// stb_truetype.h - v1.26 - public domain
// authored from 2009-2021 by Sean Barrett / RAD Game Tools
//
// =======================================================================
//
//    NO SECURITY GUARANTEE -- DO NOT USE THIS ON UNTRUSTED FONT FILES
//
// This library does no range checking of the offsets found in the file,
// meaning an attacker can use it to read arbitrary memory.
//
// =======================================================================
//
//   This library processes TrueType files:
//        parse files
//        extract glyph metrics
//        extract glyph shapes
//        render glyphs to one-channel bitmaps with antialiasing (box filter)
//        render glyphs to one-channel SDF bitmaps (signed-distance field/function)
//
//   Todo:
//        non-MS cmaps
//        crashproof on bad data
//        hinting? (no longer patented)
//        cleartype-style AA?
//        optimize: use simple memory allocator for intermediates
//        optimize: build edge-list directly from curves
//        optimize: rasterize directly from curves?
//
// ADDITIONAL CONTRIBUTORS
//
//   Mikko Mononen: compound shape support, more cmap formats
//   Tor Andersson: kerning, subpixel rendering
//   Dougall Johnson: OpenType / Type 2 font handling
//   Daniel Ribeiro Maciel: basic GPOS-based kerning
//
//   Misc other:
//       Ryan Gordon
//       Simon Glass
//       github:IntellectualKitty
//       Imanol Celaya
//       Daniel Ribeiro Maciel
//
//   Bug/warning reports/fixes:
//       "Zer" on mollyrocket       Fabian "ryg" Giesen   github:NiLuJe
//       Cass Everitt               Martins Mozeiko       github:aloucks
//       stoiko (Haemimont Games)   Cap Petschulat        github:oyvindjam
//       Brian Hook                 Omar Cornut           github:vassvik
//       Walter van Niftrik         Ryan Griege
//       David Gow                  Peter LaValle
//       David Given                Sergey Popov
//       Ivan-Assen Ivanov          Giumo X. Clanjor
//       Anthony Pesch              Higor Euripedes
//       Johan Duparc               Thomas Fields
//       Hou Qiming                 Derek Vinyard
//       Rob Loach                  Cort Stratton
//       Kenney Phillis Jr.         Brian Costabile
//       Ken Voskuil (kaesve)
//
// VERSION HISTORY
//
//   1.26 (2021-08-28) fix broken rasterizer
//   1.25 (2021-07-11) many fixes
//   1.24 (2020-02-05) fix warning
//   1.23 (2020-02-02) query SVG data for glyphs; query whole kerning table (but only kern not GPOS)
//   1.22 (2019-08-11) minimize missing-glyph duplication; fix kerning if both 'GPOS' and 'kern' are defined
//   1.21 (2019-02-25) fix warning
//   1.20 (2019-02-07) PackFontRange skips missing codepoints; GetScaleFontVMetrics()
//   1.19 (2018-02-11) GPOS kerning, STBTT_fmod
//   1.18 (2018-01-29) add missing function
//   1.17 (2017-07-23) make more arguments const; doc fix
//   1.16 (2017-07-12) SDF support
//   1.15 (2017-03-03) make more arguments const
//   1.14 (2017-01-16) num-fonts-in-TTC function
//   1.13 (2017-01-02) support OpenType fonts, certain Apple fonts
//   1.12 (2016-10-25) suppress warnings about casting away const with -Wcast-qual
//   1.11 (2016-04-02) fix unused-variable warning
//   1.10 (2016-04-02) user-defined fabs(); rare memory leak; remove duplicate typedef
//   1.09 (2016-01-16) warning fix; avoid crash on outofmem; use allocation userdata properly
//   1.08 (2015-09-13) document stbtt_Rasterize(); fixes for vertical & horizontal edges
//   1.07 (2015-08-01) allow PackFontRanges to accept arrays of sparse codepoints;
//                     variant PackFontRanges to pack and render in separate phases;
//                     fix stbtt_GetFontOFfsetForIndex (never worked for non-0 input?);
//                     fixed an assert() bug in the new rasterizer
//                     replace assert() with STBTT_assert() in new rasterizer
//
//   Full history can be found at the end of this file.
//
// LICENSE
//
//   See end of file for license information.
//
// USAGE
//
//   Include this file in whatever places need to refer to it. In ONE C/C++
//   file, write:
//      #define STB_TRUETYPE_IMPLEMENTATION
//   before the #include of this file. This expands out the actual
//   implementation into that C/C++ file.
//
//   To make the implementation private to the file that generates the implementation,
//      #define STBTT_STATIC
//
//   Simple 3D API (don't ship this, but it's fine for tools and quick start)
//           stbtt_BakeFontBitmap()               -- bake a font to a bitmap for use as texture
//           stbtt_GetBakedQuad()                 -- compute quad to draw for a given char
//
//   Improved 3D API (more shippable):
//           #include "stb_rect_pack.h"           -- optional, but you really want it
//           stbtt_PackBegin()
//           stbtt_PackSetOversampling()          -- for improved quality on small fonts
//           stbtt_PackFontRanges()               -- pack and renders
//           stbtt_PackEnd()
//           stbtt_GetPackedQuad()
//
//   "Load" a font file from a memory buffer (you have to keep the buffer loaded)
//           stbtt_InitFont()
//           stbtt_GetFontOffsetForIndex()        -- indexing for TTC font collections
//           stbtt_GetNumberOfFonts()             -- number of fonts for TTC font collections
//
//   Render a unicode codepoint to a bitmap
//           stbtt_GetCodepointBitmap()           -- allocates and returns a bitmap
//           stbtt_MakeCodepointBitmap()          -- renders into bitmap you provide
//           stbtt_GetCodepointBitmapBox()        -- how big the bitmap must be
//
//   Character advance/positioning
//           stbtt_GetCodepointHMetrics()
//           stbtt_GetFontVMetrics()
//           stbtt_GetFontVMetricsOS2()
//           stbtt_GetCodepointKernAdvance()
//
//   Starting with version 1.06, the rasterizer was replaced with a new,
//   faster and generally-more-precise rasterizer. The new rasterizer more
//   accurately measures pixel coverage for anti-aliasing, except in the case
//   where multiple shapes overlap, in which case it overestimates the AA pixel
//   coverage. Thus, anti-aliasing of intersecting shapes may look wrong. If
//   this turns out to be a problem, you can re-enable the old rasterizer with
//        #define STBTT_RASTERIZER_VERSION 1
//   which will incur about a 15% speed hit.
//
// ADDITIONAL DOCUMENTATION
//
//   Immediately after this block comment are a series of sample programs.
//
//   After the sample programs is the "header file" section. This section
//   includes documentation for each API function.
//
//   Some important concepts to understand to use this library:
//
//      Codepoint
//         Characters are defined by unicode codepoints, e.g. 65 is
//         uppercase A, 231 is lowercase c with a cedilla, 0x7e30 is
//         the hiragana for "ma".
//
//      Glyph
//         A visual character shape (every codepoint is rendered as
//         some glyph)
//
//      Glyph index
//         A font-specific integer ID representing a glyph
//
//      Baseline
//         Glyph shapes are defined relative to a baseline, which is the
//         bottom of uppercase characters. Characters extend both above
//         and below the baseline.
//
//      Current Point
//         As you draw text to the screen, you keep track of a "current point"
//         which is the origin of each character. The current point's vertical
//         position is the baseline. Even "baked fonts" use this model.
//
//      Vertical Font Metrics
//         The vertical qualities of the font, used to vertically position
//         and space the characters. See docs for stbtt_GetFontVMetrics.
//
//      Font Size in Pixels or Points
//         The preferred interface for specifying font sizes in stb_truetype
//         is to specify how tall the font's vertical extent should be in pixels.
//         If that sounds good enough, skip the next paragraph.
//
//         Most font APIs instead use "points", which are a common typographic
//         measurement for describing font size, defined as 72 points per inch.
//         stb_truetype provides a point API for compatibility. However, true
//         "per inch" conventions don't make much sense on computer displays
//         since different monitors have different number of pixels per
//         inch. For example, Windows traditionally uses a convention that
//         there are 96 pixels per inch, thus making 'inch' measurements have
//         nothing to do with inches, and thus effectively defining a point to
//         be 1.333 pixels. Additionally, the TrueType font data provides
//         an explicit scale factor to scale a given font's glyphs to points,
//         but the author has observed that this scale factor is often wrong
//         for non-commercial fonts, thus making fonts scaled in points
//         according to the TrueType spec incoherently sized in practice.
//
// DETAILED USAGE:
//
//  Scale:
//    Select how high you want the font to be, in points or pixels.
//    Call ScaleForPixelHeight or ScaleForMappingEmToPixels to compute
//    a scale factor SF that will be used by all other functions.
//
//  Baseline:
//    You need to select a y-coordinate that is the baseline of where
//    your text will appear. Call GetFontBoundingBox to get the baseline-relative
//    bounding box for all characters. SF*-y0 will be the distance in pixels
//    that the worst-case character could extend above the baseline, so if
//    you want the top edge of characters to appear at the top of the
//    screen where y=0, then you would set the baseline to SF*-y0.
//
//  Current point:
//    Set the current point where the first character will appear. The
//    first character could extend left of the current point; this is font
//    dependent. You can either choose a current point that is the leftmost
//    point and hope, or add some padding, or check the bounding box or
//    left-side-bearing of the first character to be displayed and set
//    the current point based on that.
//
//  Displaying a character:
//    Compute the bounding box of the character. It will contain signed values
//    relative to <current_point, baseline>. I.e. if it returns x0,y0,x1,y1,
//    then the character should be displayed in the rectangle from
//    <current_point+SF*x0, baseline+SF*y0> to <current_point+SF*x1,baseline+SF*y1).
//
//  Advancing for the next character:
//    Call GlyphHMetrics, and compute 'current_point += SF * advance'.
//
//
// ADVANCED USAGE
//
//   Quality:
//
//    - Use the functions with Subpixel at the end to allow your characters
//      to have subpixel positioning. Since the font is anti-aliased, not
//      hinted, this is very import for quality. (This is not possible with
//      baked fonts.)
//
//    - Kerning is now supported, and if you're supporting subpixel rendering
//      then kerning is worth using to give your text a polished look.
//
//   Performance:
//
//    - Convert Unicode codepoints to glyph indexes and operate on the glyphs;
//      if you don't do this, stb_truetype is forced to do the conversion on
//      every call.
//
//    - There are a lot of memory allocations. We should modify it to take
//      a temp buffer and allocate from the temp buffer (without freeing),
//      should help performance a lot.
//
// NOTES
//
//   The system uses the raw data found in the .ttf file without changing it
//   and without building auxiliary data structures. This is a bit inefficient
//   on little-endian systems (the data is big-endian), but assuming you're
//   caching the bitmaps or glyph shapes this shouldn't be a big deal.
//
//   It appears to be very hard to programmatically determine what font a
//   given file is in a general way. I provide an API for this, but I don't
//   recommend it.
//
//
// PERFORMANCE MEASUREMENTS FOR 1.06:
//
//                      32-bit     64-bit
//   Previous release:  8.83 s     7.68 s
//   Pool allocations:  7.72 s     6.34 s
//   Inline sort     :  6.54 s     5.65 s
//   New rasterizer  :  5.63 s     5.00 s

//////////////////////////////////////////////////////////////////////////////
//////////////////////////////////////////////////////////////////////////////
////
////  SAMPLE PROGRAMS
////
//
//  Incomplete text-in-3d-api example, which draws quads properly aligned to be lossless.
//  See "tests/truetype_demo_win32.c" for a complete version.
#if 0
#define STB_TRUETYPE_IMPLEMENTATION  // force following include to generate implementation
#include "stb_truetype.h"

unsigned char ttf_buffer[1<<20];
unsigned char temp_bitmap[512*512];

stbtt_bakedchar cdata[96]; // ASCII 32..126 is 95 glyphs
GLuint ftex;

void my_stbtt_initfont(void)
{
   fread(ttf_buffer, 1, 1<<20, fopen("c:/windows/fonts/times.ttf", "rb"));
   stbtt_BakeFontBitmap(ttf_buffer,0, 32.0, temp_bitmap,512,512, 32,96, cdata); // no guarantee this fits!
   // can free ttf_buffer at this point
   glGenTextures(1, &ftex);
   glBindTexture(GL_TEXTURE_2D, ftex);
   glTexImage2D(GL_TEXTURE_2D, 0, GL_ALPHA, 512,512, 0, GL_ALPHA, GL_UNSIGNED_BYTE, temp_bitmap);
   // can free temp_bitmap at this point
   glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
}

void my_stbtt_print(float x, float y, char *text)
{
   // assume orthographic projection with units = screen pixels, origin at top left
   glEnable(GL_BLEND);
   glBlendFunc(GL_SRC_ALPHA, GL_ONE_MINUS_SRC_ALPHA);
   glEnable(GL_TEXTURE_2D);
   glBindTexture(GL_TEXTURE_2D, ftex);
   glBegin(GL_QUADS);
   while (*text) {
      if (*text >= 32 && *text < 128) {
         stbtt_aligned_quad q;
         stbtt_GetBakedQuad(cdata, 512,512, *text-32, &x,&y,&q,1);//1=opengl & d3d10+,0=d3d9
         glTexCoord2f(q.s0,q.t0); glVertex2f(q.x0,q.y0);
         glTexCoord2f(q.s1,q.t0); glVertex2f(q.x1,q.y0);
         glTexCoord2f(q.s1,q.t1); glVertex2f(q.x1,q.y1);
         glTexCoord2f(q.s0,q.t1); glVertex2f(q.x0,q.y1);
      }
      ++text;
   }
   glEnd();
}
#endif
//
//
//////////////////////////////////////////////////////////////////////////////
//
// Complete program (this compiles): get a single bitmap, print as ASCII art
//
#if 0
#include <stdio.h>
#define STB_TRUETYPE_IMPLEMENTATION  // force following include to generate implementation
#include "stb_truetype.h"

char ttf_buffer[1<<25];

int main(int argc, char **argv)
{
   stbtt_fontinfo font;
   unsigned char *bitmap;
   int w,h,i,j,c = (argc > 1 ? atoi(argv[1]) : 'a'), s = (argc > 2 ? atoi(argv[2]) : 20);

   fread(ttf_buffer, 1, 1<<25, fopen(argc > 3 ? argv[3] : "c:/windows/fonts/arialbd.ttf", "rb"));

   stbtt_InitFont(&font, ttf_buffer, stbtt_GetFontOffsetForIndex(ttf_buffer,0));
   bitmap = stbtt_GetCodepointBitmap(&font, 0,stbtt_ScaleForPixelHeight(&font, s), c, &w, &h, 0,0);

   for (j=0; j < h; ++j) {
      for (i=0; i < w; ++i)
         putchar(" .:ioVM@"[bitmap[j*w+i]>>5]);
      putchar('\n');
   }
   return 0;
}
#endif
//
// Output:
//
//     .ii.
//    @@@@@@.
//   V@Mio@@o
//   :i.  V@V
//     :oM@@M
//   :@@@MM@M
//   @@o  o@M
//  :@@.  M@M
//   @@@o@@@@
//   :M@@V:@@.
//
//////////////////////////////////////////////////////////////////////////////
//
// Complete program: print "Hello World!" banner, with bugs
//
#if 0
char buffer[24<<20];
unsigned char screen[20][79];

int main(int arg, char **argv)
{
   stbtt_fontinfo font;
   int i,j,ascent,baseline,ch=0;
   float scale, xpos=2; // leave a little padding in case the character extends left
   char *text = "Heljo World!"; // intentionally misspelled to show 'lj' brokenness

   fread(buffer, 1, 1000000, fopen("c:/windows/fonts/arialbd.ttf", "rb"));
   stbtt_InitFont(&font, buffer, 0);

   scale = stbtt_ScaleForPixelHeight(&font, 15);
   stbtt_GetFontVMetrics(&font, &ascent,0,0);
   baseline = (int) (ascent*scale);

   while (text[ch]) {
      int advance,lsb,x0,y0,x1,y1;
      float x_shift = xpos - (float) floor(xpos);
      stbtt_GetCodepointHMetrics(&font, text[ch], &advance, &lsb);
      stbtt_GetCodepointBitmapBoxSubpixel(&font, text[ch], scale,scale,x_shift,0, &x0,&y0,&x1,&y1);
      stbtt_MakeCodepointBitmapSubpixel(&font, &screen[baseline + y0][(int) xpos + x0], x1-x0,y1-y0, 79, scale,scale,x_shift,0, text[ch]);
      // note that this stomps the old data, so where character boxes overlap (e.g. 'lj') it's wrong
      // because this API is really for baking character bitmaps into textures. if you want to render
      // a sequence of characters, you really need to render each bitmap to a temp buffer, then
      // "alpha blend" that into the working buffer
      xpos += (advance * scale);
      if (text[ch+1])
         xpos += scale*stbtt_GetCodepointKernAdvance(&font, text[ch],text[ch+1]);
      ++ch;
   }

   for (j=0; j < 20; ++j) {
      for (i=0; i < 78; ++i)
         putchar(" .:ioVM@"[screen[j][i]>>5]);
      putchar('\n');
   }

   return 0;
}
#endif


//////////////////////////////////////////////////////////////////////////////
//////////////////////////////////////////////////////////////////////////////
////
////   INTEGRATION WITH YOUR CODEBASE
////
////   The following sections allow you to supply alternate definitions
////   of C library functions used by stb_truetype, e.g. if you don't
////   link with the C runtime library.

