#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <stdbool.h>

#include <wchar.h>

#include <assert.h>

#include <unistd.h>
#include <fcntl.h>
#include <math.h>

#include <X11/Xlib.h>
#include <X11/Xutil.h>

#include <sys/ioctl.h>
#include <sys/mman.h>
#include <sys/types.h>
#include <sys/stat.h>

#define ASSERT(expr) if (!expr) { fprintf(stderr, "[ASSERT] Assertion fail %s:%d\n", __FILE__, __LINE__); abort(); }

#define MIN(a, b) ((a) < (b) ? (a) : (b))
#define MAX(a, b) ((a) > (b) ? (a) : (b))

#define MAX_INTERSECTIONS 16
#define F32EPS 1e-5f
#define MAX_CONTOURS 256

#define TTF_ENGINE_STACK 4096

typedef int64_t s64;
typedef int32_t s32;
typedef int16_t s16;
typedef int8_t  s8;

typedef uint64_t u64;
typedef uint32_t u32;
typedef uint16_t u16;
typedef uint8_t  u8;

typedef float  f32;
typedef double f64;


/************* TTF *************/

struct v2 {
    int x;
    int y;
};

struct v2u32 {
    u32 x;
    u32 y;
};

struct v2f {
    f32 x;
    f32 y;
};

struct line {
    struct v2f a;
    struct v2f b;
};

struct line_contour {
    int ncontours;
    int from[MAX_CONTOURS];
    struct line *data;
};

struct intersection {
    f32 x;
    int dir;
};

enum rgb_channel {
    RED = 0x000000FF,
    GREEN = 0x0000FF00,
    BLUE = 0x00FF0000,
};

struct font_buffer { 
    u8 *data; 
    u64 offset; 
    u64 size; 
};

struct font_directory { 
    u32 cmap_offset;
    u32 head_offset;
    u32 hhea_offset;
    u32 loca_offset;
    u32 glyf_offset;
    u32 hmtx_offset;
    u32 maxp_offset;
    u32 post_offset;
    
    u32 cvt_offset;
    u32 cvt_size;
    
    u32 prep_offset;
    u32 prep_size;
    
    u32 fpgm_offset;
    u32 fpgm_size;
};

union glyph_flag {
    struct {
        u8 on_curve : 1;
        u8 x_short : 1;
        u8 y_short : 1;
        u8 repeat : 1;
        u8 x_short_pos : 1;
        u8 y_short_pos : 1;
        u8 reserved1 : 1;
        u8 reserved2 : 1;
    };
    u8 flag;
};

enum compund_glyph_flag {
    ARG_1_AND_2_ARE_WORDS    = 0x1,
    ARGS_ARE_XY_VALUES       = 0x2,
    ROUND_XY_TO_GRID         = 0x4,
    WE_HAVE_A_SCALE          = 0x8,
    
    MORE_COMPONENTS          = 0x20,
    WE_HAVE_AN_X_AND_Y_SCALE = 0x40,
    WE_HAVE_A_TWO_BY_TWO     = 0x80,
    WE_HAVE_INSTRUCTIONS     = 0x100,
    USE_MY_METRICS           = 0x200,
    OVERLAP_COMPOUND         = 0x400
};

struct maxp_table {
    u16 max_component_points;
    u16 max_component_contours;
    u16 max_storage;
    u16 max_fdefs;
};

struct head_table {
    int itl_format;
    int units_per_em;
};

struct hhea_table {
    int ascent;
    int descent;
    int line_gap;
    int max_advance;
};

struct glyph_point {
    s16 x;
    s16 y;
    bool on_curve;
};

struct glyph_segment {
    bool is_curve;
    struct v2f p0;
    struct v2f p1;
    struct v2f p2;
};

struct glyph {
    s16 xmin, ymin;
    s16 xmax, ymax;
    
    s32 baseline;
    u16 advance;
    s16 lsb;
    
    u16 ncontours;
    u16 *end_pts_of_contours;
    struct glyph_point *points;
};

struct ttf_font {
    struct font_buffer file;
    
    struct maxp_table maxp;
    struct head_table head;
    struct hhea_table hhea;
    
    s16 *cvt;
    int cmap_format;
    int is_monospace;
    
    struct font_directory dir;
    struct ttf_function *functions;
    
    u32 *storage;
    
    char *name;
};

