#include "block_renderer.h"
#include "block_core.h"
#include <stdlib.h>
#include <string.h>
#include <math.h>

// 5x7 位图字体 (列优先, LSB 在上)
#include "block_font.inc"

int block_renderer_init(BlockRenderer* renderer, void* device, void* context) {
    memset(renderer, 0, sizeof(BlockRenderer));
    renderer->device = (ID3D11Device*)device;
    renderer->context = (ID3D11DeviceContext*)context;
    renderer->vertex_capacity = 65536;
    renderer->vertex_data = (float*)malloc(renderer->vertex_capacity * 7 * sizeof(float));
    renderer->index_capacity = 65536;
    renderer->index_data = (uint32_t*)malloc(renderer->index_capacity * sizeof(uint32_t));
    return 1;
}

void block_renderer_shutdown(BlockRenderer* renderer) {
    if (renderer->vertex_data) free(renderer->vertex_data);
    if (renderer->index_data) free(renderer->index_data);
    if (renderer->vertex_buffer) renderer->vertex_buffer->lpVtbl->Release(renderer->vertex_buffer);
    if (renderer->index_buffer) renderer->index_buffer->lpVtbl->Release(renderer->index_buffer);
    if (renderer->vertex_shader) renderer->vertex_shader->lpVtbl->Release(renderer->vertex_shader);
    if (renderer->pixel_shader) renderer->pixel_shader->lpVtbl->Release(renderer->pixel_shader);
    if (renderer->input_layout) renderer->input_layout->lpVtbl->Release(renderer->input_layout);
    memset(renderer, 0, sizeof(BlockRenderer));
}

void block_renderer_begin(BlockRenderer* renderer) {
    renderer->vertex_count = 0;
    renderer->index_count = 0;
}

void block_renderer_end(BlockRenderer* renderer) {
    if (renderer->vertex_count == 0) return;
    ID3D11DeviceContext* ctx = renderer->context;

    D3D11_MAPPED_SUBRESOURCE mapped;
    if (SUCCEEDED(ctx->lpVtbl->Map(ctx, (ID3D11Resource*)renderer->vertex_buffer, 0, D3D11_MAP_WRITE_DISCARD, 0, &mapped))) {
        memcpy(mapped.pData, renderer->vertex_data, renderer->vertex_count * 7 * sizeof(float));
        ctx->lpVtbl->Unmap(ctx, (ID3D11Resource*)renderer->vertex_buffer, 0);
    }

    if (renderer->index_count > 0) {
        if (SUCCEEDED(ctx->lpVtbl->Map(ctx, (ID3D11Resource*)renderer->index_buffer, 0, D3D11_MAP_WRITE_DISCARD, 0, &mapped))) {
            memcpy(mapped.pData, renderer->index_data, renderer->index_count * sizeof(uint32_t));
            ctx->lpVtbl->Unmap(ctx, (ID3D11Resource*)renderer->index_buffer, 0);
        }
    }

    UINT stride = 7 * sizeof(float);
    UINT offset = 0;
    ID3D11Buffer* vb = renderer->vertex_buffer;
    ctx->lpVtbl->IASetVertexBuffers(ctx, 0, 1, &vb, &stride, &offset);

    if (renderer->index_buffer) {
        ctx->lpVtbl->IASetIndexBuffer(ctx, renderer->index_buffer, DXGI_FORMAT_R32_UINT, 0);
    }
    ctx->lpVtbl->IASetInputLayout(ctx, renderer->input_layout);
    ctx->lpVtbl->IASetPrimitiveTopology(ctx, D3D11_PRIMITIVE_TOPOLOGY_TRIANGLELIST);
    ctx->lpVtbl->VSSetShader(ctx, renderer->vertex_shader, NULL, 0);
    ctx->lpVtbl->PSSetShader(ctx, renderer->pixel_shader, NULL, 0);

    if (renderer->index_count > 0) {
        ctx->lpVtbl->DrawIndexed(ctx, renderer->index_count, 0, 0);
    } else {
        ctx->lpVtbl->Draw(ctx, renderer->vertex_count, 0);
    }
}

void block_renderer_add_quad(BlockRenderer* renderer, float x, float y, float w, float h, float r, float g, float b, float a) {
    if (renderer->vertex_count + 4 > renderer->vertex_capacity) return;
    if (renderer->index_count + 6 > renderer->index_capacity) return;

    uint32_t base = renderer->vertex_count;
    float* v = renderer->vertex_data + base * 7;

    v[0*7+0] = x;     v[0*7+1] = y;     v[0*7+2] = 0; v[0*7+3] = r; v[0*7+4] = g; v[0*7+5] = b; v[0*7+6] = a;
    v[1*7+0] = x + w; v[1*7+1] = y;     v[1*7+2] = 0; v[1*7+3] = r; v[1*7+4] = g; v[1*7+5] = b; v[1*7+6] = a;
    v[2*7+0] = x + w; v[2*7+1] = y + h; v[2*7+2] = 0; v[2*7+3] = r; v[2*7+4] = g; v[2*7+5] = b; v[2*7+6] = a;
    v[3*7+0] = x;     v[3*7+1] = y + h; v[3*7+2] = 0; v[3*7+3] = r; v[3*7+4] = g; v[3*7+5] = b; v[3*7+6] = a;

    renderer->vertex_count += 4;