#ifdef STB_TRUETYPE_IMPLEMENTATION
   // #define your own (u)stbtt_int8/16/32 before including to override this
   #ifndef stbtt_uint8
   typedef unsigned char   stbtt_uint8;
   typedef signed   char   stbtt_int8;
   typedef unsigned short  stbtt_uint16;
   typedef signed   short  stbtt_int16;
   typedef unsigned int    stbtt_uint32;
   typedef signed   int    stbtt_int32;
   #endif

   typedef char stbtt__check_size32[sizeof(stbtt_int32)==4 ? 1 : -1];
   typedef char stbtt__check_size16[sizeof(stbtt_int16)==2 ? 1 : -1];

   // e.g. #define your own STBTT_ifloor/STBTT_iceil() to avoid math.h
   #ifndef STBTT_ifloor
   #include <math.h>
   #define STBTT_ifloor(x)   ((int) floor(x))
   #define STBTT_iceil(x)    ((int) ceil(x))
   #endif

   #ifndef STBTT_sqrt
   #include <math.h>
   #define STBTT_sqrt(x)      sqrt(x)
   #define STBTT_pow(x,y)     pow(x,y)
   #endif

   #ifndef STBTT_fmod
   #include <math.h>
   #define STBTT_fmod(x,y)    fmod(x,y)
   #endif

   #ifndef STBTT_cos
   #include <math.h>
   #define STBTT_cos(x)       cos(x)
   #define STBTT_acos(x)      acos(x)
   #endif

   #ifndef STBTT_fabs
   #include <math.h>
   #define STBTT_fabs(x)      fabs(x)
   #endif

   // #define your own functions "STBTT_malloc" / "STBTT_free" to avoid malloc.h
   #ifndef STBTT_malloc
   #include <stdlib.h>
   #define STBTT_malloc(x,u)  ((void)(u),FONTSTASH_MALLOC(x))
   #define STBTT_free(x,u)    ((void)(u),FONTSTASH_FREE(x))
   #endif

   #ifndef STBTT_assert
   #include <assert.h>
   #define STBTT_assert(x)    assert(x)
   #endif

   #ifndef STBTT_strlen
   #include <string.h>
   #define STBTT_strlen(x)    strlen(x)
   #endif

   #ifndef STBTT_memcpy
   #include <string.h>
   #define STBTT_memcpy       memcpy
   #define STBTT_memset       memset
   #endif
#endif

///////////////////////////////////////////////////////////////////////////////
///////////////////////////////////////////////////////////////////////////////
////
////   INTERFACE
////
////

#ifndef __STB_INCLUDE_STB_TRUETYPE_H__
#define __STB_INCLUDE_STB_TRUETYPE_H__

#ifdef STBTT_STATIC
#define STBTT_DEF static
#else
#define STBTT_DEF extern
#endif

#ifdef __cplusplus
extern "C" {
#endif

// private structure
typedef struct
{
   unsigned char *data;
   int cursor;
   int size;
} stbtt__buf;

//////////////////////////////////////////////////////////////////////////////
//
// TEXTURE BAKING API
//
// If you use this API, you only have to call two functions ever.
//

typedef struct
{
   unsigned short x0,y0,x1,y1; // coordinates of bbox in bitmap
   float xoff,yoff,xadvance;
} stbtt_bakedchar;

STBTT_DEF int stbtt_BakeFontBitmap(const unsigned char *data, int offset,  // font location (use offset=0 for plain .ttf)
                                float pixel_height,                     // height of font in pixels
                                unsigned char *pixels, int pw, int ph,  // bitmap to be filled in
                                int first_char, int num_chars,          // characters to bake
                                stbtt_bakedchar *chardata);             // you allocate this, it's num_chars long
// if return is positive, the first unused row of the bitmap
// if return is negative, returns the negative of the number of characters that fit
// if return is 0, no characters fit and no rows were used
// This uses a very crappy packing.

typedef struct
{
   float x0,y0,s0,t0; // top-left
   float x1,y1,s1,t1; // bottom-right
} stbtt_aligned_quad;

STBTT_DEF void stbtt_GetBakedQuad(const stbtt_bakedchar *chardata, int pw, int ph,  // same data as above
                               int char_index,             // character to display
                               float *xpos, float *ypos,   // pointers to current position in screen pixel space
                               stbtt_aligned_quad *q,      // output: quad to draw
                               int opengl_fillrule);       // true if opengl fill rule; false if DX9 or earlier
// Call GetBakedQuad with char_index = 'character - first_char', and it
// creates the quad you need to draw and advances the current position.
//
// The coordinate system used assumes y increases downwards.
//
// Characters will extend both above and below the current position;
// see discussion of "BASELINE" above.
//
// It's inefficient; you might want to c&p it and optimize it.

STBTT_DEF void stbtt_GetScaledFontVMetrics(const unsigned char *fontdata, int index, float size, float *ascent, float *descent, float *lineGap);
// Query the font vertical metrics without having to create a font first.


//////////////////////////////////////////////////////////////////////////////
//
// NEW TEXTURE BAKING API
//
// This provides options for packing multiple fonts into one atlas, not
// perfectly but better than nothing.

typedef struct
{
   unsigned short x0,y0,x1,y1; // coordinates of bbox in bitmap
   float xoff,yoff,xadvance;
   float xoff2,yoff2;
} stbtt_packedchar;

typedef struct stbtt_pack_context stbtt_pack_context;
typedef struct stbtt_fontinfo stbtt_fontinfo;
#ifndef STB_RECT_PACK_VERSION
typedef struct stbrp_rect stbrp_rect;
#endif

STBTT_DEF int  stbtt_PackBegin(stbtt_pack_context *spc, unsigned char *pixels, int width, int height, int stride_in_bytes, int padding, void *alloc_context);
// Initializes a packing context stored in the passed-in stbtt_pack_context.
// Future calls using this context will pack characters into the bitmap passed
// in here: a 1-channel bitmap that is width * height. stride_in_bytes is
// the distance from one row to the next (or 0 to mean they are packed tightly
// together). "padding" is the amount of padding to leave between each
// character (normally you want '1' for bitmaps you'll use as textures with
// bilinear filtering).
//
// Returns 0 on failure, 1 on success.

STBTT_DEF void stbtt_PackEnd  (stbtt_pack_context *spc);
// Cleans up the packing context and frees all memory.

#define STBTT_POINT_SIZE(x)   (-(x))

STBTT_DEF int  stbtt_PackFontRange(stbtt_pack_context *spc, const unsigned char *fontdata, int font_index, float font_size,
                                int first_unicode_char_in_range, int num_chars_in_range, stbtt_packedchar *chardata_for_range);
// Creates character bitmaps from the font_index'th font found in fontdata (use
// font_index=0 if you don't know what that is). It creates num_chars_in_range
// bitmaps for characters with unicode values starting at first_unicode_char_in_range
// and increasing. Data for how to render them is stored in chardata_for_range;
// pass these to stbtt_GetPackedQuad to get back renderable quads.
//
// font_size is the full height of the character from ascender to descender,
// as computed by stbtt_ScaleForPixelHeight. To use a point size as computed
// by stbtt_ScaleForMappingEmToPixels, wrap the point size in STBTT_POINT_SIZE()
// and pass that result as 'font_size':
//       ...,                  20 , ... // font max minus min y is 20 pixels tall
//       ..., STBTT_POINT_SIZE(20), ... // 'M' is 20 pixels tall

typedef struct
{
   float font_size;
   int first_unicode_codepoint_in_range;  // if non-zero, then the chars are continuous, and this is the first codepoint
   int *array_of_unicode_codepoints;       // if non-zero, then this is an array of unicode codepoints
   int num_chars;
   stbtt_packedchar *chardata_for_range; // output
   unsigned char h_oversample, v_oversample; // don't set these, they're used internally
} stbtt_pack_range;

STBTT_DEF int  stbtt_PackFontRanges(stbtt_pack_context *spc, const unsigned char *fontdata, int font_index, stbtt_pack_range *ranges, int num_ranges);
// Creates character bitmaps from multiple ranges of characters stored in
// ranges. This will usually create a better-packed bitmap than multiple
// calls to stbtt_PackFontRange. Note that you can call this multiple
// times within a single PackBegin/PackEnd.

STBTT_DEF void stbtt_PackSetOversampling(stbtt_pack_context *spc, unsigned int h_oversample, unsigned int v_oversample);
// Oversampling a font increases the quality by allowing higher-quality subpixel
// positioning, and is especially valuable at smaller text sizes.
//
// This function sets the amount of oversampling for all following calls to
// stbtt_PackFontRange(s) or stbtt_PackFontRangesGatherRects for a given
// pack context. The default (no oversampling) is achieved by h_oversample=1
// and v_oversample=1. The total number of pixels required is
// h_oversample*v_oversample larger than the default; for example, 2x2
// oversampling requires 4x the storage of 1x1. For best results, render
// oversampled textures with bilinear filtering. Look at the readme in
// stb/tests/oversample for information about oversampled fonts
//
// To use with PackFontRangesGather etc., you must set it before calls
// call to PackFontRangesGatherRects.

STBTT_DEF void stbtt_PackSetSkipMissingCodepoints(stbtt_pack_context *spc, int skip);
// If skip != 0, this tells stb_truetype to skip any codepoints for which
// there is no corresponding glyph. If skip=0, which is the default, then
// codepoints without a glyph received the font's "missing character" glyph,
// typically an empty box by convention.

STBTT_DEF void stbtt_GetPackedQuad(const stbtt_packedchar *chardata, int pw, int ph,  // same data as above
                               int char_index,             // character to display
                               float *xpos, float *ypos,   // pointers to current position in screen pixel space
                               stbtt_aligned_quad *q,      // output: quad to draw
                               int align_to_integer);

STBTT_DEF int  stbtt_PackFontRangesGatherRects(stbtt_pack_context *spc, const stbtt_fontinfo *info, stbtt_pack_range *ranges, int num_ranges, stbrp_rect *rects);
STBTT_DEF void stbtt_PackFontRangesPackRects(stbtt_pack_context *spc, stbrp_rect *rects, int num_rects);
STBTT_DEF int  stbtt_PackFontRangesRenderIntoRects(stbtt_pack_context *spc, const stbtt_fontinfo *info, stbtt_pack_range *ranges, int num_ranges, stbrp_rect *rects);
// Calling these functions in sequence is roughly equivalent to calling
// stbtt_PackFontRanges(). If you more control over the packing of multiple
// fonts, or if you want to pack custom data into a font texture, take a look
// at the source to of stbtt_PackFontRanges() and create a custom version
// using these functions, e.g. call GatherRects multiple times,
// building up a single array of rects, then call PackRects once,
// then call RenderIntoRects repeatedly. This may result in a
// better packing than calling PackFontRanges multiple times
// (or it may not).

// this is an opaque structure that you shouldn't mess with which holds
// all the context needed from PackBegin to PackEnd.
struct stbtt_pack_context {
   void *user_allocator_context;
   void *pack_info;
   int   width;
   int   height;
   int   stride_in_bytes;
   int   padding;
   int   skip_missing;
   unsigned int   h_oversample, v_oversample;
   unsigned char *pixels;
   void  *nodes;
};

//////////////////////////////////////////////////////////////////////////////
//
// FONT LOADING
//
//

STBTT_DEF int stbtt_GetNumberOfFonts(const unsigned char *data);
// This function will determine the number of fonts in a font file.  TrueType
// collection (.ttc) files may contain multiple fonts, while TrueType font
// (.ttf) files only contain one font. The number of fonts can be used for
// indexing with the previous function where the index is between zero and one
// less than the total fonts. If an error occurs, -1 is returned.

STBTT_DEF int stbtt_GetFontOffsetForIndex(const unsigned char *data, int index);
// Each .ttf/.ttc file may have more than one font. Each font has a sequential
// index number starting from 0. Call this function to get the font offset for
// a given index; it returns -1 if the index is out of range. A regular .ttf
// file will only define one font and it always be at offset 0, so it will
// return '0' for index 0, and -1 for all other indices.

// The following structure is defined publicly so you can declare one on
// the stack or as a global or etc, but you should treat it as opaque.
struct stbtt_fontinfo
{
   void           * userdata;
   unsigned char  * data;              // pointer to .ttf file
   int              fontstart;         // offset of start of font

   int numGlyphs;                     // number of glyphs, needed for range checking

   int loca,head,glyf,hhea,hmtx,kern,gpos,svg; // table locations as offset from start of .ttf
   int index_map;                     // a cmap mapping for our chosen character encoding
   int indexToLocFormat;              // format needed to map from glyph index to glyph

   stbtt__buf cff;                    // cff font data
   stbtt__buf charstrings;            // the charstring index
   stbtt__buf gsubrs;                 // global charstring subroutines index
   stbtt__buf subrs;                  // private charstring subroutines index
   stbtt__buf fontdicts;              // array of font dicts
   stbtt__buf fdselect;               // map from glyph to fontdict
};

STBTT_DEF int stbtt_InitFont(stbtt_fontinfo *info, const unsigned char *data, int offset);
// Given an offset into the file that defines a font, this function builds
// the necessary cached info for the rest of the system. You must allocate
// the stbtt_fontinfo yourself, and stbtt_InitFont will fill it out. You don't
// need to do anything special to free it, because the contents are pure
// value data with no additional data structures. Returns 0 on failure.


//////////////////////////////////////////////////////////////////////////////
//
// CHARACTER TO GLYPH-INDEX CONVERSIOn

STBTT_DEF int stbtt_FindGlyphIndex(const stbtt_fontinfo *info, int unicode_codepoint);
// If you're going to perform multiple operations on the same character
// and you want a speed-up, call this function with the character you're
// going to process, then use glyph-based functions instead of the
// codepoint-based functions.
// Returns 0 if the character codepoint is not defined in the font.


//////////////////////////////////////////////////////////////////////////////
//
// CHARACTER PROPERTIES
//

STBTT_DEF float stbtt_ScaleForPixelHeight(const stbtt_fontinfo *info, float pixels);
// computes a scale factor to produce a font whose "height" is 'pixels' tall.
// Height is measured as the distance from the highest ascender to the lowest
// descender; in other words, it's equivalent to calling stbtt_GetFontVMetrics
// and computing:
//       scale = pixels / (ascent - descent)
// so if you prefer to measure height by the ascent only, use a similar calculation.

STBTT_DEF float stbtt_ScaleForMappingEmToPixels(const stbtt_fontinfo *info, float pixels);
// computes a scale factor to produce a font whose EM size is mapped to
// 'pixels' tall. This is probably what traditional APIs compute, but
// I'm not positive.

STBTT_DEF void stbtt_GetFontVMetrics(const stbtt_fontinfo *info, int *ascent, int *descent, int *lineGap);
// ascent is the coordinate above the baseline the font extends; descent
// is the coordinate below the baseline the font extends (i.e. it is typically negative)
// lineGap is the spacing between one row's descent and the next row's ascent...
// so you should advance the vertical position by "*ascent - *descent + *lineGap"
//   these are expressed in unscaled coordinates, so you must multiply by
//   the scale factor for a given size

STBTT_DEF int  stbtt_GetFontVMetricsOS2(const stbtt_fontinfo *info, int *typoAscent, int *typoDescent, int *typoLineGap);
// analogous to GetFontVMetrics, but returns the "typographic" values from the OS/2
// table (specific to MS/Windows TTF files).
//
// Returns 1 on success (table present), 0 on failure.

