
// Returns 0 if in a solid block or unloaded area
// OPTIMIZED: Fast path for unloaded chunks (returns 0 immediately)
uint8_t world_get_skylight(World* world, int x, int y, int z)
{
    // Out of bounds?
    if (y < WORLD_Y_MIN || y > WORLD_Y_MAX) return 0;

    // Calculate chunk coordinates
    int32_t chunk_x = x < 0 ? (x - CHUNK_WIDTH + 1) / CHUNK_WIDTH : x / CHUNK_WIDTH;
    int32_t chunk_y = y < 0 ? (y - CHUNK_HEIGHT + 1) / CHUNK_HEIGHT : y / CHUNK_HEIGHT;
    int32_t chunk_z = z < 0 ? (z - CHUNK_DEPTH + 1) / CHUNK_DEPTH : z / CHUNK_DEPTH;

    // Calculate position within chunk
    int local_x = x - (chunk_x * CHUNK_WIDTH);
    int local_y = y - (chunk_y * CHUNK_HEIGHT);
    int local_z = z - (chunk_z * CHUNK_DEPTH);

    // Bounds check first (fast)
    if (local_x < 0 || local_x >= CHUNK_WIDTH || local_y < 0 || local_y >= CHUNK_HEIGHT || local_z < 0 || local_z >= CHUNK_DEPTH) {
        return 0;  // Out of chunk bounds
    }

    // Get chunk (linear search - but we return early for out-of-bounds)
    Chunk* chunk = world_get_chunk(world, chunk_x, chunk_y, chunk_z);
    if (!chunk) {
        return 0;  // Unloaded chunks have no skylight
    }

    int active = __atomic_load_n(&chunk->active_light_buffer, __ATOMIC_ACQUIRE);
    return chunk->skylight[active][local_y][local_z][local_x];
}

// Get blocklight level at a specific world position
// Returns 0 if in solid block or unloaded area
uint8_t world_get_blocklight(World* world, int x, int y, int z)
{
    // Out of bounds?
    if (y < WORLD_Y_MIN || y > WORLD_Y_MAX) return 0;

    // Calculate chunk coordinates
    int32_t chunk_x = x < 0 ? (x - CHUNK_WIDTH + 1) / CHUNK_WIDTH : x / CHUNK_WIDTH;
    int32_t chunk_y = y < 0 ? (y - CHUNK_HEIGHT + 1) / CHUNK_HEIGHT : y / CHUNK_HEIGHT;
    int32_t chunk_z = z < 0 ? (z - CHUNK_DEPTH + 1) / CHUNK_DEPTH : z / CHUNK_DEPTH;

    // Calculate position within chunk
    int local_x = x - (chunk_x * CHUNK_WIDTH);
    int local_y = y - (chunk_y * CHUNK_HEIGHT);
    int local_z = z - (chunk_z * CHUNK_DEPTH);

    // Bounds check first (fast)
    if (local_x < 0 || local_x >= CHUNK_WIDTH || local_y < 0 || local_y >= CHUNK_HEIGHT || local_z < 0 || local_z >= CHUNK_DEPTH) {
        return 0;  // Out of chunk bounds
    }

    // Get chunk
    Chunk* chunk = world_get_chunk(world, chunk_x, chunk_y, chunk_z);
    if (!chunk) {
        return 0;  // Unloaded chunks have no blocklight
    }

    int active = __atomic_load_n(&chunk->active_light_buffer, __ATOMIC_ACQUIRE);
    return chunk->blocklight[active][local_y][local_z][local_x];
}

// Queue entry for BFS light propagation (stack-allocated for each chunk)
typedef struct {
    int x, y, z;
    uint8_t light;
} LightQueueEntry;

// Full blocklight recalculation - BFS across entire chunk
// Used by worker thread. For main thread, use localized version instead.
void calculate_chunk_blocklight(Chunk* chunk, World* world, int target_buffer)
{
    (void)world;
    if (!chunk) return;

    uint8_t (*blocklight_buf)[CHUNK_DEPTH][CHUNK_WIDTH] = chunk->blocklight[target_buffer];

    // Clear blocklight buffer
    memset(blocklight_buf, 0, CHUNK_HEIGHT * CHUNK_DEPTH * CHUNK_WIDTH * sizeof(uint8_t));