/*******************************/
static int
clamp_u32(u32 val, u32 cap) 
{
    u32 result = val;
    if (val > cap) result = cap;
    return(result);
}

static f32
abs_f32(f32 v)
{
    f32 result = (v > 0 ? v : -v);
    return(result);
}

static int
round_f32(f32 v)
{
#if 1
    int towards_zero = (int) v;
    f32 diff = abs_f32(v - towards_zero);
    
    if (diff >= 0.5f) {
        return(v > 0 ? towards_zero + 1 : towards_zero - 1);
    } else {
        return(towards_zero);
    }
#else
    return(v);
#endif
}

static int
ceil_f32(f32 v)
{
    int trunc = v;
    if (v - trunc < F32EPS) {
        return(trunc);
    }
    return(trunc + 1);
}


#include "ttf_engine.c"
#include "ttf2.c"

static Display *display;
static Window window;
static GC default_gc;
static XImage* xwindow_buffer;

static int
scanline_intersects_line(f32 y, struct v2f p0, struct v2f p1, f32 lasty, f32 *vx)
{
    bool goes_up = (p0.y > p1.y);
    f32 t1 = (y - p0.y) / (p1.y - p0.y); /* NOTE(aolo2): no horizontal lines by design */
    
    if (t1 == 0) {
        f32 x1 = p0.x + (p1.x - p0.x) * t1;
        if ((lasty < p0.y) && (p0.y > p1.y)) return(0);
        if ((lasty > p0.y) && (p0.y < p1.y)) return(0);
        int result = (goes_up ? 1 : -1);
        *vx = x1;
        return(result);
    }
    
    if (0 < t1 && t1 < 1.0f) {
        f32 x1 = p0.x + (p1.x - p0.x) * t1;
        *vx = x1;
        int result = (goes_up ? 1 : -1);
        return(result);
    }
    
    return(0);
}

static void
render_line(struct v2 from, struct v2 to, u32 *pixels, int width, int height, u32 color)
{
    if (from.x >= width) { 
        from.x = width - 1;
    }
    
    if (to.x >= width) {
        to.x = width - 1;
    }
    
    if (from.y >= height) {
        from.y = height - 1;
    }
    
    if (to.y >= height) {
        to.y = height - 1;
    }
    
    if (from.y == to.y) {
        if (to.x < from.x) {
            int tmp = to.x;
            to.x = from.x;
            from.x = tmp;
        }
        
        u32 *out = pixels + width * from.y + from.x;
        // printf("%ld %ld -> %ld %ld\n", from.x, from.y, to.x, to.y);
        
        for (int x = from.x; x <= to.x; ++x) {
            // printf("%d\n", x); 
            *out = color;
            ++out;
        }
    } else if (from.x == to.x) {
        if (to.y < from.y) {
            int tmp = to.y;
            to.y = from.y;
            from.y = tmp;
        }
        
        u32 *out = pixels + width * from.y + from.x;
        
        //printf("%ld %ld -> %ld %ld\n", from.x, from.y, to.x, to.y);
        
        for (int y = from.y; y <= to.y; ++y) {
            out = pixels + width * y + from.x;
            *out = color;
        }
    } else {
        int x0 = from.x;
        int y0 = from.y;
        int x1 = to.x;
        int y1 = to.y;
        
        int dx = abs(x1 - x0);
        int sx = (x0 < x1 ? 1 : -1);
        int dy = -abs(y1 - y0);
        int sy = (y0 < y1 ? 1 : -1);
        int err = dx+dy;
        
        while (0 <= x0 && x0 < width && 0 <= y0 && y0 < height) {
            u32 *out = pixels + width * y0 + x0;
            *out = color;
            
            if (x0 == x1 && y0 == y1) {
                break;
            }
            
            int e2 = 2 * err;
            
            if (e2 >= dy) {
                err += dy;
                x0 += sx;
            }
            
            if (e2 <= dx) {
                err += dx;
                y0 += sy;
            }
        }
    }
}

static void
render_grid(int x_step, int y_step, u32 *pixels, int width, int height)
{
    for (int y = 0; y < height; ++y) {
        for (int x = 0; x < width; ++x) {
            if (x % x_step == 0 || y % y_step == 0) {
                pixels[y * width + x] = 0xFFEEEEE;
            }
        }
    }
}

static int
intersect_glyph(struct line_contour *lines, f32 y, struct intersection *intersections)
{
    int nints = 0;
    
    for (int c = 0; c < lines->ncontours; ++c) {
        int from = lines->from[c];
        int to = lines->from[c + 1];
        
        for (int i = from; i < to; ++i) {
            int lasti = (i > from ? i - 1 : to - 1);
            f32 lasty = lines->data[lasti].a.y;
            f32 vx;
            int r = scanline_intersects_line(y, lines->data[i].a, lines->data[i].b, lasty, &vx);
            