STBTT_DEF void stbtt_GetFontBoundingBox(const stbtt_fontinfo *info, int *x0, int *y0, int *x1, int *y1);
// the bounding box around all possible characters

STBTT_DEF void stbtt_GetCodepointHMetrics(const stbtt_fontinfo *info, int codepoint, int *advanceWidth, int *leftSideBearing);
// leftSideBearing is the offset from the current horizontal position to the left edge of the character
// advanceWidth is the offset from the current horizontal position to the next horizontal position
//   these are expressed in unscaled coordinates

STBTT_DEF int  stbtt_GetCodepointKernAdvance(const stbtt_fontinfo *info, int ch1, int ch2);
// an additional amount to add to the 'advance' value between ch1 and ch2

STBTT_DEF int stbtt_GetCodepointBox(const stbtt_fontinfo *info, int codepoint, int *x0, int *y0, int *x1, int *y1);
// Gets the bounding box of the visible part of the glyph, in unscaled coordinates

STBTT_DEF void stbtt_GetGlyphHMetrics(const stbtt_fontinfo *info, int glyph_index, int *advanceWidth, int *leftSideBearing);
STBTT_DEF int  stbtt_GetGlyphKernAdvance(const stbtt_fontinfo *info, int glyph1, int glyph2);
STBTT_DEF int  stbtt_GetGlyphBox(const stbtt_fontinfo *info, int glyph_index, int *x0, int *y0, int *x1, int *y1);
// as above, but takes one or more glyph indices for greater efficiency

typedef struct stbtt_kerningentry
{
   int glyph1; // use stbtt_FindGlyphIndex
   int glyph2;
   int advance;
} stbtt_kerningentry;

STBTT_DEF int  stbtt_GetKerningTableLength(const stbtt_fontinfo *info);
STBTT_DEF int  stbtt_GetKerningTable(const stbtt_fontinfo *info, stbtt_kerningentry* table, int table_length);
// Retrieves a complete list of all of the kerning pairs provided by the font
// stbtt_GetKerningTable never writes more than table_length entries and returns how many entries it did write.
// The table will be sorted by (a.glyph1 == b.glyph1)?(a.glyph2 < b.glyph2):(a.glyph1 < b.glyph1)

//////////////////////////////////////////////////////////////////////////////
//
// GLYPH SHAPES (you probably don't need these, but they have to go before
// the bitmaps for C declaration-order reasons)
//

#ifndef STBTT_vmove // you can predefine these to use different values (but why?)
   enum {
      STBTT_vmove=1,
      STBTT_vline,
      STBTT_vcurve,
      STBTT_vcubic
   };
#endif

#ifndef stbtt_vertex // you can predefine this to use different values
                   // (we share this with other code at RAD)
   #define stbtt_vertex_type short // can't use stbtt_int16 because that's not visible in the header file
   typedef struct
   {
      stbtt_vertex_type x,y,cx,cy,cx1,cy1;
      unsigned char type,padding;
   } stbtt_vertex;
#endif

STBTT_DEF int stbtt_IsGlyphEmpty(const stbtt_fontinfo *info, int glyph_index);
// returns non-zero if nothing is drawn for this glyph

STBTT_DEF int stbtt_GetCodepointShape(const stbtt_fontinfo *info, int unicode_codepoint, stbtt_vertex **vertices);
STBTT_DEF int stbtt_GetGlyphShape(const stbtt_fontinfo *info, int glyph_index, stbtt_vertex **vertices);
// returns # of vertices and fills *vertices with the pointer to them
//   these are expressed in "unscaled" coordinates
//
// The shape is a series of contours. Each one starts with
// a STBTT_moveto, then consists of a series of mixed
// STBTT_lineto and STBTT_curveto segments. A lineto
// draws a line from previous endpoint to its x,y; a curveto
// draws a quadratic bezier from previous endpoint to
// its x,y, using cx,cy as the bezier control point.

STBTT_DEF void stbtt_FreeShape(const stbtt_fontinfo *info, stbtt_vertex *vertices);
// frees the data allocated above

STBTT_DEF unsigned char *stbtt_FindSVGDoc(const stbtt_fontinfo *info, int gl);
STBTT_DEF int stbtt_GetCodepointSVG(const stbtt_fontinfo *info, int unicode_codepoint, const char **svg);
STBTT_DEF int stbtt_GetGlyphSVG(const stbtt_fontinfo *info, int gl, const char **svg);
// fills svg with the character's SVG data.
// returns data size or 0 if SVG not found.

//////////////////////////////////////////////////////////////////////////////
//
// BITMAP RENDERING
//

STBTT_DEF void stbtt_FreeBitmap(unsigned char *bitmap, void *userdata);
// frees the bitmap allocated below

STBTT_DEF unsigned char *stbtt_GetCodepointBitmap(const stbtt_fontinfo *info, float scale_x, float scale_y, int codepoint, int *width, int *height, int *xoff, int *yoff);
// allocates a large-enough single-channel 8bpp bitmap and renders the
// specified character/glyph at the specified scale into it, with
// antialiasing. 0 is no coverage (transparent), 255 is fully covered (opaque).
// *width & *height are filled out with the width & height of the bitmap,
// which is stored left-to-right, top-to-bottom.
//
// xoff/yoff are the offset it pixel space from the glyph origin to the top-left of the bitmap

STBTT_DEF unsigned char *stbtt_GetCodepointBitmapSubpixel(const stbtt_fontinfo *info, float scale_x, float scale_y, float shift_x, float shift_y, int codepoint, int *width, int *height, int *xoff, int *yoff);
// the same as stbtt_GetCodepoitnBitmap, but you can specify a subpixel
// shift for the character

STBTT_DEF void stbtt_MakeCodepointBitmap(const stbtt_fontinfo *info, unsigned char *output, int out_w, int out_h, int out_stride, float scale_x, float scale_y, int codepoint);
// the same as stbtt_GetCodepointBitmap, but you pass in storage for the bitmap
// in the form of 'output', with row spacing of 'out_stride' bytes. the bitmap
// is clipped to out_w/out_h bytes. Call stbtt_GetCodepointBitmapBox to get the
// width and height and positioning info for it first.

STBTT_DEF void stbtt_MakeCodepointBitmapSubpixel(const stbtt_fontinfo *info, unsigned char *output, int out_w, int out_h, int out_stride, float scale_x, float scale_y, float shift_x, float shift_y, int codepoint);
// same as stbtt_MakeCodepointBitmap, but you can specify a subpixel
// shift for the character

STBTT_DEF void stbtt_MakeCodepointBitmapSubpixelPrefilter(const stbtt_fontinfo *info, unsigned char *output, int out_w, int out_h, int out_stride, float scale_x, float scale_y, float shift_x, float shift_y, int oversample_x, int oversample_y, float *sub_x, float *sub_y, int codepoint);
// same as stbtt_MakeCodepointBitmapSubpixel, but prefiltering
// is performed (see stbtt_PackSetOversampling)

STBTT_DEF void stbtt_GetCodepointBitmapBox(const stbtt_fontinfo *font, int codepoint, float scale_x, float scale_y, int *ix0, int *iy0, int *ix1, int *iy1);
// get the bbox of the bitmap centered around the glyph origin; so the
// bitmap width is ix1-ix0, height is iy1-iy0, and location to place
// the bitmap top left is (leftSideBearing*scale,iy0).
// (Note that the bitmap uses y-increases-down, but the shape uses
// y-increases-up, so CodepointBitmapBox and CodepointBox are inverted.)

STBTT_DEF void stbtt_GetCodepointBitmapBoxSubpixel(const stbtt_fontinfo *font, int codepoint, float scale_x, float scale_y, float shift_x, float shift_y, int *ix0, int *iy0, int *ix1, int *iy1);
// same as stbtt_GetCodepointBitmapBox, but you can specify a subpixel
// shift for the character

// the following functions are equivalent to the above functions, but operate
// on glyph indices instead of Unicode codepoints (for efficiency)
STBTT_DEF unsigned char *stbtt_GetGlyphBitmap(const stbtt_fontinfo *info, float scale_x, float scale_y, int glyph, int *width, int *height, int *xoff, int *yoff);
STBTT_DEF unsigned char *stbtt_GetGlyphBitmapSubpixel(const stbtt_fontinfo *info, float scale_x, float scale_y, float shift_x, float shift_y, int glyph, int *width, int *height, int *xoff, int *yoff);
STBTT_DEF void stbtt_MakeGlyphBitmap(const stbtt_fontinfo *info, unsigned char *output, int out_w, int out_h, int out_stride, float scale_x, float scale_y, int glyph);
STBTT_DEF void stbtt_MakeGlyphBitmapSubpixel(const stbtt_fontinfo *info, unsigned char *output, int out_w, int out_h, int out_stride, float scale_x, float scale_y, float shift_x, float shift_y, int glyph);
STBTT_DEF void stbtt_MakeGlyphBitmapSubpixelPrefilter(const stbtt_fontinfo *info, unsigned char *output, int out_w, int out_h, int out_stride, float scale_x, float scale_y, float shift_x, float shift_y, int oversample_x, int oversample_y, float *sub_x, float *sub_y, int glyph);
STBTT_DEF void stbtt_GetGlyphBitmapBox(const stbtt_fontinfo *font, int glyph, float scale_x, float scale_y, int *ix0, int *iy0, int *ix1, int *iy1);
STBTT_DEF void stbtt_GetGlyphBitmapBoxSubpixel(const stbtt_fontinfo *font, int glyph, float scale_x, float scale_y,float shift_x, float shift_y, int *ix0, int *iy0, int *ix1, int *iy1);


// @TODO: don't expose this structure
typedef struct
{
   int w,h,stride;
   unsigned char *pixels;
} stbtt__bitmap;

// rasterize a shape with quadratic beziers into a bitmap
STBTT_DEF void stbtt_Rasterize(stbtt__bitmap *result,        // 1-channel bitmap to draw into
                               float flatness_in_pixels,     // allowable error of curve in pixels
                               stbtt_vertex *vertices,       // array of vertices defining shape
                               int num_verts,                // number of vertices in above array
                               float scale_x, float scale_y, // scale applied to input vertices
                               float shift_x, float shift_y, // translation applied to input vertices
                               int x_off, int y_off,         // another translation applied to input
                               int invert,                   // if non-zero, vertically flip shape
                               void *userdata);              // context for to STBTT_MALLOC

//////////////////////////////////////////////////////////////////////////////
//
// Signed Distance Function (or Field) rendering

STBTT_DEF void stbtt_FreeSDF(unsigned char *bitmap, void *userdata);
// frees the SDF bitmap allocated below

STBTT_DEF unsigned char * stbtt_GetGlyphSDF(const stbtt_fontinfo *info, float scale, int glyph, int padding, unsigned char onedge_value, float pixel_dist_scale, int *width, int *height, int *xoff, int *yoff);
STBTT_DEF unsigned char * stbtt_GetCodepointSDF(const stbtt_fontinfo *info, float scale, int codepoint, int padding, unsigned char onedge_value, float pixel_dist_scale, int *width, int *height, int *xoff, int *yoff);
// These functions compute a discretized SDF field for a single character, suitable for storing
// in a single-channel texture, sampling with bilinear filtering, and testing against
// larger than some threshold to produce scalable fonts.
//        info              --  the font
//        scale             --  controls the size of the resulting SDF bitmap, same as it would be creating a regular bitmap
//        glyph/codepoint   --  the character to generate the SDF for
//        padding           --  extra "pixels" around the character which are filled with the distance to the character (not 0),
//                                 which allows effects like bit outlines
//        onedge_value      --  value 0-255 to test the SDF against to reconstruct the character (i.e. the isocontour of the character)
//        pixel_dist_scale  --  what value the SDF should increase by when moving one SDF "pixel" away from the edge (on the 0..255 scale)
//                                 if positive, > onedge_value is inside; if negative, < onedge_value is inside
//        width,height      --  output height & width of the SDF bitmap (including padding)
//        xoff,yoff         --  output origin of the character
//        return value      --  a 2D array of bytes 0..255, width*height in size
//
// pixel_dist_scale & onedge_value are a scale & bias that allows you to make
// optimal use of the limited 0..255 for your application, trading off precision
// and special effects. SDF values outside the range 0..255 are clamped to 0..255.
//
// Example:
//      scale = stbtt_ScaleForPixelHeight(22)
//      padding = 5
//      onedge_value = 180
//      pixel_dist_scale = 180/5.0 = 36.0
//
//      This will create an SDF bitmap in which the character is about 22 pixels
//      high but the whole bitmap is about 22+5+5=32 pixels high. To produce a filled
//      shape, sample the SDF at each pixel and fill the pixel if the SDF value
//      is greater than or equal to 180/255. (You'll actually want to antialias,
//      which is beyond the scope of this example.) Additionally, you can compute
//      offset outlines (e.g. to stroke the character border inside & outside,
//      or only outside). For example, to fill outside the character up to 3 SDF
//      pixels, you would compare against (180-36.0*3)/255 = 72/255. The above
//      choice of variables maps a range from 5 pixels outside the shape to
//      2 pixels inside the shape to 0..255; this is intended primarily for apply
//      outside effects only (the interior range is needed to allow proper
//      antialiasing of the font at *smaller* sizes)
//
// The function computes the SDF analytically at each SDF pixel, not by e.g.
// building a higher-res bitmap and approximating it. In theory the quality
// should be as high as possible for an SDF of this size & representation, but
// unclear if this is true in practice (perhaps building a higher-res bitmap
// and computing from that can allow drop-out prevention).
//
// The algorithm has not been optimized at all, so expect it to be slow
// if computing lots of characters or very large sizes.



//////////////////////////////////////////////////////////////////////////////
//
// Finding the right font...
//
// You should really just solve this offline, keep your own tables
// of what font is what, and don't try to get it out of the .ttf file.
// That's because getting it out of the .ttf file is really hard, because
// the names in the file can appear in many possible encodings, in many
// possible languages, and e.g. if you need a case-insensitive comparison,
// the details of that depend on the encoding & language in a complex way
// (actually underspecified in truetype, but also gigantic).
//
// But you can use the provided functions in two possible ways:
//     stbtt_FindMatchingFont() will use *case-sensitive* comparisons on
//             unicode-encoded names to try to find the font you want;
//             you can run this before calling stbtt_InitFont()
//
//     stbtt_GetFontNameString() lets you get any of the various strings
//             from the file yourself and do your own comparisons on them.
//             You have to have called stbtt_InitFont() first.