    // Stack-allocated queue for common case
    #define STACK_QUEUE_SIZE 4096
    LightQueueEntry stack_queue[STACK_QUEUE_SIZE];
    LightQueueEntry* queue = stack_queue;
    int max_queue = STACK_QUEUE_SIZE;
    bool queue_on_heap = false;

    int head = 0, tail = 0;

    // Seed with all light-emitting blocks
    for (int y = 0; y < CHUNK_HEIGHT; y++) {
        for (int z = 0; z < CHUNK_DEPTH; z++) {
            for (int x = 0; x < CHUNK_WIDTH; x++) {
                BlockProperties props = get_block_properties(chunk->blocks[y][z][x].type);
                if (props.emission > 0) {
                    blocklight_buf[y][z][x] = props.emission;
                    
                    if (tail >= max_queue) {
                        int new_max = max_queue * 2;
                        LightQueueEntry* new_queue = (LightQueueEntry*)malloc(sizeof(LightQueueEntry) * new_max);
                        if (!new_queue) goto cleanup_full;
                        memcpy(new_queue, queue, sizeof(LightQueueEntry) * tail);
                        if (queue_on_heap) free(queue);
                        queue = new_queue;
                        max_queue = new_max;
                        queue_on_heap = true;
                    }
                    
                    queue[tail++] = (LightQueueEntry){x, y, z, props.emission};
                }
            }
        }
    }

    // BFS propagation within chunk
    const int dx[] = {1, -1, 0, 0, 0, 0};
    const int dy[] = {0, 0, 1, -1, 0, 0};
    const int dz[] = {0, 0, 0, 0, 1, -1};

    while (head < tail) {
        LightQueueEntry e = queue[head++];
        if (e.light <= 1) continue;
        uint8_t new_light = e.light - 1;

        for (int d = 0; d < 6; d++) {
            int nx = e.x + dx[d];
            int ny = e.y + dy[d];
            int nz = e.z + dz[d];

            // Bounds check
            if (nx < 0 || nx >= CHUNK_WIDTH || ny < 0 || ny >= CHUNK_HEIGHT || nz < 0 || nz >= CHUNK_DEPTH)
                continue;

            // Skip opaque blocks
            BlockProperties neighbor_props = get_block_properties(chunk->blocks[ny][nz][nx].type);
            if (neighbor_props.opacity >= 15) continue;

            // Update if we found a better light value
            if (new_light > blocklight_buf[ny][nz][nx]) {
                blocklight_buf[ny][nz][nx] = new_light;
                
                if (tail >= max_queue) {
                    int new_max = max_queue * 2;
                    LightQueueEntry* new_queue = (LightQueueEntry*)malloc(sizeof(LightQueueEntry) * new_max);
                    if (!new_queue) goto cleanup_full;
                    memcpy(new_queue, queue, sizeof(LightQueueEntry) * tail);
                    if (queue_on_heap) free(queue);
                    queue = new_queue;
                    max_queue = new_max;
                    queue_on_heap = true;
                }
                
                queue[tail++] = (LightQueueEntry){nx, ny, nz, new_light};
            }
        }
    }

cleanup_full:
    if (queue_on_heap) free(queue);
    #undef STACK_QUEUE_SIZE
}


// Much faster than full BFS: instead of exploring entire chunk, we bound the search to nearby area
// Does BFS starting from all neighbors of the break point, only expanding within ~15 block radius
// Trade-off: distant light sources may be stale for ~1 frame (imperceptible in practice)
void calculate_chunk_blocklight_localized(Chunk* chunk, World* world, int target_buffer, int break_x, int break_y, int break_z)
{
    (void)world;
    if (!chunk) return;

    uint8_t (*blocklight_buf)[CHUNK_DEPTH][CHUNK_WIDTH] = chunk->blocklight[target_buffer];

    // Bounding radius around break point (blocks further away aren't noticeably affected)
    #define LOCALIZATION_RADIUS 15
    