            if (r) {
                intersections[nints].x = vx;
                intersections[nints].dir = r;
                ++nints;
            }
        }
    }
    
    return(nints);
}

static void
_D_render_glyph_points(struct glyph g, f32 scale, u32 *pixels, int width, int height)
{
    int points_from = 0;
    for (int c = 0; c < g.ncontours; ++c) {
        for (int p = points_from; p < g.end_pts_of_contours[c] + 1; ++p) {
            struct glyph_point gp = g.points[p];
            
            int x1 = round_f32((gp.x - g.xmin) * scale);
            int y1 = round_f32(gp.y * scale);
            
            int next = p + 1;
            if (p == g.end_pts_of_contours[c]) {
                next = points_from;
            }
            
            int x2 = round_f32((g.points[next].x - g.xmin) * scale);
            int y2 = round_f32(g.points[next].y * scale);
            
            render_line((struct v2) {x1, y1}, (struct v2) {x2, y2}, pixels, width, height, 0xFFFF00FF);
            //printf("%d %d\n", g.points[p].x, g.points[p].y);
            
            //XPutImage(display, window, default_gc, xwindow_buffer, 0, 0, 0, 0, width, height);
            
            if (gp.on_curve) {
                pixels[y1 * width + x1] = 0xFF0000FF;
            } else {
                pixels[y1 * width + x1] = 0xFFFFFF00;
            }
        }
        
        points_from = g.end_pts_of_contours[c] + 1;
    }
}

static void
sort_intersections(struct intersection *intersections, int size)
{
    bool swapped = true;
    
    while (swapped) {
        swapped = false;
        for (int i = 0; i < size - 1; ++i) {
            f32 x1 = intersections[i].x;
            f32 x2 = intersections[i + 1].x;
            if (x1 > x2) {
                struct intersection tmp = intersections[i];
                intersections[i] = intersections[i + 1];
                intersections[i + 1] = tmp;
                swapped = true;
            }
        }
    }
}

static void
render_glyph(struct glyph g, int px_size, struct line_contour *lines,
             f32 scale, u32 *pixels, int width, int at_x, int at_y)
{
    int oversample_y = 4;
    if (px_size <= 12) {
        oversample_y = 8;
    }
    
    f32 oversample_step = 1.0f / (oversample_y + 1);
    f32 oversample_norm = 1.0f / oversample_y;
    
    int gwidth = ceil_f32((g.xmax - g.xmin) * scale);
    int gheight = px_size;
    
    struct intersection *intersections = malloc(lines->from[lines->ncontours] * sizeof(struct intersection));
    
    f32 *accumulator = calloc(1, gwidth * gheight * sizeof(f32));
    
    for (int y = 0; y < gheight; ++y) {
        for (int yy = 1; yy <= oversample_y; ++yy) {
            u32 ncross = intersect_glyph(lines, y + oversample_step * yy, intersections);
            if (ncross) {
                sort_intersections(intersections, ncross);
                
                int state = 0;
                
                for (u32 i = 0; i < ncross - 1; ++i) {
                    struct intersection inter = intersections[i];
                    struct intersection next_inter = intersections[i + 1];
                    
                    state += inter.dir;
                    
                    if (state != 0) {
                        f32 x0 = inter.x;
                        f32 x1 = next_inter.x;
                        
                        int x_from = x0;
                        int x_to = (x1 > gwidth - 1 ? gwidth - 1 : x1);
                        
                        f32 start_brightness = (x_from + 1 - x0);
                        f32 end_brightness = (x1 - x_to);
                        
                        for (int x = x_from + 1; x < x_to; ++x) {
                            accumulator[y * gwidth + x] += oversample_norm;
                        }
                        
                        accumulator[y * gwidth + x_from] += start_brightness * oversample_norm;
                        accumulator[y * gwidth + x_to] += end_brightness * oversample_norm;
                    }
                }
            }
        }
    }
    
    for (int y = 0; y < gheight; ++y) {
        for (int x = 0; x < gwidth; ++x) {
            //printf(" %.2f", accumulator[y * gwidth + x]);
            u32 brightness = clamp_u32(accumulator[y * gwidth + x] * 255.99f, 255);
            u32 white = 0xFF000000 | brightness << 16 | brightness << 8 | brightness;
            
            if (brightness > 0) {
                pixels[(at_y + (gheight - 1 - y)) * width + (at_x + x)] = white;
            } else {
                //pixels[(at_y + (gheight - 1 - y)) * width + (at_x + x)] = 0xFFFF0000;
            }
        }
    }
    
    free(accumulator);
    