STBTT_DEF int stbtt_FindMatchingFont(const unsigned char *fontdata, const char *name, int flags);
// returns the offset (not index) of the font that matches, or -1 if none
//   if you use STBTT_MACSTYLE_DONTCARE, use a font name like "Arial Bold".
//   if you use any other flag, use a font name like "Arial"; this checks
//     the 'macStyle' header field; i don't know if fonts set this consistently
#define STBTT_MACSTYLE_DONTCARE     0
#define STBTT_MACSTYLE_BOLD         1
#define STBTT_MACSTYLE_ITALIC       2
#define STBTT_MACSTYLE_UNDERSCORE   4
#define STBTT_MACSTYLE_NONE         8   // <= not same as 0, this makes us check the bitfield is 0

STBTT_DEF int stbtt_CompareUTF8toUTF16_bigendian(const char *s1, int len1, const char *s2, int len2);
// returns 1/0 whether the first string interpreted as utf8 is identical to
// the second string interpreted as big-endian utf16... useful for strings from next func

STBTT_DEF const char *stbtt_GetFontNameString(const stbtt_fontinfo *font, int *length, int platformID, int encodingID, int languageID, int nameID);
// returns the string (which may be big-endian double byte, e.g. for unicode)
// and puts the length in bytes in *length.
//
// some of the values for the IDs are below; for more see the truetype spec:
//     http://developer.apple.com/textfonts/TTRefMan/RM06/Chap6name.html
//     http://www.microsoft.com/typography/otspec/name.htm

enum { // platformID
   STBTT_PLATFORM_ID_UNICODE   =0,
   STBTT_PLATFORM_ID_MAC       =1,
   STBTT_PLATFORM_ID_ISO       =2,
   STBTT_PLATFORM_ID_MICROSOFT =3
};

enum { // encodingID for STBTT_PLATFORM_ID_UNICODE
   STBTT_UNICODE_EID_UNICODE_1_0    =0,
   STBTT_UNICODE_EID_UNICODE_1_1    =1,
   STBTT_UNICODE_EID_ISO_10646      =2,
   STBTT_UNICODE_EID_UNICODE_2_0_BMP=3,
   STBTT_UNICODE_EID_UNICODE_2_0_FULL=4
};

enum { // encodingID for STBTT_PLATFORM_ID_MICROSOFT
   STBTT_MS_EID_SYMBOL        =0,
   STBTT_MS_EID_UNICODE_BMP   =1,
   STBTT_MS_EID_SHIFTJIS      =2,
   STBTT_MS_EID_UNICODE_FULL  =10
};

enum { // encodingID for STBTT_PLATFORM_ID_MAC; same as Script Manager codes
   STBTT_MAC_EID_ROMAN        =0,   STBTT_MAC_EID_ARABIC       =4,
   STBTT_MAC_EID_JAPANESE     =1,   STBTT_MAC_EID_HEBREW       =5,
   STBTT_MAC_EID_CHINESE_TRAD =2,   STBTT_MAC_EID_GREEK        =6,
   STBTT_MAC_EID_KOREAN       =3,   STBTT_MAC_EID_RUSSIAN      =7
};

enum { // languageID for STBTT_PLATFORM_ID_MICROSOFT; same as LCID...
       // problematic because there are e.g. 16 english LCIDs and 16 arabic LCIDs
   STBTT_MS_LANG_ENGLISH     =0x0409,   STBTT_MS_LANG_ITALIAN     =0x0410,
   STBTT_MS_LANG_CHINESE     =0x0804,   STBTT_MS_LANG_JAPANESE    =0x0411,
   STBTT_MS_LANG_DUTCH       =0x0413,   STBTT_MS_LANG_KOREAN      =0x0412,
   STBTT_MS_LANG_FRENCH      =0x040c,   STBTT_MS_LANG_RUSSIAN     =0x0419,
   STBTT_MS_LANG_GERMAN      =0x0407,   STBTT_MS_LANG_SPANISH     =0x0409,
   STBTT_MS_LANG_HEBREW      =0x040d,   STBTT_MS_LANG_SWEDISH     =0x041D
};

enum { // languageID for STBTT_PLATFORM_ID_MAC
   STBTT_MAC_LANG_ENGLISH      =0 ,   STBTT_MAC_LANG_JAPANESE     =11,
   STBTT_MAC_LANG_ARABIC       =12,   STBTT_MAC_LANG_KOREAN       =23,
   STBTT_MAC_LANG_DUTCH        =4 ,   STBTT_MAC_LANG_RUSSIAN      =32,
   STBTT_MAC_LANG_FRENCH       =1 ,   STBTT_MAC_LANG_SPANISH      =6 ,
   STBTT_MAC_LANG_GERMAN       =2 ,   STBTT_MAC_LANG_SWEDISH      =5 ,
   STBTT_MAC_LANG_HEBREW       =10,   STBTT_MAC_LANG_CHINESE_SIMPLIFIED =33,
   STBTT_MAC_LANG_ITALIAN      =3 ,   STBTT_MAC_LANG_CHINESE_TRAD =19
};

#ifdef __cplusplus
}
#endif

#endif // __STB_INCLUDE_STB_TRUETYPE_H__

///////////////////////////////////////////////////////////////////////////////
///////////////////////////////////////////////////////////////////////////////
////
////   IMPLEMENTATION
////
////

#ifdef STB_TRUETYPE_IMPLEMENTATION

#ifndef STBTT_MAX_OVERSAMPLE
#define STBTT_MAX_OVERSAMPLE   8
#endif

#if STBTT_MAX_OVERSAMPLE > 255
#error "STBTT_MAX_OVERSAMPLE cannot be > 255"
#endif

typedef int stbtt__test_oversample_pow2[(STBTT_MAX_OVERSAMPLE & (STBTT_MAX_OVERSAMPLE-1)) == 0 ? 1 : -1];

#ifndef STBTT_RASTERIZER_VERSION
#define STBTT_RASTERIZER_VERSION 2
#endif

#ifdef _MSC_VER
#define STBTT__NOTUSED(v)  (void)(v)
#else
#define STBTT__NOTUSED(v)  (void)sizeof(v)
#endif

//////////////////////////////////////////////////////////////////////////
//
// stbtt__buf helpers to parse data from file
//

static stbtt_uint8 stbtt__buf_get8(stbtt__buf *b)
{
   if (b->cursor >= b->size)
      return 0;
   return b->data[b->cursor++];
}

static stbtt_uint8 stbtt__buf_peek8(stbtt__buf *b)
{
   if (b->cursor >= b->size)
      return 0;
   return b->data[b->cursor];
}

static void stbtt__buf_seek(stbtt__buf *b, int o)
{
   STBTT_assert(!(o > b->size || o < 0));
   b->cursor = (o > b->size || o < 0) ? b->size : o;
}

static void stbtt__buf_skip(stbtt__buf *b, int o)
{
   stbtt__buf_seek(b, b->cursor + o);
}

static stbtt_uint32 stbtt__buf_get(stbtt__buf *b, int n)
{
   stbtt_uint32 v = 0;
   int i;
   STBTT_assert(n >= 1 && n <= 4);
   for (i = 0; i < n; i++)
      v = (v << 8) | stbtt__buf_get8(b);
   return v;
}

static stbtt__buf stbtt__new_buf(const void *p, size_t size)
{
   stbtt__buf r;
   STBTT_assert(size < 0x40000000);
   r.data = (stbtt_uint8*) p;
   r.size = (int) size;
   r.cursor = 0;
   return r;
}

#define stbtt__buf_get16(b)  stbtt__buf_get((b), 2)
#define stbtt__buf_get32(b)  stbtt__buf_get((b), 4)

static stbtt__buf stbtt__buf_range(const stbtt__buf *b, int o, int s)
{
   stbtt__buf r = stbtt__new_buf(NULL, 0);
   if (o < 0 || s < 0 || o > b->size || s > b->size - o) return r;
   r.data = b->data + o;
   r.size = s;
   return r;
}

static stbtt__buf stbtt__cff_get_index(stbtt__buf *b)
{
   int count, start, offsize;
   start = b->cursor;
   count = stbtt__buf_get16(b);
   if (count) {
      offsize = stbtt__buf_get8(b);
      STBTT_assert(offsize >= 1 && offsize <= 4);
      stbtt__buf_skip(b, offsize * count);
      stbtt__buf_skip(b, stbtt__buf_get(b, offsize) - 1);
   }
   return stbtt__buf_range(b, start, b->cursor - start);
}

static stbtt_uint32 stbtt__cff_int(stbtt__buf *b)
{
   int b0 = stbtt__buf_get8(b);
   if (b0 >= 32 && b0 <= 246)       return b0 - 139;
   else if (b0 >= 247 && b0 <= 250) return (b0 - 247)*256 + stbtt__buf_get8(b) + 108;
   else if (b0 >= 251 && b0 <= 254) return -(b0 - 251)*256 - stbtt__buf_get8(b) - 108;
   else if (b0 == 28)               return stbtt__buf_get16(b);
   else if (b0 == 29)               return stbtt__buf_get32(b);
   STBTT_assert(0);
   return 0;
}

static void stbtt__cff_skip_operand(stbtt__buf *b) {
   int v, b0 = stbtt__buf_peek8(b);
   STBTT_assert(b0 >= 28);
   if (b0 == 30) {
      stbtt__buf_skip(b, 1);
      while (b->cursor < b->size) {
         v = stbtt__buf_get8(b);
         if ((v & 0xF) == 0xF || (v >> 4) == 0xF)
            break;
      }
   } else {
      stbtt__cff_int(b);
   }
}

static stbtt__buf stbtt__dict_get(stbtt__buf *b, int key)
{
   stbtt__buf_seek(b, 0);
   while (b->cursor < b->size) {
      int start = b->cursor, end, op;
      while (stbtt__buf_peek8(b) >= 28)
         stbtt__cff_skip_operand(b);
      end = b->cursor;
      op = stbtt__buf_get8(b);
      if (op == 12)  op = stbtt__buf_get8(b) | 0x100;
      if (op == key) return stbtt__buf_range(b, start, end-start);
   }
   return stbtt__buf_range(b, 0, 0);
}

static void stbtt__dict_get_ints(stbtt__buf *b, int key, int outcount, stbtt_uint32 *out)
{
   int i;
   stbtt__buf operands = stbtt__dict_get(b, key);
   for (i = 0; i < outcount && operands.cursor < operands.size; i++)
      out[i] = stbtt__cff_int(&operands);
}

static int stbtt__cff_index_count(stbtt__buf *b)
{
   stbtt__buf_seek(b, 0);
   return stbtt__buf_get16(b);
}

static stbtt__buf stbtt__cff_index_get(stbtt__buf b, int i)
{
   int count, offsize, start, end;
   stbtt__buf_seek(&b, 0);
   count = stbtt__buf_get16(&b);
   offsize = stbtt__buf_get8(&b);
   STBTT_assert(i >= 0 && i < count);
   STBTT_assert(offsize >= 1 && offsize <= 4);
   stbtt__buf_skip(&b, i*offsize);
   start = stbtt__buf_get(&b, offsize);
   end = stbtt__buf_get(&b, offsize);
   return stbtt__buf_range(&b, 2+(count+1)*offsize+start, end - start);
}

//////////////////////////////////////////////////////////////////////////
//
// accessors to parse data from file
//

// on platforms that don't allow misaligned reads, if we want to allow
// truetype fonts that aren't padded to alignment, define ALLOW_UNALIGNED_TRUETYPE

#define ttBYTE(p)     (* (stbtt_uint8 *) (p))
#define ttCHAR(p)     (* (stbtt_int8 *) (p))
#define ttFixed(p)    ttLONG(p)

static stbtt_uint16 ttUSHORT(stbtt_uint8 *p) { return p[0]*256 + p[1]; }
static stbtt_int16 ttSHORT(stbtt_uint8 *p)   { return p[0]*256 + p[1]; }
static stbtt_uint32 ttULONG(stbtt_uint8 *p)  { return (p[0]<<24) + (p[1]<<16) + (p[2]<<8) + p[3]; }
static stbtt_int32 ttLONG(stbtt_uint8 *p)    { return (p[0]<<24) + (p[1]<<16) + (p[2]<<8) + p[3]; }

#define stbtt_tag4(p,c0,c1,c2,c3) ((p)[0] == (c0) && (p)[1] == (c1) && (p)[2] == (c2) && (p)[3] == (c3))
#define stbtt_tag(p,str)           stbtt_tag4(p,str[0],str[1],str[2],str[3])

static int stbtt__isfont(stbtt_uint8 *font)
{
   // check the version number
   if (stbtt_tag4(font, '1',0,0,0))  return 1; // TrueType 1
   if (stbtt_tag(font, "typ1"))   return 1; // TrueType with type 1 font -- we don't support this!
   if (stbtt_tag(font, "OTTO"))   return 1; // OpenType with CFF
   if (stbtt_tag4(font, 0,1,0,0)) return 1; // OpenType 1.0
   if (stbtt_tag(font, "true"))   return 1; // Apple specification for TrueType fonts
   return 0;
}

// @OPTIMIZE: binary search
static stbtt_uint32 stbtt__find_table(stbtt_uint8 *data, stbtt_uint32 fontstart, const char *tag)
{
   stbtt_int32 num_tables = ttUSHORT(data+fontstart+4);
   stbtt_uint32 tabledir = fontstart + 12;
   stbtt_int32 i;
   for (i=0; i < num_tables; ++i) {
      stbtt_uint32 loc = tabledir + 16*i;
      if (stbtt_tag(data+loc+0, tag))
         return ttULONG(data+loc+8);
   }
   return 0;
}

static int stbtt_GetFontOffsetForIndex_internal(unsigned char *font_collection, int index)
{
   // if it's just a font, there's only one valid index
   if (stbtt__isfont(font_collection))
      return index == 0 ? 0 : -1;

   // check if it's a TTC
   if (stbtt_tag(font_collection, "ttcf")) {
      // version 1?
      if (ttULONG(font_collection+4) == 0x00010000 || ttULONG(font_collection+4) == 0x00020000) {
         stbtt_int32 n = ttLONG(font_collection+8);
         if (index >= n)
            return -1;
         return ttULONG(font_collection+12+index*4);
      }
   }
   return -1;
}

static int stbtt_GetNumberOfFonts_internal(unsigned char *font_collection)
{
   // if it's just a font, there's only one valid font
   if (stbtt__isfont(font_collection))
      return 1;

   // check if it's a TTC
   if (stbtt_tag(font_collection, "ttcf")) {
      // version 1?
      if (ttULONG(font_collection+4) == 0x00010000 || ttULONG(font_collection+4) == 0x00020000) {
         return ttLONG(font_collection+8);
      }
   }
   return 0;
}

static stbtt__buf stbtt__get_subrs(stbtt__buf cff, stbtt__buf fontdict)
{
   stbtt_uint32 subrsoff = 0, private_loc[2] = { 0, 0 };
   stbtt__buf pdict;
   stbtt__dict_get_ints(&fontdict, 18, 2, private_loc);
   if (!private_loc[1] || !private_loc[0]) return stbtt__new_buf(NULL, 0);
   pdict = stbtt__buf_range(&cff, private_loc[1], private_loc[0]);
   stbtt__dict_get_ints(&pdict, 19, 1, &subrsoff);
   if (!subrsoff) return stbtt__new_buf(NULL, 0);
   stbtt__buf_seek(&cff, private_loc[1]+subrsoff);
   return stbtt__cff_get_index(&cff);
}

// since most people won't use this, find this table the first time it's needed
static int stbtt__get_svg(stbtt_fontinfo *info)
{
   stbtt_uint32 t;
   if (info->svg < 0) {
      t = stbtt__find_table(info->data, info->fontstart, "SVG ");
      if (t) {
         stbtt_uint32 offset = ttULONG(info->data + t + 2);
         info->svg = t + offset;
      } else {
         info->svg = 0;
      }
   }
   return info->svg;
}

static int stbtt_InitFont_internal(stbtt_fontinfo *info, unsigned char *data, int fontstart)
{
   stbtt_uint32 cmap, t;
   stbtt_int32 i,numTables;

   info->data = data;
   info->fontstart = fontstart;
   info->cff = stbtt__new_buf(NULL, 0);

   cmap = stbtt__find_table(data, fontstart, "cmap");       // required
   info->loca = stbtt__find_table(data, fontstart, "loca"); // required
   info->head = stbtt__find_table(data, fontstart, "head"); // required
   info->glyf = stbtt__find_table(data, fontstart, "glyf"); // required
   info->hhea = stbtt__find_table(data, fontstart, "hhea"); // required
   info->hmtx = stbtt__find_table(data, fontstart, "hmtx"); // required
   info->kern = stbtt__find_table(data, fontstart, "kern"); // not required
   info->gpos = stbtt__find_table(data, fontstart, "GPOS"); // not required

   if (!cmap || !info->head || !info->hhea || !info->hmtx)
      return 0;
   if (info->glyf) {
      // required for truetype
      if (!info->loca) return 0;
   } else {
      // initialization for CFF / Type2 fonts (OTF)
      stbtt__buf b, topdict, topdictidx;
      stbtt_uint32 cstype = 2, charstrings = 0, fdarrayoff = 0, fdselectoff = 0;
      stbtt_uint32 cff;

      cff = stbtt__find_table(data, fontstart, "CFF ");
      if (!cff) return 0;

      info->fontdicts = stbtt__new_buf(NULL, 0);
      info->fdselect = stbtt__new_buf(NULL, 0);

      // @TODO this should use size from table (not 512MB)
      info->cff = stbtt__new_buf(data+cff, 512*1024*1024);
      b = info->cff;

      // read the header
      stbtt__buf_skip(&b, 2);
      stbtt__buf_seek(&b, stbtt__buf_get8(&b)); // hdrsize