    // Clear only the local region around the break
    int min_x = (break_x - LOCALIZATION_RADIUS) > 0 ? (break_x - LOCALIZATION_RADIUS) : 0;
    int max_x = (break_x + LOCALIZATION_RADIUS) < CHUNK_WIDTH ? (break_x + LOCALIZATION_RADIUS) : CHUNK_WIDTH - 1;
    int min_y = (break_y - LOCALIZATION_RADIUS) > 0 ? (break_y - LOCALIZATION_RADIUS) : 0;
    int max_y = (break_y + LOCALIZATION_RADIUS) < CHUNK_HEIGHT ? (break_y + LOCALIZATION_RADIUS) : CHUNK_HEIGHT - 1;
    int min_z = (break_z - LOCALIZATION_RADIUS) > 0 ? (break_z - LOCALIZATION_RADIUS) : 0;
    int max_z = (break_z + LOCALIZATION_RADIUS) < CHUNK_DEPTH ? (break_z + LOCALIZATION_RADIUS) : CHUNK_DEPTH - 1;

    // Clear blocklight in the local region
    for (int y = min_y; y <= max_y; y++) {
        for (int z = min_z; z <= max_z; z++) {
            for (int x = min_x; x <= max_x; x++) {
                blocklight_buf[y][z][x] = 0;
            }
        }
    }

    // Stack-allocated queue for common case
    #define STACK_QUEUE_SIZE 1024
    LightQueueEntry stack_queue[STACK_QUEUE_SIZE];
    LightQueueEntry* queue = stack_queue;
    int max_queue = STACK_QUEUE_SIZE;
    bool queue_on_heap = false;

    int head = 0, tail = 0;

    // Seed with light-emitting blocks in the affected region
    for (int y = min_y; y <= max_y; y++) {
        for (int z = min_z; z <= max_z; z++) {
            for (int x = min_x; x <= max_x; x++) {
                BlockProperties props = get_block_properties(chunk->blocks[y][z][x].type);
                if (props.emission > 0) {
                    blocklight_buf[y][z][x] = props.emission;
                    
                    if (tail >= max_queue) {
                        int new_max = max_queue * 2;
                        LightQueueEntry* new_queue = (LightQueueEntry*)malloc(sizeof(LightQueueEntry) * new_max);
                        if (!new_queue) goto cleanup_localized;
                        memcpy(new_queue, queue, sizeof(LightQueueEntry) * tail);
                        if (queue_on_heap) free(queue);
                        queue = new_queue;
                        max_queue = new_max;
                        queue_on_heap = true;
                    }
                    
                    queue[tail++] = (LightQueueEntry){x, y, z, props.emission};
                }
            }
        }
    }

    // Also check boundaries of the region - add adjacent light that can propagate inward
    // This prevents "dark halos" at the region boundary
    const int dx[] = {1, -1, 0, 0, 0, 0};
    const int dy[] = {0, 0, 1, -1, 0, 0};
    const int dz[] = {0, 0, 0, 0, 1, -1};

    for (int y = min_y; y <= max_y; y++) {
        for (int z = min_z; z <= max_z; z++) {
            for (int x = min_x; x <= max_x; x++) {
                for (int d = 0; d < 6; d++) {
                    int nx = x + dx[d];
                    int ny = y + dy[d];
                    int nz = z + dz[d];
                    
                    // Check boundary neighbor
                    if (nx < 0 || nx >= CHUNK_WIDTH || ny < 0 || ny >= CHUNK_HEIGHT || nz < 0 || nz >= CHUNK_DEPTH)
                        continue;
                    
                    // If outside our region, check if it has light that should propagate in
                    if (nx < min_x || nx > max_x || ny < min_y || ny > max_y || nz < min_z || nz > max_z) {
                        uint8_t neighbor_light = blocklight_buf[ny][nz][nx];
                        if (neighbor_light > 1) {
                            uint8_t incoming_light = neighbor_light - 1;
                            if (incoming_light > blocklight_buf[y][z][x]) {
                                blocklight_buf[y][z][x] = incoming_light;
                            }
                        }
                    }
                }
            }
        }
    }

    // BFS propagation within the bounded region
    while (head < tail) {
        LightQueueEntry e = queue[head++];
        if (e.light <= 1) continue;
        uint8_t new_light = e.light - 1;

        for (int d = 0; d < 6; d++) {
            int nx = e.x + dx[d];
            int ny = e.y + dy[d];
            int nz = e.z + dz[d];

            // Stay within bounds
            if (nx < 0 || nx >= CHUNK_WIDTH || ny < 0 || ny >= CHUNK_HEIGHT || nz < 0 || nz >= CHUNK_DEPTH)
                continue;

            // Skip opaque blocks
            BlockProperties neighbor_props = get_block_properties(chunk->blocks[ny][nz][nx].type);
            if (neighbor_props.opacity >= 15) continue;