    //exit(1);
    
    free(intersections);
}

static void
outline_to_lines(struct glyph g, f32 scale, int max_descent, struct line_contour *dest, int *cnt) 
{
    int nlines = 0;
    int points_from = 0;
    int curve_segments = 5;
    
    for (int c = 0; c < g.ncontours; ++c) {
        for (int p = points_from; p < g.end_pts_of_contours[c] + 1; ++p) {
            struct glyph_point gp = g.points[p];
            
            int nexti = (p + 1 < g.end_pts_of_contours[c] + 1 ? p + 1 : points_from);
            struct glyph_point nextgp = g.points[nexti];
            
            if (p == points_from && !gp.on_curve) {
                continue;
            }
            
            f32 x1 = gp.x * scale;
            f32 y1 = (gp.y - g.baseline - max_descent) * scale;
            
            f32 x2 = nextgp.x * scale;
            f32 y2 = (nextgp.y - g.baseline - max_descent) * scale;
            
            if (nextgp.on_curve) {
                if (gp.y != nextgp.y) {
                    if (dest->data) {
                        dest->data[nlines].a = (struct v2f) { x1, y1 };
                        dest->data[nlines].b = (struct v2f) { x2, y2 };
                    }
                    ++nlines;
                }
            } else {
                int nextnexti = (nexti + 1 < g.end_pts_of_contours[c] + 1 ? nexti + 1 : points_from);
                struct glyph_point nextnextgp = g.points[nextnexti];
                
                f32 x3 = nextnextgp.x * scale;
                f32 y3 = (nextnextgp.y - g.baseline - max_descent) * scale;
                
                /* P(t) = P0*t^2 + P1*2*t*(1-t) + P2*(1-t)^2 */
                f32 t_step = 1.0f / curve_segments;
                f32 x_prev = x1;
                f32 y_prev = y1;
                
                /* s = 1 for exact beginning */
                for (int s = 1; s <= curve_segments; ++s) {
                    f32 t_now = t_step * s;
                    f32 x_now;
                    f32 y_now;
                    
                    if (s < curve_segments) {
                        x_now = x3 * t_now * t_now 
                            + x2 * 2.0f * t_now * (1.0f - t_now) 
                            + x1 * (1.0f - t_now) * (1.0f - t_now);
                        
                        y_now = y3 * t_now * t_now 
                            + y2 * 2.0f * t_now * (1.0f - t_now) 
                            + y1 * (1.0f - t_now) * (1.0f - t_now);
                    } else {
                        /* For exact match between neighbours */
                        x_now = x3;
                        y_now = y3;
                    }
                    
                    if (abs_f32(y_now - y_prev) > F32EPS) {
                        if (dest->data) {
                            dest->data[nlines].a = (struct v2f) { x_prev, y_prev };
                            dest->data[nlines].b = (struct v2f) { x_now, y_now };
                        }
                        ++nlines;
                    }
                    
                    x_prev = x_now;
                    y_prev = y_now;
                }
                
                ++p;
            }
        }
        
        dest->from[c + 1] = nlines;
        
        points_from = g.end_pts_of_contours[c] + 1;
    }
    
    dest->ncontours = g.ncontours;
    
    if (cnt) {
        *cnt = nlines;
    }
}

static void
render_utf_string(struct ttf_font font, int px_size, u32 *pixels, u32 width, wchar_t *string, int at_x, int at_y)
{
    u32 offset_x = at_x;
    u32 offset_y = at_y;
    f32 scale = (f32) px_size / ((f32) (font.hhea.ascent - font.hhea.descent));
    
    u32 len = wcslen(string);
    
    for (u32 i = 0; i < len; ++i) {
        u16 codepoint = string[i];
        if (codepoint != ' ') {
            struct glyph g = get_outline(&font, codepoint);
            
            //exit(0);
            
            struct line_contour lines = { 0 };
            int nlines = 0;
            outline_to_lines(g, scale, font.hhea.descent, &lines, &nlines);
            lines.data = malloc(nlines * sizeof(struct line));
            outline_to_lines(g, scale, font.hhea.descent, &lines, 0);
            
            render_glyph(g, px_size, &lines, scale, pixels, width, offset_x + g.lsb * scale, offset_y);
            
            offset_x += ceil_f32(scale * g.advance);
            
            free(lines.data);
        } else {
            offset_x += px_size / 3;
        }
        