      // @TODO the name INDEX could list multiple fonts,
      // but we just use the first one.
      stbtt__cff_get_index(&b);  // name INDEX
      topdictidx = stbtt__cff_get_index(&b);
      topdict = stbtt__cff_index_get(topdictidx, 0);
      stbtt__cff_get_index(&b);  // string INDEX
      info->gsubrs = stbtt__cff_get_index(&b);

      stbtt__dict_get_ints(&topdict, 17, 1, &charstrings);
      stbtt__dict_get_ints(&topdict, 0x100 | 6, 1, &cstype);
      stbtt__dict_get_ints(&topdict, 0x100 | 36, 1, &fdarrayoff);
      stbtt__dict_get_ints(&topdict, 0x100 | 37, 1, &fdselectoff);
      info->subrs = stbtt__get_subrs(b, topdict);

      // we only support Type 2 charstrings
      if (cstype != 2) return 0;
      if (charstrings == 0) return 0;

      if (fdarrayoff) {
         // looks like a CID font
         if (!fdselectoff) return 0;
         stbtt__buf_seek(&b, fdarrayoff);
         info->fontdicts = stbtt__cff_get_index(&b);
         info->fdselect = stbtt__buf_range(&b, fdselectoff, b.size-fdselectoff);
      }

      stbtt__buf_seek(&b, charstrings);
      info->charstrings = stbtt__cff_get_index(&b);
   }

   t = stbtt__find_table(data, fontstart, "maxp");
   if (t)
      info->numGlyphs = ttUSHORT(data+t+4);
   else
      info->numGlyphs = 0xffff;

   info->svg = -1;

   // find a cmap encoding table we understand *now* to avoid searching
   // later. (todo: could make this installable)
   // the same regardless of glyph.
   numTables = ttUSHORT(data + cmap + 2);
   info->index_map = 0;
   for (i=0; i < numTables; ++i) {
      stbtt_uint32 encoding_record = cmap + 4 + 8 * i;
      // find an encoding we understand:
      switch(ttUSHORT(data+encoding_record)) {
         case STBTT_PLATFORM_ID_MICROSOFT:
            switch (ttUSHORT(data+encoding_record+2)) {
               case STBTT_MS_EID_UNICODE_BMP:
               case STBTT_MS_EID_UNICODE_FULL:
                  // MS/Unicode
                  info->index_map = cmap + ttULONG(data+encoding_record+4);
                  break;
            }
            break;
        case STBTT_PLATFORM_ID_UNICODE:
            // Mac/iOS has these
            // all the encodingIDs are unicode, so we don't bother to check it
            info->index_map = cmap + ttULONG(data+encoding_record+4);
            break;
      }
   }
   if (info->index_map == 0)
      return 0;

   info->indexToLocFormat = ttUSHORT(data+info->head + 50);
   return 1;
}

STBTT_DEF int stbtt_FindGlyphIndex(const stbtt_fontinfo *info, int unicode_codepoint)
{
   stbtt_uint8 *data = info->data;
   stbtt_uint32 index_map = info->index_map;

   stbtt_uint16 format = ttUSHORT(data + index_map + 0);
   if (format == 0) { // apple byte encoding
      stbtt_int32 bytes = ttUSHORT(data + index_map + 2);
      if (unicode_codepoint < bytes-6)
         return ttBYTE(data + index_map + 6 + unicode_codepoint);
      return 0;
   } else if (format == 6) {
      stbtt_uint32 first = ttUSHORT(data + index_map + 6);
      stbtt_uint32 count = ttUSHORT(data + index_map + 8);
      if ((stbtt_uint32) unicode_codepoint >= first && (stbtt_uint32) unicode_codepoint < first+count)
         return ttUSHORT(data + index_map + 10 + (unicode_codepoint - first)*2);
      return 0;
   } else if (format == 2) {
      STBTT_assert(0); // @TODO: high-byte mapping for japanese/chinese/korean
      return 0;
   } else if (format == 4) { // standard mapping for windows fonts: binary search collection of ranges
      stbtt_uint16 segcount = ttUSHORT(data+index_map+6) >> 1;
      stbtt_uint16 searchRange = ttUSHORT(data+index_map+8) >> 1;
      stbtt_uint16 entrySelector = ttUSHORT(data+index_map+10);
      stbtt_uint16 rangeShift = ttUSHORT(data+index_map+12) >> 1;

      // do a binary search of the segments
      stbtt_uint32 endCount = index_map + 14;
      stbtt_uint32 search = endCount;

      if (unicode_codepoint > 0xffff)
         return 0;

      // they lie from endCount .. endCount + segCount
      // but searchRange is the nearest power of two, so...
      if (unicode_codepoint >= ttUSHORT(data + search + rangeShift*2))
         search += rangeShift*2;

      // now decrement to bias correctly to find smallest
      search -= 2;
      while (entrySelector) {
         stbtt_uint16 end;
         searchRange >>= 1;
         end = ttUSHORT(data + search + searchRange*2);
         if (unicode_codepoint > end)
            search += searchRange*2;
         --entrySelector;
      }
      search += 2;

      {
         stbtt_uint16 offset, start, last;
         stbtt_uint16 item = (stbtt_uint16) ((search - endCount) >> 1);

         start = ttUSHORT(data + index_map + 14 + segcount*2 + 2 + 2*item);
         last = ttUSHORT(data + endCount + 2*item);
         if (unicode_codepoint < start || unicode_codepoint > last)
            return 0;

         offset = ttUSHORT(data + index_map + 14 + segcount*6 + 2 + 2*item);
         if (offset == 0)
            return (stbtt_uint16) (unicode_codepoint + ttSHORT(data + index_map + 14 + segcount*4 + 2 + 2*item));

         return ttUSHORT(data + offset + (unicode_codepoint-start)*2 + index_map + 14 + segcount*6 + 2 + 2*item);
      }
   } else if (format == 12 || format == 13) {
      stbtt_uint32 ngroups = ttULONG(data+index_map+12);
      stbtt_int32 low,high;
      low = 0; high = (stbtt_int32)ngroups;
      // Binary search the right group.
      while (low < high) {
         stbtt_int32 mid = low + ((high-low) >> 1); // rounds down, so low <= mid < high
         stbtt_uint32 start_char = ttULONG(data+index_map+16+mid*12);
         stbtt_uint32 end_char = ttULONG(data+index_map+16+mid*12+4);
         if ((stbtt_uint32) unicode_codepoint < start_char)
            high = mid;
         else if ((stbtt_uint32) unicode_codepoint > end_char)
            low = mid+1;
         else {
            stbtt_uint32 start_glyph = ttULONG(data+index_map+16+mid*12+8);
            if (format == 12)
               return start_glyph + unicode_codepoint-start_char;
            else // format == 13
               return start_glyph;
         }
      }
      return 0; // not found
   }
   // @TODO
   STBTT_assert(0);
   return 0;
}

STBTT_DEF int stbtt_GetCodepointShape(const stbtt_fontinfo *info, int unicode_codepoint, stbtt_vertex **vertices)
{
   return stbtt_GetGlyphShape(info, stbtt_FindGlyphIndex(info, unicode_codepoint), vertices);
}

static void stbtt_setvertex(stbtt_vertex *v, stbtt_uint8 type, stbtt_int32 x, stbtt_int32 y, stbtt_int32 cx, stbtt_int32 cy)
{
   v->type = type;
   v->x = (stbtt_int16) x;
   v->y = (stbtt_int16) y;
   v->cx = (stbtt_int16) cx;
   v->cy = (stbtt_int16) cy;
}

static int stbtt__GetGlyfOffset(const stbtt_fontinfo *info, int glyph_index)
{
   int g1,g2;

   STBTT_assert(!info->cff.size);

   if (glyph_index >= info->numGlyphs) return -1; // glyph index out of range
   if (info->indexToLocFormat >= 2)    return -1; // unknown index->glyph map format

   if (info->indexToLocFormat == 0) {
      g1 = info->glyf + ttUSHORT(info->data + info->loca + glyph_index * 2) * 2;
      g2 = info->glyf + ttUSHORT(info->data + info->loca + glyph_index * 2 + 2) * 2;
   } else {
      g1 = info->glyf + ttULONG (info->data + info->loca + glyph_index * 4);
      g2 = info->glyf + ttULONG (info->data + info->loca + glyph_index * 4 + 4);
   }

   return g1==g2 ? -1 : g1; // if length is 0, return -1
}

static int stbtt__GetGlyphInfoT2(const stbtt_fontinfo *info, int glyph_index, int *x0, int *y0, int *x1, int *y1);

STBTT_DEF int stbtt_GetGlyphBox(const stbtt_fontinfo *info, int glyph_index, int *x0, int *y0, int *x1, int *y1)
{
   if (info->cff.size) {
      stbtt__GetGlyphInfoT2(info, glyph_index, x0, y0, x1, y1);
   } else {
      int g = stbtt__GetGlyfOffset(info, glyph_index);
      if (g < 0) return 0;

      if (x0) *x0 = ttSHORT(info->data + g + 2);
      if (y0) *y0 = ttSHORT(info->data + g + 4);
      if (x1) *x1 = ttSHORT(info->data + g + 6);
      if (y1) *y1 = ttSHORT(info->data + g + 8);
   }
   return 1;
}

STBTT_DEF int stbtt_GetCodepointBox(const stbtt_fontinfo *info, int codepoint, int *x0, int *y0, int *x1, int *y1)
{
   return stbtt_GetGlyphBox(info, stbtt_FindGlyphIndex(info,codepoint), x0,y0,x1,y1);
}

STBTT_DEF int stbtt_IsGlyphEmpty(const stbtt_fontinfo *info, int glyph_index)
{
   stbtt_int16 numberOfContours;
   int g;
   if (info->cff.size)
      return stbtt__GetGlyphInfoT2(info, glyph_index, NULL, NULL, NULL, NULL) == 0;
   g = stbtt__GetGlyfOffset(info, glyph_index);
   if (g < 0) return 1;
   numberOfContours = ttSHORT(info->data + g);
   return numberOfContours == 0;
}

static int stbtt__close_shape(stbtt_vertex *vertices, int num_vertices, int was_off, int start_off,
    stbtt_int32 sx, stbtt_int32 sy, stbtt_int32 scx, stbtt_int32 scy, stbtt_int32 cx, stbtt_int32 cy)
{
   if (start_off) {
      if (was_off)
         stbtt_setvertex(&vertices[num_vertices++], STBTT_vcurve, (cx+scx)>>1, (cy+scy)>>1, cx,cy);
      stbtt_setvertex(&vertices[num_vertices++], STBTT_vcurve, sx,sy,scx,scy);
   } else {
      if (was_off)
         stbtt_setvertex(&vertices[num_vertices++], STBTT_vcurve,sx,sy,cx,cy);
      else
         stbtt_setvertex(&vertices[num_vertices++], STBTT_vline,sx,sy,0,0);
   }
   return num_vertices;
}

static int stbtt__GetGlyphShapeTT(const stbtt_fontinfo *info, int glyph_index, stbtt_vertex **pvertices)
{
   stbtt_int16 numberOfContours;
   stbtt_uint8 *endPtsOfContours;
   stbtt_uint8 *data = info->data;
   stbtt_vertex *vertices=0;
   int num_vertices=0;
   int g = stbtt__GetGlyfOffset(info, glyph_index);

   *pvertices = NULL;

   if (g < 0) return 0;

   numberOfContours = ttSHORT(data + g);

   if (numberOfContours > 0) {
      stbtt_uint8 flags=0,flagcount;
      stbtt_int32 ins, i,j=0,m,n, next_move, was_off=0, off, start_off=0;
      stbtt_int32 x,y,cx,cy,sx,sy, scx,scy;
      stbtt_uint8 *points;
      endPtsOfContours = (data + g + 10);
      ins = ttUSHORT(data + g + 10 + numberOfContours * 2);
      points = data + g + 10 + numberOfContours * 2 + 2 + ins;

      n = 1+ttUSHORT(endPtsOfContours + numberOfContours*2-2);

      m = n + 2*numberOfContours;  // a loose bound on how many vertices we might need
      vertices = (stbtt_vertex *) STBTT_malloc(m * sizeof(vertices[0]), info->userdata);
      if (vertices == 0)
         return 0;

      next_move = 0;
      flagcount=0;

      // in first pass, we load uninterpreted data into the allocated array
      // above, shifted to the end of the array so we won't overwrite it when
      // we create our final data starting from the front

      off = m - n; // starting offset for uninterpreted data, regardless of how m ends up being calculated

      // first load flags

      for (i=0; i < n; ++i) {
         if (flagcount == 0) {
            flags = *points++;
            if (flags & 8)
               flagcount = *points++;
         } else
            --flagcount;
         vertices[off+i].type = flags;
      }

      // now load x coordinates
      x=0;
      for (i=0; i < n; ++i) {
         flags = vertices[off+i].type;
         if (flags & 2) {
            stbtt_int16 dx = *points++;
            x += (flags & 16) ? dx : -dx; // ???
         } else {
            if (!(flags & 16)) {
               x = x + (stbtt_int16) (points[0]*256 + points[1]);
               points += 2;
            }
         }
         vertices[off+i].x = (stbtt_int16) x;
      }

      // now load y coordinates
      y=0;
      for (i=0; i < n; ++i) {
         flags = vertices[off+i].type;
         if (flags & 4) {
            stbtt_int16 dy = *points++;
            y += (flags & 32) ? dy : -dy; // ???
         } else {
            if (!(flags & 32)) {
               y = y + (stbtt_int16) (points[0]*256 + points[1]);
               points += 2;
            }
         }
         vertices[off+i].y = (stbtt_int16) y;
      }

      // now convert them to our format
      num_vertices=0;
      sx = sy = cx = cy = scx = scy = 0;
      for (i=0; i < n; ++i) {
         flags = vertices[off+i].type;
         x     = (stbtt_int16) vertices[off+i].x;
         y     = (stbtt_int16) vertices[off+i].y;

         if (next_move == i) {
            if (i != 0)
               num_vertices = stbtt__close_shape(vertices, num_vertices, was_off, start_off, sx,sy,scx,scy,cx,cy);

            // now start the new one
            start_off = !(flags & 1);
            if (start_off) {
               // if we start off with an off-curve point, then when we need to find a point on the curve
               // where we can start, and we need to save some state for when we wraparound.
               scx = x;
               scy = y;
               if (!(vertices[off+i+1].type & 1)) {
                  // next point is also a curve point, so interpolate an on-point curve
                  sx = (x + (stbtt_int32) vertices[off+i+1].x) >> 1;
                  sy = (y + (stbtt_int32) vertices[off+i+1].y) >> 1;
               } else {
                  // otherwise just use the next point as our start point
                  sx = (stbtt_int32) vertices[off+i+1].x;
                  sy = (stbtt_int32) vertices[off+i+1].y;
                  ++i; // we're using point i+1 as the starting point, so skip it
               }
            } else {
               sx = x;
               sy = y;
            }
            stbtt_setvertex(&vertices[num_vertices++], STBTT_vmove,sx,sy,0,0);
            was_off = 0;
            next_move = 1 + ttUSHORT(endPtsOfContours+j*2);
            ++j;
         } else {
            if (!(flags & 1)) { // if it's a curve
               if (was_off) // two off-curve control points in a row means interpolate an on-curve midpoint
                  stbtt_setvertex(&vertices[num_vertices++], STBTT_vcurve, (cx+x)>>1, (cy+y)>>1, cx, cy);
               cx = x;
               cy = y;
               was_off = 1;
            } else {
               if (was_off)
                  stbtt_setvertex(&vertices[num_vertices++], STBTT_vcurve, x,y, cx, cy);
               else
                  stbtt_setvertex(&vertices[num_vertices++], STBTT_vline, x,y,0,0);
               was_off = 0;
            }
         }
      }
      num_vertices = stbtt__close_shape(vertices, num_vertices, was_off, start_off, sx,sy,scx,scy,cx,cy);
   } else if (numberOfContours < 0) {
      // Compound shapes.
      int more = 1;
      stbtt_uint8 *comp = data + g + 10;
      num_vertices = 0;
      vertices = 0;
      while (more) {
         stbtt_uint16 flags, gidx;
         int comp_num_verts = 0, i;
         stbtt_vertex *comp_verts = 0, *tmp = 0;
         float mtx[6] = {1,0,0,1,0,0}, m, n;

         flags = ttSHORT(comp); comp+=2;
         gidx = ttSHORT(comp); comp+=2;

         if (flags & 2) { // XY values
            if (flags & 1) { // shorts
               mtx[4] = ttSHORT(comp); comp+=2;
               mtx[5] = ttSHORT(comp); comp+=2;
            } else {
               mtx[4] = ttCHAR(comp); comp+=1;
               mtx[5] = ttCHAR(comp); comp+=1;
            }
         }
         else {
            // @TODO handle matching point
            STBTT_assert(0);
         }
         if (flags & (1<<3)) { // WE_HAVE_A_SCALE
            mtx[0] = mtx[3] = ttSHORT(comp)/16384.0f; comp+=2;
            mtx[1] = mtx[2] = 0;
         } else if (flags & (1<<6)) { // WE_HAVE_AN_X_AND_YSCALE
            mtx[0] = ttSHORT(comp)/16384.0f; comp+=2;
            mtx[1] = mtx[2] = 0;
            mtx[3] = ttSHORT(comp)/16384.0f; comp+=2;
         } else if (flags & (1<<7)) { // WE_HAVE_A_TWO_BY_TWO
            mtx[0] = ttSHORT(comp)/16384.0f; comp+=2;
            mtx[1] = ttSHORT(comp)/16384.0f; comp+=2;
            mtx[2] = ttSHORT(comp)/16384.0f; comp+=2;
            mtx[3] = ttSHORT(comp)/16384.0f; comp+=2;
         }