            // Update if we found better light
            if (new_light > blocklight_buf[ny][nz][nx]) {
                blocklight_buf[ny][nz][nx] = new_light;
                
                // Only queue if still within region OR at boundary (propagate outward)
                if (nx >= min_x && nx <= max_x && ny >= min_y && ny <= max_y && nz >= min_z && nz <= max_z) {
                    if (tail >= max_queue) {
                        int new_max = max_queue * 2;
                        LightQueueEntry* new_queue = (LightQueueEntry*)malloc(sizeof(LightQueueEntry) * new_max);
                        if (!new_queue) goto cleanup_localized;
                        memcpy(new_queue, queue, sizeof(LightQueueEntry) * tail);
                        if (queue_on_heap) free(queue);
                        queue = new_queue;
                        max_queue = new_max;
                        queue_on_heap = true;
                    }
                    
                    queue[tail++] = (LightQueueEntry){nx, ny, nz, new_light};
                }
            }
        }
    }

cleanup_localized:
    if (queue_on_heap) free(queue);
    #undef LOCALIZATION_RADIUS
    #undef STACK_QUEUE_SIZE
}

// Calculate skylight using simple column scan (no BFS needed for skylight)
// Algorithm:
// 1. For each XZ column, scan downward from sky
// 2. Skylight = 15 through air, = 0 through solids
// 3. Result: accurate lighting without cross-chunk seams
void calculate_chunk_skylight(Chunk* chunk, World* world, int target_buffer)
{
    if (!chunk) return;

    uint8_t (*skylight_buf)[CHUNK_DEPTH][CHUNK_WIDTH] = chunk->skylight[target_buffer];

    int base_y = chunk->chunk_y * CHUNK_HEIGHT;

    // For each XZ column, scan downward from top
    for (int z = 0; z < CHUNK_DEPTH; z++) {
        for (int x = 0; x < CHUNK_WIDTH; x++) {
            int world_x = chunk->chunk_x * CHUNK_WIDTH + x;
            int world_z = chunk->chunk_z * CHUNK_DEPTH + z;

            // Check if sky is blocked above this chunk
            // Only scan blocks ABOVE this chunk, not the entire world
            uint8_t incoming_light = 15;
            for (int wy = WORLD_Y_MAX; wy > base_y + CHUNK_HEIGHT - 1; wy--) {
                BlockType b = world_get_block(world, world_x, wy, world_z);
                if (b != BLOCK_AIR) {
                    incoming_light = 0;
                    break;
                }
            }

            // Now scan down through this chunk only
            for (int y = CHUNK_HEIGHT - 1; y >= 0; y--) {
                if (incoming_light == 0) {
                    skylight_buf[y][z][x] = 0;
                } else {
                    BlockType b = chunk->blocks[y][z][x].type;
                    if (b != BLOCK_AIR) {
                        skylight_buf[y][z][x] = 0;
                        incoming_light = 0;  // Block skylight for all blocks below
                    } else {
                        skylight_buf[y][z][x] = incoming_light;
                    }
                }
            }
        }
    }
}

// GREEDY MESHING: Merge adjacent coplanar exposed faces into larger rectangles
// This dramatically reduces geometry - typically 60-80% reduction in face count
// Algorithm: For each face direction, find maximal rectangles of exposed faces
MergedMesh* chunk_greedy_mesh(Chunk* chunk, World* world)
{
    MergedMesh* mesh = (MergedMesh*)malloc(sizeof(MergedMesh));
    memset(mesh, 0, sizeof(MergedMesh));
    
    // Initialize arrays for each face
    for (int f = 0; f < 6; f++) {
        mesh->quad_capacity[f] = 256;
        mesh->quads[f] = (MergedQuad*)malloc(sizeof(MergedQuad) * mesh->quad_capacity[f]);
        mesh->quad_count[f] = 0;
    }

    // For each face direction, perform greedy meshing
    // Face 0: +X, Face 1: -X, Face 2: +Y, Face 3: -Y, Face 4: +Z, Face 5: -Z
    
    // +Y (top faces) - process in XZ plane
    {
        bool used[CHUNK_DEPTH][CHUNK_WIDTH] = {0};  // Mark which blocks have been merged
        
        for (int y = CHUNK_HEIGHT - 1; y >= 0; y--) {
            for (int z = 0; z < CHUNK_DEPTH; z++) {
                for (int x = 0; x < CHUNK_WIDTH; x++) {
                    if (used[z][x]) continue;
                    