    renderer->index_data[renderer->index_count++] = base + 0;
    renderer->index_data[renderer->index_count++] = base + 1;
    renderer->index_data[renderer->index_count++] = base + 2;
    renderer->index_data[renderer->index_count++] = base + 0;
    renderer->index_data[renderer->index_count++] = base + 2;
    renderer->index_data[renderer->index_count++] = base + 3;
}

void block_renderer_add_line(BlockRenderer* renderer, float x1, float y1, float x2, float y2, float r, float g, float b, float a) {
    float dx = x2 - x1;
    float dy = y2 - y1;
    float len = sqrtf(dx * dx + dy * dy);
    if (len < 0.001f) return;

    const float thickness = 1.0f;
    float nx = -dy / len * thickness * 0.5f;
    float ny = dx / len * thickness * 0.5f;

    if (renderer->vertex_count + 4 > renderer->vertex_capacity) return;
    if (renderer->index_count + 6 > renderer->index_capacity) return;

    uint32_t base = renderer->vertex_count;
    float* v = renderer->vertex_data + base * 7;

    v[0*7+0] = x1 + nx; v[0*7+1] = y1 + ny; v[0*7+2] = 0; v[0*7+3] = r; v[0*7+4] = g; v[0*7+5] = b; v[0*7+6] = a;
    v[1*7+0] = x2 + nx; v[1*7+1] = y2 + ny; v[1*7+2] = 0; v[1*7+3] = r; v[1*7+4] = g; v[1*7+5] = b; v[1*7+6] = a;
    v[2*7+0] = x2 - nx; v[2*7+1] = y2 - ny; v[2*7+2] = 0; v[2*7+3] = r; v[2*7+4] = g; v[2*7+5] = b; v[2*7+6] = a;
    v[3*7+0] = x1 - nx; v[3*7+1] = y1 - ny; v[3*7+2] = 0; v[3*7+3] = r; v[3*7+4] = g; v[3*7+5] = b; v[3*7+6] = a;

    renderer->vertex_count += 4;

    renderer->index_data[renderer->index_count++] = base + 0;
    renderer->index_data[renderer->index_count++] = base + 1;
    renderer->index_data[renderer->index_count++] = base + 2;
    renderer->index_data[renderer->index_count++] = base + 0;
    renderer->index_data[renderer->index_count++] = base + 2;
    renderer->index_data[renderer->index_count++] = base + 3;
}

static void block_renderer_add_bezier(BlockRenderer* renderer, float x1, float y1, float cx1, float cy1, float cx2, float cy2, float x2, float y2, float r, float g, float b, float a) {
    int segments = 16;
    float prev_x = x1, prev_y = y1;
    for (int i = 1; i <= segments; i++) {
        float t = (float)i / segments;
        float mt = 1.0f - t;
        float px = mt*mt*mt*x1 + 3*mt*mt*t*cx1 + 3*mt*t*t*cx2 + t*t*t*x2;
        float py = mt*mt*mt*y1 + 3*mt*mt*t*cy1 + 3*mt*t*t*cy2 + t*t*t*y2;
        block_renderer_add_line(renderer, prev_x, prev_y, px, py, r, g, b, a);
        prev_x = px;
        prev_y = py;
    }
}

void block_renderer_draw_block(BlockRenderer* renderer, Block* block, float view_x, float view_y, float zoom) {
    float x = (block->x - view_x) * zoom;
    float y = (block->y - view_y) * zoom;
    float w = block->width * zoom;
    float h = block->height * zoom;

    block_renderer_add_quad(renderer, x, y, w, h, 0.3f, 0.3f, 0.35f, 1.0f);
    block_renderer_add_quad(renderer, x, y, w, 20 * zoom, 0.2f, 0.2f, 0.25f, 1.0f);

    if (block->label[0]) {
        block_renderer_draw_text(renderer, block->label, x + 4.0f * zoom, y + 4.0f * zoom, 0.9f, 0.9f, 0.9f);
    }
}

void block_renderer_draw_connection(BlockRenderer* renderer, BlockSlot* from, BlockSlot* to, float view_x, float view_y, float zoom) {
    float x1 = (from->owner->x + from->x - view_x) * zoom;
    float y1 = (from->owner->y + from->y - view_y) * zoom;
    float x2 = (to->owner->x + to->x - view_x) * zoom;
    float y2 = (to->owner->y + to->y - view_y) * zoom;

    float dx = fabsf(x2 - x1) * 0.5f;
    if (dx < 50.0f) dx = 50.0f;

    block_renderer_add_bezier(renderer, x1, y1, x1 + dx, y1, x2 - dx, y2, x2, y2, 0.8f, 0.8f, 0.2f, 1.0f);
}

void block_renderer_draw_text(BlockRenderer* renderer, const char* text, float x, float y, float r, float g, float b) {
    if (!text) return;
    const float scale = 2.0f;
    const float advance = 5.0f * scale + 1.0f;
    float cx = x;
    for (const char* p = text; *p; ++p) {
        unsigned char ch = (unsigned char)*p;
        if (ch < 32 || ch > 127) ch = '?';
        int idx = ch - 32;
        for (int col = 0; col < 5; ++col) {
            uint8_t column = BLOCK_FONT[idx][col];
            for (int row = 0; row < 7; ++row) {
                if (column & (1u << row)) {
                    block_renderer_add_quad(renderer,
                        cx + col * scale,
                        y + row * scale,
                        scale, scale,
                        r, g, b, 1.0f);
                }
            }
        }
        cx += advance;
    }
}