         // Find transformation scales.
         m = (float) STBTT_sqrt(mtx[0]*mtx[0] + mtx[1]*mtx[1]);
         n = (float) STBTT_sqrt(mtx[2]*mtx[2] + mtx[3]*mtx[3]);

         // Get indexed glyph.
         comp_num_verts = stbtt_GetGlyphShape(info, gidx, &comp_verts);
         if (comp_num_verts > 0) {
            // Transform vertices.
            for (i = 0; i < comp_num_verts; ++i) {
               stbtt_vertex* v = &comp_verts[i];
               stbtt_vertex_type x,y;
               x=v->x; y=v->y;
               v->x = (stbtt_vertex_type)(m * (mtx[0]*x + mtx[2]*y + mtx[4]));
               v->y = (stbtt_vertex_type)(n * (mtx[1]*x + mtx[3]*y + mtx[5]));
               x=v->cx; y=v->cy;
               v->cx = (stbtt_vertex_type)(m * (mtx[0]*x + mtx[2]*y + mtx[4]));
               v->cy = (stbtt_vertex_type)(n * (mtx[1]*x + mtx[3]*y + mtx[5]));
            }
            // Append vertices.
            tmp = (stbtt_vertex*)STBTT_malloc((num_vertices+comp_num_verts)*sizeof(stbtt_vertex), info->userdata);
            if (!tmp) {
               if (vertices) STBTT_free(vertices, info->userdata);
               if (comp_verts) STBTT_free(comp_verts, info->userdata);
               return 0;
            }
            if (num_vertices > 0 && vertices) STBTT_memcpy(tmp, vertices, num_vertices*sizeof(stbtt_vertex));
            STBTT_memcpy(tmp+num_vertices, comp_verts, comp_num_verts*sizeof(stbtt_vertex));
            if (vertices) STBTT_free(vertices, info->userdata);
            vertices = tmp;
            STBTT_free(comp_verts, info->userdata);
            num_vertices += comp_num_verts;
         }
         // More components ?
         more = flags & (1<<5);
      }
   } else {
      // numberOfCounters == 0, do nothing
   }

   *pvertices = vertices;
   return num_vertices;
}

typedef struct
{
   int bounds;
   int started;
   float first_x, first_y;
   float x, y;
   stbtt_int32 min_x, max_x, min_y, max_y;

   stbtt_vertex *pvertices;
   int num_vertices;
} stbtt__csctx;

#define STBTT__CSCTX_INIT(bounds) {bounds,0, 0,0, 0,0, 0,0,0,0, NULL, 0}

static void stbtt__track_vertex(stbtt__csctx *c, stbtt_int32 x, stbtt_int32 y)
{
   if (x > c->max_x || !c->started) c->max_x = x;
   if (y > c->max_y || !c->started) c->max_y = y;
   if (x < c->min_x || !c->started) c->min_x = x;
   if (y < c->min_y || !c->started) c->min_y = y;
   c->started = 1;
}

static void stbtt__csctx_v(stbtt__csctx *c, stbtt_uint8 type, stbtt_int32 x, stbtt_int32 y, stbtt_int32 cx, stbtt_int32 cy, stbtt_int32 cx1, stbtt_int32 cy1)
{
   if (c->bounds) {
      stbtt__track_vertex(c, x, y);
      if (type == STBTT_vcubic) {
         stbtt__track_vertex(c, cx, cy);
         stbtt__track_vertex(c, cx1, cy1);
      }
   } else {
      stbtt_setvertex(&c->pvertices[c->num_vertices], type, x, y, cx, cy);
      c->pvertices[c->num_vertices].cx1 = (stbtt_int16) cx1;
      c->pvertices[c->num_vertices].cy1 = (stbtt_int16) cy1;
   }
   c->num_vertices++;
}

static void stbtt__csctx_close_shape(stbtt__csctx *ctx)
{
   if (ctx->first_x != ctx->x || ctx->first_y != ctx->y)
      stbtt__csctx_v(ctx, STBTT_vline, (int)ctx->first_x, (int)ctx->first_y, 0, 0, 0, 0);
}

static void stbtt__csctx_rmove_to(stbtt__csctx *ctx, float dx, float dy)
{
   stbtt__csctx_close_shape(ctx);
   ctx->first_x = ctx->x = ctx->x + dx;
   ctx->first_y = ctx->y = ctx->y + dy;
   stbtt__csctx_v(ctx, STBTT_vmove, (int)ctx->x, (int)ctx->y, 0, 0, 0, 0);
}

static void stbtt__csctx_rline_to(stbtt__csctx *ctx, float dx, float dy)
{
   ctx->x += dx;
   ctx->y += dy;
   stbtt__csctx_v(ctx, STBTT_vline, (int)ctx->x, (int)ctx->y, 0, 0, 0, 0);
}

static void stbtt__csctx_rccurve_to(stbtt__csctx *ctx, float dx1, float dy1, float dx2, float dy2, float dx3, float dy3)
{
   float cx1 = ctx->x + dx1;
   float cy1 = ctx->y + dy1;
   float cx2 = cx1 + dx2;
   float cy2 = cy1 + dy2;
   ctx->x = cx2 + dx3;
   ctx->y = cy2 + dy3;
   stbtt__csctx_v(ctx, STBTT_vcubic, (int)ctx->x, (int)ctx->y, (int)cx1, (int)cy1, (int)cx2, (int)cy2);
}

static stbtt__buf stbtt__get_subr(stbtt__buf idx, int n)
{
   int count = stbtt__cff_index_count(&idx);
   int bias = 107;
   if (count >= 33900)
      bias = 32768;
   else if (count >= 1240)
      bias = 1131;
   n += bias;
   if (n < 0 || n >= count)
      return stbtt__new_buf(NULL, 0);
   return stbtt__cff_index_get(idx, n);
}

static stbtt__buf stbtt__cid_get_glyph_subrs(const stbtt_fontinfo *info, int glyph_index)
{
   stbtt__buf fdselect = info->fdselect;
   int nranges, start, end, v, fmt, fdselector = -1, i;

   stbtt__buf_seek(&fdselect, 0);
   fmt = stbtt__buf_get8(&fdselect);
   if (fmt == 0) {
      // untested
      stbtt__buf_skip(&fdselect, glyph_index);
      fdselector = stbtt__buf_get8(&fdselect);
   } else if (fmt == 3) {
      nranges = stbtt__buf_get16(&fdselect);
      start = stbtt__buf_get16(&fdselect);
      for (i = 0; i < nranges; i++) {
         v = stbtt__buf_get8(&fdselect);
         end = stbtt__buf_get16(&fdselect);
         if (glyph_index >= start && glyph_index < end) {
            fdselector = v;
            break;
         }
         start = end;
      }
   }
   if (fdselector == -1) stbtt__new_buf(NULL, 0);
   return stbtt__get_subrs(info->cff, stbtt__cff_index_get(info->fontdicts, fdselector));
}

static int stbtt__run_charstring(const stbtt_fontinfo *info, int glyph_index, stbtt__csctx *c)
{
   int in_header = 1, maskbits = 0, subr_stack_height = 0, sp = 0, v, i, b0;
   int has_subrs = 0, clear_stack;
   float s[48];
   stbtt__buf subr_stack[10], subrs = info->subrs, b;
   float f;

#define STBTT__CSERR(s) (0)

   // this currently ignores the initial width value, which isn't needed if we have hmtx
   b = stbtt__cff_index_get(info->charstrings, glyph_index);
   while (b.cursor < b.size) {
      i = 0;
      clear_stack = 1;
      b0 = stbtt__buf_get8(&b);
      switch (b0) {
      // @TODO implement hinting
      case 0x13: // hintmask
      case 0x14: // cntrmask
         if (in_header)
            maskbits += (sp / 2); // implicit "vstem"
         in_header = 0;
         stbtt__buf_skip(&b, (maskbits + 7) / 8);
         break;

      case 0x01: // hstem
      case 0x03: // vstem
      case 0x12: // hstemhm
      case 0x17: // vstemhm
         maskbits += (sp / 2);
         break;

      case 0x15: // rmoveto
         in_header = 0;
         if (sp < 2) return STBTT__CSERR("rmoveto stack");
         stbtt__csctx_rmove_to(c, s[sp-2], s[sp-1]);
         break;
      case 0x04: // vmoveto
         in_header = 0;
         if (sp < 1) return STBTT__CSERR("vmoveto stack");
         stbtt__csctx_rmove_to(c, 0, s[sp-1]);
         break;
      case 0x16: // hmoveto
         in_header = 0;
         if (sp < 1) return STBTT__CSERR("hmoveto stack");
         stbtt__csctx_rmove_to(c, s[sp-1], 0);
         break;

      case 0x05: // rlineto
         if (sp < 2) return STBTT__CSERR("rlineto stack");
         for (; i + 1 < sp; i += 2)
            stbtt__csctx_rline_to(c, s[i], s[i+1]);
         break;

      // hlineto/vlineto and vhcurveto/hvcurveto alternate horizontal and vertical
      // starting from a different place.

      case 0x07: // vlineto
         if (sp < 1) return STBTT__CSERR("vlineto stack");
         goto vlineto;
      case 0x06: // hlineto
         if (sp < 1) return STBTT__CSERR("hlineto stack");
         for (;;) {
            if (i >= sp) break;
            stbtt__csctx_rline_to(c, s[i], 0);
            i++;
      vlineto:
            if (i >= sp) break;
            stbtt__csctx_rline_to(c, 0, s[i]);
            i++;
         }
         break;

      case 0x1F: // hvcurveto
         if (sp < 4) return STBTT__CSERR("hvcurveto stack");
         goto hvcurveto;
      case 0x1E: // vhcurveto
         if (sp < 4) return STBTT__CSERR("vhcurveto stack");
         for (;;) {
            if (i + 3 >= sp) break;
            stbtt__csctx_rccurve_to(c, 0, s[i], s[i+1], s[i+2], s[i+3], (sp - i == 5) ? s[i + 4] : 0.0f);
            i += 4;
      hvcurveto:
            if (i + 3 >= sp) break;
            stbtt__csctx_rccurve_to(c, s[i], 0, s[i+1], s[i+2], (sp - i == 5) ? s[i+4] : 0.0f, s[i+3]);
            i += 4;
         }
         break;

      case 0x08: // rrcurveto
         if (sp < 6) return STBTT__CSERR("rcurveline stack");
         for (; i + 5 < sp; i += 6)
            stbtt__csctx_rccurve_to(c, s[i], s[i+1], s[i+2], s[i+3], s[i+4], s[i+5]);
         break;

      case 0x18: // rcurveline
         if (sp < 8) return STBTT__CSERR("rcurveline stack");
         for (; i + 5 < sp - 2; i += 6)
            stbtt__csctx_rccurve_to(c, s[i], s[i+1], s[i+2], s[i+3], s[i+4], s[i+5]);
         if (i + 1 >= sp) return STBTT__CSERR("rcurveline stack");
         stbtt__csctx_rline_to(c, s[i], s[i+1]);
         break;

      case 0x19: // rlinecurve
         if (sp < 8) return STBTT__CSERR("rlinecurve stack");
         for (; i + 1 < sp - 6; i += 2)
            stbtt__csctx_rline_to(c, s[i], s[i+1]);
         if (i + 5 >= sp) return STBTT__CSERR("rlinecurve stack");
         stbtt__csctx_rccurve_to(c, s[i], s[i+1], s[i+2], s[i+3], s[i+4], s[i+5]);
         break;

      case 0x1A: // vvcurveto
      case 0x1B: // hhcurveto
         if (sp < 4) return STBTT__CSERR("(vv|hh)curveto stack");
         f = 0.0;
         if (sp & 1) { f = s[i]; i++; }
         for (; i + 3 < sp; i += 4) {
            if (b0 == 0x1B)
               stbtt__csctx_rccurve_to(c, s[i], f, s[i+1], s[i+2], s[i+3], 0.0);
            else
               stbtt__csctx_rccurve_to(c, f, s[i], s[i+1], s[i+2], 0.0, s[i+3]);
            f = 0.0;
         }
         break;

      case 0x0A: // callsubr
         if (!has_subrs) {
            if (info->fdselect.size)
               subrs = stbtt__cid_get_glyph_subrs(info, glyph_index);
            has_subrs = 1;
         }
         // FALLTHROUGH
      case 0x1D: // callgsubr
         if (sp < 1) return STBTT__CSERR("call(g|)subr stack");
         v = (int) s[--sp];
         if (subr_stack_height >= 10) return STBTT__CSERR("recursion limit");
         subr_stack[subr_stack_height++] = b;
         b = stbtt__get_subr(b0 == 0x0A ? subrs : info->gsubrs, v);
         if (b.size == 0) return STBTT__CSERR("subr not found");
         b.cursor = 0;
         clear_stack = 0;
         break;

      case 0x0B: // return
         if (subr_stack_height <= 0) return STBTT__CSERR("return outside subr");
         b = subr_stack[--subr_stack_height];
         clear_stack = 0;
         break;

      case 0x0E: // endchar
         stbtt__csctx_close_shape(c);
         return 1;

      case 0x0C: { // two-byte escape
         float dx1, dx2, dx3, dx4, dx5, dx6, dy1, dy2, dy3, dy4, dy5, dy6;
         float dx, dy;
         int b1 = stbtt__buf_get8(&b);
         switch (b1) {
         // @TODO These "flex" implementations ignore the flex-depth and resolution,
         // and always draw beziers.
         case 0x22: // hflex
            if (sp < 7) return STBTT__CSERR("hflex stack");
            dx1 = s[0];
            dx2 = s[1];
            dy2 = s[2];
            dx3 = s[3];
            dx4 = s[4];
            dx5 = s[5];
            dx6 = s[6];
            stbtt__csctx_rccurve_to(c, dx1, 0, dx2, dy2, dx3, 0);
            stbtt__csctx_rccurve_to(c, dx4, 0, dx5, -dy2, dx6, 0);
            break;

         case 0x23: // flex
            if (sp < 13) return STBTT__CSERR("flex stack");
            dx1 = s[0];
            dy1 = s[1];
            dx2 = s[2];
            dy2 = s[3];
            dx3 = s[4];
            dy3 = s[5];
            dx4 = s[6];
            dy4 = s[7];
            dx5 = s[8];
            dy5 = s[9];
            dx6 = s[10];
            dy6 = s[11];
            //fd is s[12]
            stbtt__csctx_rccurve_to(c, dx1, dy1, dx2, dy2, dx3, dy3);
            stbtt__csctx_rccurve_to(c, dx4, dy4, dx5, dy5, dx6, dy6);
            break;

         case 0x24: // hflex1
            if (sp < 9) return STBTT__CSERR("hflex1 stack");
            dx1 = s[0];
            dy1 = s[1];
            dx2 = s[2];
            dy2 = s[3];
            dx3 = s[4];
            dx4 = s[5];
            dx5 = s[6];
            dy5 = s[7];
            dx6 = s[8];
            stbtt__csctx_rccurve_to(c, dx1, dy1, dx2, dy2, dx3, 0);
            stbtt__csctx_rccurve_to(c, dx4, 0, dx5, dy5, dx6, -(dy1+dy2+dy5));
            break;

         case 0x25: // flex1
            if (sp < 11) return STBTT__CSERR("flex1 stack");
            dx1 = s[0];
            dy1 = s[1];
            dx2 = s[2];
            dy2 = s[3];
            dx3 = s[4];
            dy3 = s[5];
            dx4 = s[6];
            dy4 = s[7];
            dx5 = s[8];
            dy5 = s[9];
            dx6 = dy6 = s[10];
            dx = dx1+dx2+dx3+dx4+dx5;
            dy = dy1+dy2+dy3+dy4+dy5;
            if (STBTT_fabs(dx) > STBTT_fabs(dy))
               dy6 = -dy;
            else
               dx6 = -dx;
            stbtt__csctx_rccurve_to(c, dx1, dy1, dx2, dy2, dx3, dy3);
            stbtt__csctx_rccurve_to(c, dx4, dy4, dx5, dy5, dx6, dy6);
            break;

         default:
            return STBTT__CSERR("unimplemented");
         }
      } break;

      default:
         if (b0 != 255 && b0 != 28 && b0 < 32)
            return STBTT__CSERR("reserved operator");