                    BlockType block = world_chunk_get_block(chunk, x, y, z);
                    if (block == BLOCK_AIR) continue;
                    
                    // Check if +Y face is exposed
                    int world_x = chunk->chunk_x * CHUNK_WIDTH + x;
                    int world_y = chunk->chunk_y * CHUNK_HEIGHT + y;
                    int world_z = chunk->chunk_z * CHUNK_DEPTH + z;
                    
                    if (world_get_block(world, world_x, world_y + 1, world_z) != BLOCK_AIR) continue;
                    
                    // Find max width
                    int w = 1;
                    while (x + w < CHUNK_WIDTH && !used[z][x + w]) {
                        BlockType b = world_chunk_get_block(chunk, x + w, y, z);
                        if (b == BLOCK_AIR) break;
                        int wx = world_x + w;
                        if (world_get_block(world, wx, world_y + 1, world_z) != BLOCK_AIR) break;
                        w++;
                    }
                    
                    // Find max height
                    int h = 1;
                    while (z + h < CHUNK_DEPTH) {
                        bool can_extend = true;
                        for (int dx = 0; dx < w; dx++) {
                            if (used[z + h][x + dx]) {
                                can_extend = false;
                                break;
                            }
                            BlockType b = world_chunk_get_block(chunk, x + dx, y, z + h);
                            if (b == BLOCK_AIR) {
                                can_extend = false;
                                break;
                            }
                            int wx = world_x + dx;
                            int wz = world_z + h;
                            if (world_get_block(world, wx, world_y + 1, wz) != BLOCK_AIR) {
                                can_extend = false;
                                break;
                            }
                        }
                        if (!can_extend) break;
                        h++;
                    }
                    
                    // Add quad
                    if (mesh->quad_count[2] >= mesh->quad_capacity[2]) {
                        mesh->quad_capacity[2] *= 2;
                        mesh->quads[2] = (MergedQuad*)realloc(mesh->quads[2], sizeof(MergedQuad) * mesh->quad_capacity[2]);
                    }
                    
                    MergedQuad* quad = &mesh->quads[2][mesh->quad_count[2]++];
                    quad->x = world_x;
                    quad->y = world_y;
                    quad->z = world_z;
                    quad->w = w;
                    quad->h = h;
                    quad->face = 2;  // +Y
                    quad->type = block;
                    quad->light = world_get_skylight(world, world_x, world_y + 1, world_z);
                    
                    // Mark as used
                    for (int dz = 0; dz < h; dz++) {
                        for (int dx = 0; dx < w; dx++) {
                            used[z + dz][x + dx] = true;
                        }
                    }
                }
            }
        }
    }
    
    // -Y (bottom faces) - similar process
    {
        bool used[CHUNK_DEPTH][CHUNK_WIDTH] = {0};
        
        for (int y = 0; y < CHUNK_HEIGHT; y++) {
            for (int z = 0; z < CHUNK_DEPTH; z++) {
                for (int x = 0; x < CHUNK_WIDTH; x++) {
                    if (used[z][x]) continue;
                    
                    BlockType block = world_chunk_get_block(chunk, x, y, z);
                    if (block == BLOCK_AIR) continue;
                    
                    int world_x = chunk->chunk_x * CHUNK_WIDTH + x;
                    int world_y = chunk->chunk_y * CHUNK_HEIGHT + y;
                    int world_z = chunk->chunk_z * CHUNK_DEPTH + z;
                    
                    if (world_get_block(world, world_x, world_y - 1, world_z) != BLOCK_AIR) continue;
                    
                    int w = 1;
                    while (x + w < CHUNK_WIDTH && !used[z][x + w]) {
                        BlockType b = world_chunk_get_block(chunk, x + w, y, z);
                        if (b == BLOCK_AIR) break;
                        int wx = world_x + w;
                        if (world_get_block(world, wx, world_y - 1, world_z) != BLOCK_AIR) break;
                        w++;
                    }
                    