        //XPutImage(display, window, default_gc, xwindow_buffer, 0, 0, 0, 0, width, height);
        //sleep(1);
    }
}

int
main(int argc, char **argv)
{
    if (argc != 2) {
        fprintf(stderr, "Usage: %s filename.ttf\n", argv[0]);
        return(1);
    }
    
    display = XOpenDisplay(0);
    Window root_window = DefaultRootWindow(display);
    
    int default_screen = DefaultScreen(display);
    int screen_bit_depth = 24;
    int width = 1280;
    int height = 720;
    
    XVisualInfo visinfo = { 0 };
    XMatchVisualInfo(display, default_screen, screen_bit_depth, TrueColor, &visinfo);
    
    XSetWindowAttributes window_attr;
    window_attr.bit_gravity = StaticGravity;
    window_attr.background_pixel = 0;
    window_attr.colormap = XCreateColormap(display, root_window, visinfo.visual, AllocNone);
    window_attr.event_mask = StructureNotifyMask | KeyPressMask | KeyReleaseMask;
    
    unsigned long attribute_mask = CWBitGravity | CWBackPixel | CWColormap | CWEventMask;
    
    window = XCreateWindow(display, root_window,
                           0, 0,
                           width, height, 0,
                           visinfo.depth, InputOutput,
                           visinfo.visual, attribute_mask, &window_attr);
    
    XGrabPointer(display, window, False, ButtonPressMask | PointerMotionMask,
                 GrabModeAsync, GrabModeAsync, None, None, CurrentTime);
    
    XSelectInput(display, window, KeyPress | KeyRelease | ButtonRelease | PointerMotionMask);
    
    XStoreName(display, window, "ttf");
    XMapWindow(display, window);
    XFlush(display);
    
    Atom WM_DELETE_WINDOW = XInternAtom(display, "WM_DELETE_WINDOW", False);
    XSetWMProtocols(display, window, &WM_DELETE_WINDOW, 1);
    
    int pixel_bits = 32;
    int pixel_bytes = pixel_bits / 8;
    int window_buffer_size = width * height * pixel_bytes;
    void *hc_vram = malloc(window_buffer_size);
    
    xwindow_buffer = XCreateImage(display, visinfo.visual, visinfo.depth, ZPixmap, 0, hc_vram, width, height,
                                  pixel_bits, 0);
    
    default_gc = DefaultGC(display, default_screen);
    
    struct ttf_font font = parse_ttf_file(argv[1], "Inter");
    printf("Loaded font\n");
    
    u32 *pixels = hc_vram;
    int t = 0;
    
    for (;;) {
        //memset(pixels, 0xFF, width * height * 4);
        //memset(pixels, 0xFF, width * height * 4);
        memset(pixels, 0x00, width * height * 4);
        
        int at = 75;
        //render_utf_string(font, 18, pixels, width, L"The quick brown fox jumps over the lazy dog", 100, 150);
        //render_utf_string(font, 12, pixels, width, L"This text is seriously small", 100, at); at += 12;
        //render_utf_string(font, 13, pixels, width, L"This text is seriously small", 100, at); at += 13;
        render_utf_string(font, 14, pixels, width, L"This text is seriously small", 100, at); at += 14;
        //render_utf_string(font, 15, pixels, width, L"This text is seriously small", 100, at); at += 14;
        render_utf_string(font, 26, pixels, width, L"memset(pixels, 0x00, width * height * 4);", 10, at); at += 24;
        render_utf_string(font, 24, pixels, width, L"The quick brown fox jumps over the lazy dog", 10, at); at += 24;
        
        wchar_t *string1 = L"iiiiiiiiiiiiiiiiiii";
        wchar_t *string2 = L"something-some";
        wchar_t *string3 = L"MORE MORE";
        
        int w1 = get_string_width(&font, 32, string1, wcslen(string1));
        int w2 = get_string_width(&font, 32, string2, wcslen(string2));
        
        render_utf_string(font, 32, pixels, width, string1, 10, at); 
        render_utf_string(font, 32, pixels, width, string2, 10 + w1, at);
        render_utf_string(font, 32, pixels, width, string3, 10 + w1 + w2, at);
        at += 80;
        render_utf_string(font, 80, pixels, width, L"~!@#$%^&*()_+;:,./\\||||", 10, at);
        
        XPutImage(display, window, default_gc, xwindow_buffer, 0, 0, 0, 0, width, height);
        
        sleep(1);
        
        ++t;
    }
    
    return(0);
}