         // push immediate
         if (b0 == 255) {
            f = (float)(stbtt_int32)stbtt__buf_get32(&b) / 0x10000;
         } else {
            stbtt__buf_skip(&b, -1);
            f = (float)(stbtt_int16)stbtt__cff_int(&b);
         }
         if (sp >= 48) return STBTT__CSERR("push stack overflow");
         s[sp++] = f;
         clear_stack = 0;
         break;
      }
      if (clear_stack) sp = 0;
   }
   return STBTT__CSERR("no endchar");

#undef STBTT__CSERR
}

static int stbtt__GetGlyphShapeT2(const stbtt_fontinfo *info, int glyph_index, stbtt_vertex **pvertices)
{
   // runs the charstring twice, once to count and once to output (to avoid realloc)
   stbtt__csctx count_ctx = STBTT__CSCTX_INIT(1);
   stbtt__csctx output_ctx = STBTT__CSCTX_INIT(0);
   if (stbtt__run_charstring(info, glyph_index, &count_ctx)) {
      *pvertices = (stbtt_vertex*)STBTT_malloc(count_ctx.num_vertices*sizeof(stbtt_vertex), info->userdata);
      output_ctx.pvertices = *pvertices;
      if (stbtt__run_charstring(info, glyph_index, &output_ctx)) {
         STBTT_assert(output_ctx.num_vertices == count_ctx.num_vertices);
         return output_ctx.num_vertices;
      }
   }
   *pvertices = NULL;
   return 0;
}

static int stbtt__GetGlyphInfoT2(const stbtt_fontinfo *info, int glyph_index, int *x0, int *y0, int *x1, int *y1)
{
   stbtt__csctx c = STBTT__CSCTX_INIT(1);
   int r = stbtt__run_charstring(info, glyph_index, &c);
   if (x0)  *x0 = r ? c.min_x : 0;
   if (y0)  *y0 = r ? c.min_y : 0;
   if (x1)  *x1 = r ? c.max_x : 0;
   if (y1)  *y1 = r ? c.max_y : 0;
   return r ? c.num_vertices : 0;
}

STBTT_DEF int stbtt_GetGlyphShape(const stbtt_fontinfo *info, int glyph_index, stbtt_vertex **pvertices)
{
   if (!info->cff.size)
      return stbtt__GetGlyphShapeTT(info, glyph_index, pvertices);
   else
      return stbtt__GetGlyphShapeT2(info, glyph_index, pvertices);
}

STBTT_DEF void stbtt_GetGlyphHMetrics(const stbtt_fontinfo *info, int glyph_index, int *advanceWidth, int *leftSideBearing)
{
   stbtt_uint16 numOfLongHorMetrics = ttUSHORT(info->data+info->hhea + 34);
   if (glyph_index < numOfLongHorMetrics) {
      if (advanceWidth)     *advanceWidth    = ttSHORT(info->data + info->hmtx + 4*glyph_index);
      if (leftSideBearing)  *leftSideBearing = ttSHORT(info->data + info->hmtx + 4*glyph_index + 2);
   } else {
      if (advanceWidth)     *advanceWidth    = ttSHORT(info->data + info->hmtx + 4*(numOfLongHorMetrics-1));
      if (leftSideBearing)  *leftSideBearing = ttSHORT(info->data + info->hmtx + 4*numOfLongHorMetrics + 2*(glyph_index - numOfLongHorMetrics));
   }
}

STBTT_DEF int  stbtt_GetKerningTableLength(const stbtt_fontinfo *info)
{
   stbtt_uint8 *data = info->data + info->kern;

   // we only look at the first table. it must be 'horizontal' and format 0.
   if (!info->kern)
      return 0;
   if (ttUSHORT(data+2) < 1) // number of tables, need at least 1
      return 0;
   if (ttUSHORT(data+8) != 1) // horizontal flag must be set in format
      return 0;

   return ttUSHORT(data+10);
}

STBTT_DEF int stbtt_GetKerningTable(const stbtt_fontinfo *info, stbtt_kerningentry* table, int table_length)
{
   stbtt_uint8 *data = info->data + info->kern;
   int k, length;

   // we only look at the first table. it must be 'horizontal' and format 0.
   if (!info->kern)
      return 0;
   if (ttUSHORT(data+2) < 1) // number of tables, need at least 1
      return 0;
   if (ttUSHORT(data+8) != 1) // horizontal flag must be set in format
      return 0;

   length = ttUSHORT(data+10);
   if (table_length < length)
      length = table_length;

   for (k = 0; k < length; k++)
   {
      table[k].glyph1 = ttUSHORT(data+18+(k*6));
      table[k].glyph2 = ttUSHORT(data+20+(k*6));
      table[k].advance = ttSHORT(data+22+(k*6));
   }

   return length;
}

static int stbtt__GetGlyphKernInfoAdvance(const stbtt_fontinfo *info, int glyph1, int glyph2)
{
   stbtt_uint8 *data = info->data + info->kern;
   stbtt_uint32 needle, straw;
   int l, r, m;

   // we only look at the first table. it must be 'horizontal' and format 0.
   if (!info->kern)
      return 0;
   if (ttUSHORT(data+2) < 1) // number of tables, need at least 1
      return 0;
   if (ttUSHORT(data+8) != 1) // horizontal flag must be set in format
      return 0;

   l = 0;
   r = ttUSHORT(data+10) - 1;
   needle = glyph1 << 16 | glyph2;
   while (l <= r) {
      m = (l + r) >> 1;
      straw = ttULONG(data+18+(m*6)); // note: unaligned read
      if (needle < straw)
         r = m - 1;
      else if (needle > straw)
         l = m + 1;
      else
         return ttSHORT(data+22+(m*6));
   }
   return 0;
}

static stbtt_int32 stbtt__GetCoverageIndex(stbtt_uint8 *coverageTable, int glyph)
{
   stbtt_uint16 coverageFormat = ttUSHORT(coverageTable);
   switch (coverageFormat) {
      case 1: {
         stbtt_uint16 glyphCount = ttUSHORT(coverageTable + 2);

         // Binary search.
         stbtt_int32 l=0, r=glyphCount-1, m;
         int straw, needle=glyph;
         while (l <= r) {
            stbtt_uint8 *glyphArray = coverageTable + 4;
            stbtt_uint16 glyphID;
            m = (l + r) >> 1;
            glyphID = ttUSHORT(glyphArray + 2 * m);
            straw = glyphID;
            if (needle < straw)
               r = m - 1;
            else if (needle > straw)
               l = m + 1;
            else {
               return m;
            }
         }
         break;
      }

      case 2: {
         stbtt_uint16 rangeCount = ttUSHORT(coverageTable + 2);
         stbtt_uint8 *rangeArray = coverageTable + 4;

         // Binary search.
         stbtt_int32 l=0, r=rangeCount-1, m;
         int strawStart, strawEnd, needle=glyph;
         while (l <= r) {
            stbtt_uint8 *rangeRecord;
            m = (l + r) >> 1;
            rangeRecord = rangeArray + 6 * m;
            strawStart = ttUSHORT(rangeRecord);
            strawEnd = ttUSHORT(rangeRecord + 2);
            if (needle < strawStart)
               r = m - 1;
            else if (needle > strawEnd)
               l = m + 1;
            else {
               stbtt_uint16 startCoverageIndex = ttUSHORT(rangeRecord + 4);
               return startCoverageIndex + glyph - strawStart;
            }
         }
         break;
      }

      default: return -1; // unsupported
   }

   return -1;
}

static stbtt_int32  stbtt__GetGlyphClass(stbtt_uint8 *classDefTable, int glyph)
{
   stbtt_uint16 classDefFormat = ttUSHORT(classDefTable);
   switch (classDefFormat)
   {
      case 1: {
         stbtt_uint16 startGlyphID = ttUSHORT(classDefTable + 2);
         stbtt_uint16 glyphCount = ttUSHORT(classDefTable + 4);
         stbtt_uint8 *classDef1ValueArray = classDefTable + 6;

         if (glyph >= startGlyphID && glyph < startGlyphID + glyphCount)
            return (stbtt_int32)ttUSHORT(classDef1ValueArray + 2 * (glyph - startGlyphID));
         break;
      }

      case 2: {
         stbtt_uint16 classRangeCount = ttUSHORT(classDefTable + 2);
         stbtt_uint8 *classRangeRecords = classDefTable + 4;

         // Binary search.
         stbtt_int32 l=0, r=classRangeCount-1, m;
         int strawStart, strawEnd, needle=glyph;
         while (l <= r) {
            stbtt_uint8 *classRangeRecord;
            m = (l + r) >> 1;
            classRangeRecord = classRangeRecords + 6 * m;
            strawStart = ttUSHORT(classRangeRecord);
            strawEnd = ttUSHORT(classRangeRecord + 2);
            if (needle < strawStart)
               r = m - 1;
            else if (needle > strawEnd)
               l = m + 1;
            else
               return (stbtt_int32)ttUSHORT(classRangeRecord + 4);
         }
         break;
      }

      default:
         return -1; // Unsupported definition type, return an error.
   }

   // "All glyphs not assigned to a class fall into class 0". (OpenType spec)
   return 0;
}

// Define to STBTT_assert(x) if you want to break on unimplemented formats.
#define STBTT_GPOS_TODO_assert(x)

static stbtt_int32 stbtt__GetGlyphGPOSInfoAdvance(const stbtt_fontinfo *info, int glyph1, int glyph2)
{
   stbtt_uint16 lookupListOffset;
   stbtt_uint8 *lookupList;
   stbtt_uint16 lookupCount;
   stbtt_uint8 *data;
   stbtt_int32 i, sti;

   if (!info->gpos) return 0;

   data = info->data + info->gpos;

   if (ttUSHORT(data+0) != 1) return 0; // Major version 1
   if (ttUSHORT(data+2) != 0) return 0; // Minor version 0

   lookupListOffset = ttUSHORT(data+8);
   lookupList = data + lookupListOffset;
   lookupCount = ttUSHORT(lookupList);

   for (i=0; i<lookupCount; ++i) {
      stbtt_uint16 lookupOffset = ttUSHORT(lookupList + 2 + 2 * i);
      stbtt_uint8 *lookupTable = lookupList + lookupOffset;

      stbtt_uint16 lookupType = ttUSHORT(lookupTable);
      stbtt_uint16 subTableCount = ttUSHORT(lookupTable + 4);
      stbtt_uint8 *subTableOffsets = lookupTable + 6;
      if (lookupType != 2) // Pair Adjustment Positioning Subtable
         continue;

      for (sti=0; sti<subTableCount; sti++) {
         stbtt_uint16 subtableOffset = ttUSHORT(subTableOffsets + 2 * sti);
         stbtt_uint8 *table = lookupTable + subtableOffset;
         stbtt_uint16 posFormat = ttUSHORT(table);
         stbtt_uint16 coverageOffset = ttUSHORT(table + 2);
         stbtt_int32 coverageIndex = stbtt__GetCoverageIndex(table + coverageOffset, glyph1);
         if (coverageIndex == -1) continue;

         switch (posFormat) {
            case 1: {
               stbtt_int32 l, r, m;
               int straw, needle;
               stbtt_uint16 valueFormat1 = ttUSHORT(table + 4);
               stbtt_uint16 valueFormat2 = ttUSHORT(table + 6);
               if (valueFormat1 == 4 && valueFormat2 == 0) { // Support more formats?
                  stbtt_int32 valueRecordPairSizeInBytes = 2;
                  stbtt_uint16 pairSetCount = ttUSHORT(table + 8);
                  stbtt_uint16 pairPosOffset = ttUSHORT(table + 10 + 2 * coverageIndex);
                  stbtt_uint8 *pairValueTable = table + pairPosOffset;
                  stbtt_uint16 pairValueCount = ttUSHORT(pairValueTable);
                  stbtt_uint8 *pairValueArray = pairValueTable + 2;

                  if (coverageIndex >= pairSetCount) return 0;

                  needle=glyph2;
                  r=pairValueCount-1;
                  l=0;

                  // Binary search.
                  while (l <= r) {
                     stbtt_uint16 secondGlyph;
                     stbtt_uint8 *pairValue;
                     m = (l + r) >> 1;
                     pairValue = pairValueArray + (2 + valueRecordPairSizeInBytes) * m;
                     secondGlyph = ttUSHORT(pairValue);
                     straw = secondGlyph;
                     if (needle < straw)
                        r = m - 1;
                     else if (needle > straw)
                        l = m + 1;
                     else {
                        stbtt_int16 xAdvance = ttSHORT(pairValue + 2);
                        return xAdvance;
                     }
                  }
               } else
                  return 0;
               break;
            }

            case 2: {
               stbtt_uint16 valueFormat1 = ttUSHORT(table + 4);
               stbtt_uint16 valueFormat2 = ttUSHORT(table + 6);
               if (valueFormat1 == 4 && valueFormat2 == 0) { // Support more formats?
                  stbtt_uint16 classDef1Offset = ttUSHORT(table + 8);
                  stbtt_uint16 classDef2Offset = ttUSHORT(table + 10);
                  int glyph1class = stbtt__GetGlyphClass(table + classDef1Offset, glyph1);
                  int glyph2class = stbtt__GetGlyphClass(table + classDef2Offset, glyph2);

                  stbtt_uint16 class1Count = ttUSHORT(table + 12);
                  stbtt_uint16 class2Count = ttUSHORT(table + 14);
                  stbtt_uint8 *class1Records, *class2Records;
                  stbtt_int16 xAdvance;

                  if (glyph1class < 0 || glyph1class >= class1Count) return 0; // malformed
                  if (glyph2class < 0 || glyph2class >= class2Count) return 0; // malformed

                  class1Records = table + 16;
                  class2Records = class1Records + 2 * (glyph1class * class2Count);
                  xAdvance = ttSHORT(class2Records + 2 * glyph2class);
                  return xAdvance;
               } else
                  return 0;
               break;
            }

            default:
               return 0; // Unsupported position format
         }
      }
   }

   return 0;
}

STBTT_DEF int  stbtt_GetGlyphKernAdvance(const stbtt_fontinfo *info, int g1, int g2)
{
   int xAdvance = 0;

   if (info->gpos)
      xAdvance += stbtt__GetGlyphGPOSInfoAdvance(info, g1, g2);
   else if (info->kern)
      xAdvance += stbtt__GetGlyphKernInfoAdvance(info, g1, g2);

   return xAdvance;
}

STBTT_DEF int  stbtt_GetCodepointKernAdvance(const stbtt_fontinfo *info, int ch1, int ch2)
{
   if (!info->kern && !info->gpos) // if no kerning table, don't waste time looking up both codepoint->glyphs
      return 0;
   return stbtt_GetGlyphKernAdvance(info, stbtt_FindGlyphIndex(info,ch1), stbtt_FindGlyphIndex(info,ch2));
}

STBTT_DEF void stbtt_GetCodepointHMetrics(const stbtt_fontinfo *info, int codepoint, int *advanceWidth, int *leftSideBearing)
{
   stbtt_GetGlyphHMetrics(info, stbtt_FindGlyphIndex(info,codepoint), advanceWidth, leftSideBearing);
}