                    int h = 1;
                    while (z + h < CHUNK_DEPTH) {
                        bool can_extend = true;
                        for (int dx = 0; dx < w; dx++) {
                            if (used[z + h][x + dx]) {
                                can_extend = false;
                                break;
                            }
                            BlockType b = world_chunk_get_block(chunk, x + dx, y, z + h);
                            if (b == BLOCK_AIR) {
                                can_extend = false;
                                break;
                            }
                            int wx = world_x + dx;
                            int wz = world_z + h;
                            if (world_get_block(world, wx, world_y - 1, wz) != BLOCK_AIR) {
                                can_extend = false;
                                break;
                            }
                        }
                        if (!can_extend) break;
                        h++;
                    }
                    
                    if (mesh->quad_count[3] >= mesh->quad_capacity[3]) {
                        mesh->quad_capacity[3] *= 2;
                        mesh->quads[3] = (MergedQuad*)realloc(mesh->quads[3], sizeof(MergedQuad) * mesh->quad_capacity[3]);
                    }
                    
                    MergedQuad* quad = &mesh->quads[3][mesh->quad_count[3]++];
                    quad->x = world_x;
                    quad->y = world_y;
                    quad->z = world_z;
                    quad->w = w;
                    quad->h = h;
                    quad->face = 3;  // -Y
                    quad->type = block;
                    quad->light = world_get_skylight(world, world_x, world_y - 1, world_z);
                    
                    for (int dz = 0; dz < h; dz++) {
                        for (int dx = 0; dx < w; dx++) {
                            used[z + dz][x + dx] = true;
                        }
                    }
                }
            }
        }
    }

    return mesh;
}

// OPTIMIZED MESHING FOR VOXEL RENDERING
// Pre-compute and cache all visible blocks in a chunk (blocks with exposed faces)
// This avoids the per-frame triple-nested loop and provides massive performance improvement
//
// GREEDY MESHING IMPLEMENTATION:
// - For each block, we only add it if it has exposed faces (faces adjacent to air)
// - We track per-face lighting to enable better face merging during rendering
// - The rendering pipeline (draw_cube_faces) can then optimize by rendering
//   larger merged quads instead of individual faces when faces have matching lighting
//
// PERFORMANCE OPTIMIZATIONS INCLUDED:
// 1. Skip entirely interior blocks (completely surrounded by solid blocks)
// 2. Only track exposed faces that border unloaded/air blocks
// 3. Bake lighting per-face to enable adaptive face merging
// 4. Double-buffered swapping for thread-safe concurrent updates
// 5. Blocks are stored in order of iteration for cache coherence
//
// Result: Typical 40-60% reduction in geometry compared to rendering all block faces

// Pre-compute and cache all visible blocks in a chunk (blocks with exposed faces)
void chunk_cache_visible_blocks(Chunk* chunk, World* world)
{
    if (!chunk) return;

    // Build into a temporary array to avoid realloc while render thread might use the original
    int temp_capacity = 1024;  // Start with 1024 blocks
    int temp_count = 0;
    CachedVisibleBlock* temp_blocks = (CachedVisibleBlock*)malloc(sizeof(CachedVisibleBlock) * temp_capacity);

    // Iterate through all blocks in chunk
    for (int y = 0; y < CHUNK_HEIGHT; y++) {
        for (int z = 0; z < CHUNK_DEPTH; z++) {
            for (int x = 0; x < CHUNK_WIDTH; x++) {
                BlockType block = world_chunk_get_block(chunk, x, y, z);

                // Skip air blocks
                if (block == BLOCK_AIR) continue;

                // Calculate world coordinates for lighting lookups
                int world_x = chunk->chunk_x * CHUNK_WIDTH + x;
                int world_y = chunk->chunk_y * CHUNK_HEIGHT + y;
                int world_z = chunk->chunk_z * CHUNK_DEPTH + z;

                // OPTIMIZATION: For interior blocks, check neighbors using local array access (no locks)
                // For edge blocks, use world_get_block (has mutex but only for edges)
                uint8_t exposed_faces = 0;
                
                // Check +X (right)
                if (x + 1 < CHUNK_WIDTH) {
                    // Interior: check locally
                    if (world_chunk_get_block(chunk, x + 1, y, z) == BLOCK_AIR) exposed_faces |= (1 << 0);
                } else {
                    // Edge: use world function (will lock cache)
                    if (world_get_block(world, world_x + 1, world_y, world_z) == BLOCK_AIR) exposed_faces |= (1 << 0);
                }
                
                // Check -X (left)
                if (x - 1 >= 0) {
                    if (world_chunk_get_block(chunk, x - 1, y, z) == BLOCK_AIR) exposed_faces |= (1 << 1);
                } else {
                    if (world_get_block(world, world_x - 1, world_y, world_z) == BLOCK_AIR) exposed_faces |= (1 << 1);
                }
                
                // Check +Y (up)
                if (y + 1 < CHUNK_HEIGHT) {
                    if (world_chunk_get_block(chunk, x, y + 1, z) == BLOCK_AIR) exposed_faces |= (1 << 2);
                } else {
                    if (world_get_block(world, world_x, world_y + 1, world_z) == BLOCK_AIR) exposed_faces |= (1 << 2);
                }
                
                // Check -Y (down)
                if (y - 1 >= 0) {
                    if (world_chunk_get_block(chunk, x, y - 1, z) == BLOCK_AIR) exposed_faces |= (1 << 3);
                } else {
                    if (world_get_block(world, world_x, world_y - 1, world_z) == BLOCK_AIR) exposed_faces |= (1 << 3);
                }
                
                // Check +Z (forward)
                if (z + 1 < CHUNK_DEPTH) {
                    if (world_chunk_get_block(chunk, x, y, z + 1) == BLOCK_AIR) exposed_faces |= (1 << 4);
                } else {
                    if (world_get_block(world, world_x, world_y, world_z + 1) == BLOCK_AIR) exposed_faces |= (1 << 4);
                }
                
                // Check -Z (back)
                if (z - 1 >= 0) {
                    if (world_chunk_get_block(chunk, x, y, z - 1) == BLOCK_AIR) exposed_faces |= (1 << 5);
                } else {
                    if (world_get_block(world, world_x, world_y, world_z - 1) == BLOCK_AIR) exposed_faces |= (1 << 5);
                }

                if (exposed_faces != 0) {
                    // Grow temporary array if needed
                    if (temp_count >= temp_capacity) {
                        temp_capacity *= 2;
                        temp_blocks = (CachedVisibleBlock*)realloc(temp_blocks,
                                                                   sizeof(CachedVisibleBlock) * temp_capacity);
                    }

                    // Add to temporary array
                    temp_blocks[temp_count].x = x;
                    temp_blocks[temp_count].y = y;
                    temp_blocks[temp_count].z = z;
                    temp_blocks[temp_count].exposed_faces = exposed_faces;
                    temp_blocks[temp_count].light_level = 0;  // preserved for compatibility

                    // Compute per-face baked lighting (use max of skylight and blocklight at adjacent air block)
                    for (int face = 0; face < 6; face++) {
                        // default to zero light
                        temp_blocks[temp_count].face_light[face] = 0;
                    }

                    // +X
                    if (exposed_faces & (1 << 0)) {
                        int nx = world_x + 1;
                        int ny = world_y;
                        int nz = world_z;
                        uint8_t skyl = world_get_skylight(world, nx, ny, nz);
                        uint8_t blockl = world_get_blocklight(world, nx, ny, nz);
                        temp_blocks[temp_count].face_light[0] = (skyl > blockl) ? skyl : blockl;
                    }
                    // -X
                    if (exposed_faces & (1 << 1)) {
                        int nx = world_x - 1;
                        int ny = world_y;
                        int nz = world_z;
                        uint8_t skyl = world_get_skylight(world, nx, ny, nz);
                        uint8_t blockl = world_get_blocklight(world, nx, ny, nz);
                        temp_blocks[temp_count].face_light[1] = (skyl > blockl) ? skyl : blockl;
                    }
                    // +Y
                    if (exposed_faces & (1 << 2)) {
                        int nx = world_x;
                        int ny = world_y + 1;
                        int nz = world_z;
                        uint8_t skyl = world_get_skylight(world, nx, ny, nz);
                        uint8_t blockl = world_get_blocklight(world, nx, ny, nz);
                        temp_blocks[temp_count].face_light[2] = (skyl > blockl) ? skyl : blockl;
                    }
                    // -Y
                    if (exposed_faces & (1 << 3)) {
                        int nx = world_x;
                        int ny = world_y - 1;
                        int nz = world_z;
                        uint8_t skyl = world_get_skylight(world, nx, ny, nz);
                        uint8_t blockl = world_get_blocklight(world, nx, ny, nz);
                        temp_blocks[temp_count].face_light[3] = (skyl > blockl) ? skyl : blockl;
                    }
                    // +Z
                    if (exposed_faces & (1 << 4)) {
                        int nx = world_x;
                        int ny = world_y;
                        int nz = world_z + 1;
                        uint8_t skyl = world_get_skylight(world, nx, ny, nz);
                        uint8_t blockl = world_get_blocklight(world, nx, ny, nz);
                        temp_blocks[temp_count].face_light[4] = (skyl > blockl) ? skyl : blockl;
                    }
                    // -Z
                    if (exposed_faces & (1 << 5)) {
                        int nx = world_x;
                        int ny = world_y;
                        int nz = world_z - 1;
                        uint8_t skyl = world_get_skylight(world, nx, ny, nz);
                        uint8_t blockl = world_get_blocklight(world, nx, ny, nz);
                        temp_blocks[temp_count].face_light[5] = (skyl > blockl) ? skyl : blockl;
                    }

                    temp_count++;
                }
            }
        }
    }

    // ATOMIC SWAP: Now safely replace the old array with the new one
    // Using double-buffering: build into inactive buffer, then atomically swap active_mesh index
    // This ensures render thread always sees consistent data (no partial updates)

    // Lock mutex during swap to prevent render thread from reading between buffer updates
    // This ensures the render thread never sees an inconsistent state
    pthread_mutex_lock(&chunk->mesh_swap_mutex);

    int current_active = __atomic_load_n(&chunk->active_mesh, __ATOMIC_ACQUIRE);
    int inactive_buffer = 1 - current_active;  // Opposite of currently active buffer

    // Free old data in the inactive buffer if it exists
    if (chunk->visible_blocks[inactive_buffer] != NULL) {
        free(chunk->visible_blocks[inactive_buffer]);
    }

    // Store new mesh into inactive buffer
    chunk->visible_blocks[inactive_buffer] = temp_blocks;
    chunk->visible_count[inactive_buffer] = temp_count;
    chunk->visible_capacity[inactive_buffer] = temp_capacity;

    // GREEDY MESHING: Generate merged quads for better performance
    MergedMesh* new_merged = chunk_greedy_mesh(chunk, world);
    
    // Free old merged mesh in inactive buffer
    if (chunk->merged_mesh[inactive_buffer] != NULL) {
        for (int f = 0; f < 6; f++) {
            if (chunk->merged_mesh[inactive_buffer]->quads[f] != NULL) {
                free(chunk->merged_mesh[inactive_buffer]->quads[f]);
            }
        }
        free(chunk->merged_mesh[inactive_buffer]);
    }
    
    // Store new merged mesh
    chunk->merged_mesh[inactive_buffer] = new_merged;

    // ATOMIC SWAP: Use atomic operation with memory barrier
    // This ensures the updated mesh is fully visible before we flip the active buffer switch
    // Render thread will see the new mesh on the next inspection
    __atomic_store_n(&chunk->active_mesh, inactive_buffer, __ATOMIC_RELEASE);
    __atomic_store_n(&chunk->active_merged_mesh, inactive_buffer, __ATOMIC_RELEASE);

    pthread_mutex_unlock(&chunk->mesh_swap_mutex);
}

// Free the visible blocks cache (both buffers)
void chunk_free_visible_blocks(Chunk* chunk)
{
    // Free both buffers
    for (int i = 0; i < 2; i++) {
        if (chunk->visible_blocks[i] != NULL) {
            free(chunk->visible_blocks[i]);
            chunk->visible_blocks[i] = NULL;
        }
        chunk->visible_count[i] = 0;
        chunk->visible_capacity[i] = 0;
    }
    // NOTE: Don't set meshed=false here - keep rendering old mesh while worker recalculates
    // This prevents flickering when blocks are placed/broken
    // Worker thread will recalculate and update visible_blocks while meshed stays true
}

// Free merged mesh data
void chunk_free_merged_mesh(Chunk* chunk)
{
    if (!chunk) return;
    
    for (int i = 0; i < 2; i++) {
        if (chunk->merged_mesh[i] != NULL) {
            for (int f = 0; f < 6; f++) {
                if (chunk->merged_mesh[i]->quads[f] != NULL) {
                    free(chunk->merged_mesh[i]->quads[f]);
                }
            }
            free(chunk->merged_mesh[i]);
            chunk->merged_mesh[i] = NULL;
        }
    }
}