STBTT_DEF void stbtt_GetFontVMetrics(const stbtt_fontinfo *info, int *ascent, int *descent, int *lineGap)
{
   if (ascent ) *ascent  = ttSHORT(info->data+info->hhea + 4);
   if (descent) *descent = ttSHORT(info->data+info->hhea + 6);
   if (lineGap) *lineGap = ttSHORT(info->data+info->hhea + 8);
}

STBTT_DEF int  stbtt_GetFontVMetricsOS2(const stbtt_fontinfo *info, int *typoAscent, int *typoDescent, int *typoLineGap)
{
   int tab = stbtt__find_table(info->data, info->fontstart, "OS/2");
   if (!tab)
      return 0;
   if (typoAscent ) *typoAscent  = ttSHORT(info->data+tab + 68);
   if (typoDescent) *typoDescent = ttSHORT(info->data+tab + 70);
   if (typoLineGap) *typoLineGap = ttSHORT(info->data+tab + 72);
   return 1;
}

STBTT_DEF void stbtt_GetFontBoundingBox(const stbtt_fontinfo *info, int *x0, int *y0, int *x1, int *y1)
{
   *x0 = ttSHORT(info->data + info->head + 36);
   *y0 = ttSHORT(info->data + info->head + 38);
   *x1 = ttSHORT(info->data + info->head + 40);
   *y1 = ttSHORT(info->data + info->head + 42);
}

STBTT_DEF float stbtt_ScaleForPixelHeight(const stbtt_fontinfo *info, float height)
{
   int fheight = ttSHORT(info->data + info->hhea + 4) - ttSHORT(info->data + info->hhea + 6);
   return (float) height / fheight;
}

STBTT_DEF float stbtt_ScaleForMappingEmToPixels(const stbtt_fontinfo *info, float pixels)
{
   int unitsPerEm = ttUSHORT(info->data + info->head + 18);
   return pixels / unitsPerEm;
}

STBTT_DEF void stbtt_FreeShape(const stbtt_fontinfo *info, stbtt_vertex *v)
{
   STBTT_free(v, info->userdata);
}

STBTT_DEF stbtt_uint8 *stbtt_FindSVGDoc(const stbtt_fontinfo *info, int gl)
{
   int i;
   stbtt_uint8 *data = info->data;
   stbtt_uint8 *svg_doc_list = data + stbtt__get_svg((stbtt_fontinfo *) info);

   int numEntries = ttUSHORT(svg_doc_list);
   stbtt_uint8 *svg_docs = svg_doc_list + 2;

   for(i=0; i<numEntries; i++) {
      stbtt_uint8 *svg_doc = svg_docs + (12 * i);
      if ((gl >= ttUSHORT(svg_doc)) && (gl <= ttUSHORT(svg_doc + 2)))
         return svg_doc;
   }
   return 0;
}

STBTT_DEF int stbtt_GetGlyphSVG(const stbtt_fontinfo *info, int gl, const char **svg)
{
   stbtt_uint8 *data = info->data;
   stbtt_uint8 *svg_doc;

   if (info->svg == 0)
      return 0;

   svg_doc = stbtt_FindSVGDoc(info, gl);
   if (svg_doc != NULL) {
      *svg = (char *) data + info->svg + ttULONG(svg_doc + 4);
      return ttULONG(svg_doc + 8);
   } else {
      return 0;
   }
}

STBTT_DEF int stbtt_GetCodepointSVG(const stbtt_fontinfo *info, int unicode_codepoint, const char **svg)
{
   return stbtt_GetGlyphSVG(info, stbtt_FindGlyphIndex(info, unicode_codepoint), svg);
}

//////////////////////////////////////////////////////////////////////////////
//
// antialiasing software rasterizer
//

STBTT_DEF void stbtt_GetGlyphBitmapBoxSubpixel(const stbtt_fontinfo *font, int glyph, float scale_x, float scale_y,float shift_x, float shift_y, int *ix0, int *iy0, int *ix1, int *iy1)
{
   int x0=0,y0=0,x1,y1; // =0 suppresses compiler warning
   if (!stbtt_GetGlyphBox(font, glyph, &x0,&y0,&x1,&y1)) {
      // e.g. space character
      if (ix0) *ix0 = 0;
      if (iy0) *iy0 = 0;
      if (ix1) *ix1 = 0;
      if (iy1) *iy1 = 0;
   } else {
      // move to integral bboxes (treating pixels as little squares, what pixels get touched)?
      if (ix0) *ix0 = STBTT_ifloor( x0 * scale_x + shift_x);
      if (iy0) *iy0 = STBTT_ifloor(-y1 * scale_y + shift_y);
      if (ix1) *ix1 = STBTT_iceil ( x1 * scale_x + shift_x);
      if (iy1) *iy1 = STBTT_iceil (-y0 * scale_y + shift_y);
   }
}

STBTT_DEF void stbtt_GetGlyphBitmapBox(const stbtt_fontinfo *font, int glyph, float scale_x, float scale_y, int *ix0, int *iy0, int *ix1, int *iy1)
{
   stbtt_GetGlyphBitmapBoxSubpixel(font, glyph, scale_x, scale_y,0.0f,0.0f, ix0, iy0, ix1, iy1);
}

STBTT_DEF void stbtt_GetCodepointBitmapBoxSubpixel(const stbtt_fontinfo *font, int codepoint, float scale_x, float scale_y, float shift_x, float shift_y, int *ix0, int *iy0, int *ix1, int *iy1)
{
   stbtt_GetGlyphBitmapBoxSubpixel(font, stbtt_FindGlyphIndex(font,codepoint), scale_x, scale_y,shift_x,shift_y, ix0,iy0,ix1,iy1);
}

STBTT_DEF void stbtt_GetCodepointBitmapBox(const stbtt_fontinfo *font, int codepoint, float scale_x, float scale_y, int *ix0, int *iy0, int *ix1, int *iy1)
{
   stbtt_GetCodepointBitmapBoxSubpixel(font, codepoint, scale_x, scale_y,0.0f,0.0f, ix0,iy0,ix1,iy1);
}

//////////////////////////////////////////////////////////////////////////////
//
//  Rasterizer

typedef struct stbtt__hheap_chunk
{
   struct stbtt__hheap_chunk *next;
} stbtt__hheap_chunk;

typedef struct stbtt__hheap
{
   struct stbtt__hheap_chunk *head;
   void   *first_free;
   int    num_remaining_in_head_chunk;
} stbtt__hheap;

static void *stbtt__hheap_alloc(stbtt__hheap *hh, size_t size, void *userdata)
{
   if (hh->first_free) {
      void *p = hh->first_free;
      hh->first_free = * (void **) p;
      return p;
   } else {
      if (hh->num_remaining_in_head_chunk == 0) {
         int count = (size < 32 ? 2000 : size < 128 ? 800 : 100);
         stbtt__hheap_chunk *c = (stbtt__hheap_chunk *) STBTT_malloc(sizeof(stbtt__hheap_chunk) + size * count, userdata);
         if (c == NULL)
            return NULL;
         c->next = hh->head;
         hh->head = c;
         hh->num_remaining_in_head_chunk = count;
      }
      --hh->num_remaining_in_head_chunk;
      return (char *) (hh->head) + sizeof(stbtt__hheap_chunk) + size * hh->num_remaining_in_head_chunk;
   }
}

static void stbtt__hheap_free(stbtt__hheap *hh, void *p)
{
   *(void **) p = hh->first_free;
   hh->first_free = p;
}

static void stbtt__hheap_cleanup(stbtt__hheap *hh, void *userdata)
{
   stbtt__hheap_chunk *c = hh->head;
   while (c) {
      stbtt__hheap_chunk *n = c->next;
      STBTT_free(c, userdata);
      c = n;
   }
}

typedef struct stbtt__edge {
   float x0,y0, x1,y1;
   int invert;
} stbtt__edge;


typedef struct stbtt__active_edge
{
   struct stbtt__active_edge *next;
   #if STBTT_RASTERIZER_VERSION==1
   int x,dx;
   float ey;
   int direction;
   #elif STBTT_RASTERIZER_VERSION==2
   float fx,fdx,fdy;
   float direction;
   float sy;
   float ey;
   #else
   #error "Unrecognized value of STBTT_RASTERIZER_VERSION"
   #endif
} stbtt__active_edge;

#if STBTT_RASTERIZER_VERSION == 1
#define STBTT_FIXSHIFT   10
#define STBTT_FIX        (1 << STBTT_FIXSHIFT)
#define STBTT_FIXMASK    (STBTT_FIX-1)

static stbtt__active_edge *stbtt__new_active(stbtt__hheap *hh, stbtt__edge *e, int off_x, float start_point, void *userdata)
{
   stbtt__active_edge *z = (stbtt__active_edge *) stbtt__hheap_alloc(hh, sizeof(*z), userdata);
   float dxdy = (e->x1 - e->x0) / (e->y1 - e->y0);
   STBTT_assert(z != NULL);
   if (!z) return z;

   // round dx down to avoid overshooting
   if (dxdy < 0)
      z->dx = -STBTT_ifloor(STBTT_FIX * -dxdy);
   else
      z->dx = STBTT_ifloor(STBTT_FIX * dxdy);

   z->x = STBTT_ifloor(STBTT_FIX * e->x0 + z->dx * (start_point - e->y0)); // use z->dx so when we offset later it's by the same amount
   z->x -= off_x * STBTT_FIX;

   z->ey = e->y1;
   z->next = 0;
   z->direction = e->invert ? 1 : -1;
   return z;
}
#elif STBTT_RASTERIZER_VERSION == 2
static stbtt__active_edge *stbtt__new_active(stbtt__hheap *hh, stbtt__edge *e, int off_x, float start_point, void *userdata)
{
   stbtt__active_edge *z = (stbtt__active_edge *) stbtt__hheap_alloc(hh, sizeof(*z), userdata);
   float dxdy = (e->x1 - e->x0) / (e->y1 - e->y0);
   STBTT_assert(z != NULL);
   //STBTT_assert(e->y0 <= start_point);
   if (!z) return z;
   z->fdx = dxdy;
   z->fdy = dxdy != 0.0f ? (1.0f/dxdy) : 0.0f;
   z->fx = e->x0 + dxdy * (start_point - e->y0);
   z->fx -= off_x;
   z->direction = e->invert ? 1.0f : -1.0f;
   z->sy = e->y0;
   z->ey = e->y1;
   z->next = 0;
   return z;
}
#else
#error "Unrecognized value of STBTT_RASTERIZER_VERSION"
#endif

#if STBTT_RASTERIZER_VERSION == 1
// note: this routine clips fills that extend off the edges... ideally this
// wouldn't happen, but it could happen if the truetype glyph bounding boxes
// are wrong, or if the user supplies a too-small bitmap
static void stbtt__fill_active_edges(unsigned char *scanline, int len, stbtt__active_edge *e, int max_weight)
{
   // non-zero winding fill
   int x0=0, w=0;

   while (e) {
      if (w == 0) {
         // if we're currently at zero, we need to record the edge start point
         x0 = e->x; w += e->direction;
      } else {
         int x1 = e->x; w += e->direction;
         // if we went to zero, we need to draw
         if (w == 0) {
            int i = x0 >> STBTT_FIXSHIFT;
            int j = x1 >> STBTT_FIXSHIFT;

            if (i < len && j >= 0) {
               if (i == j) {
                  // x0,x1 are the same pixel, so compute combined coverage
                  scanline[i] = scanline[i] + (stbtt_uint8) ((x1 - x0) * max_weight >> STBTT_FIXSHIFT);
               } else {
                  if (i >= 0) // add antialiasing for x0
                     scanline[i] = scanline[i] + (stbtt_uint8) (((STBTT_FIX - (x0 & STBTT_FIXMASK)) * max_weight) >> STBTT_FIXSHIFT);
                  else
                     i = -1; // clip

                  if (j < len) // add antialiasing for x1
                     scanline[j] = scanline[j] + (stbtt_uint8) (((x1 & STBTT_FIXMASK) * max_weight) >> STBTT_FIXSHIFT);
                  else
                     j = len; // clip

                  for (++i; i < j; ++i) // fill pixels between x0 and x1
                     scanline[i] = scanline[i] + (stbtt_uint8) max_weight;
               }
            }
         }
      }

      e = e->next;
   }
}

static void stbtt__rasterize_sorted_edges(stbtt__bitmap *result, stbtt__edge *e, int n, int vsubsample, int off_x, int off_y, void *userdata)
{
   stbtt__hheap hh = { 0, 0, 0 };
   stbtt__active_edge *active = NULL;
   int y,j=0;
   int max_weight = (255 / vsubsample);  // weight per vertical scanline
   int s; // vertical subsample index
   unsigned char scanline_data[512], *scanline;

   if (result->w > 512)
      scanline = (unsigned char *) STBTT_malloc(result->w, userdata);
   else
      scanline = scanline_data;

   y = off_y * vsubsample;
   e[n].y0 = (off_y + result->h) * (float) vsubsample + 1;

   while (j < result->h) {
      STBTT_memset(scanline, 0, result->w);
      for (s=0; s < vsubsample; ++s) {
         // find center of pixel for this scanline
         float scan_y = y + 0.5f;
         stbtt__active_edge **step = &active;

         // update all active edges;
         // remove all active edges that terminate before the center of this scanline
         while (*step) {
            stbtt__active_edge * z = *step;
            if (z->ey <= scan_y) {
               *step = z->next; // delete from list
               STBTT_assert(z->direction);
               z->direction = 0;
               stbtt__hheap_free(&hh, z);
            } else {
               z->x += z->dx; // advance to position for current scanline
               step = &((*step)->next); // advance through list
            }
         }

         // resort the list if needed
         for(;;) {
            int changed=0;
            step = &active;
            while (*step && (*step)->next) {
               if ((*step)->x > (*step)->next->x) {
                  stbtt__active_edge *t = *step;
                  stbtt__active_edge *q = t->next;

                  t->next = q->next;
                  q->next = t;
                  *step = q;
                  changed = 1;
               }
               step = &(*step)->next;
            }
            if (!changed) break;
         }

         // insert all edges that start before the center of this scanline -- omit ones that also end on this scanline
         while (e->y0 <= scan_y) {
            if (e->y1 > scan_y) {
               stbtt__active_edge *z = stbtt__new_active(&hh, e, off_x, scan_y, userdata);
               if (z != NULL) {
                  // find insertion point
                  if (active == NULL)
                     active = z;
                  else if (z->x < active->x) {
                     // insert at front
                     z->next = active;
                     active = z;
                  } else {
                     // find thing to insert AFTER
                     stbtt__active_edge *p = active;
                     while (p->next && p->next->x < z->x)
                        p = p->next;
                     // at this point, p->next->x is NOT < z->x
                     z->next = p->next;
                     p->next = z;
                  }
               }
            }
            ++e;
         }

         // now process all active edges in XOR fashion
         if (active)
            stbtt__fill_active_edges(scanline, result->w, active, max_weight);

         ++y;
      }
      STBTT_memcpy(result->pixels + j * result->stride, scanline, result->w);
      ++j;
   }

   stbtt__hheap_cleanup(&hh, userdata);

   if (scanline != scanline_data)
      STBTT_free(scanline, userdata);
}

#elif STBTT_RASTERIZER_VERSION == 2

// the edge passed in here does not cross the vertical line at x or the vertical line at x+1
// (i.e. it has already been clipped to those)
static void stbtt__handle_clipped_edge(float *scanline, int x, stbtt__active_edge *e, float x0, float y0, float x1, float y1)
{
   if (y0 == y1) return;
   STBTT_assert(y0 < y1);
   STBTT_assert(e->sy <= e->ey);
   if (y0 > e->ey) return;
   if (y1 < e->sy) return;
   if (y0 < e->sy) {
      x0 += (x1-x0) * (e->sy - y0) / (y1-y0);
      y0 = e->sy;
   }
   if (y1 > e->ey) {
      x1 += (x1-x0) * (e->ey - y1) / (y1-y0);
      y1 = e->ey;
   }

   if (x0 == x)
      STBTT_assert(x1 <= x+1);
   else if (x0 == x+1)
      STBTT_assert(x1 >= x);
   else if (x0 <= x)
      STBTT_assert(x1 <= x);
   else if (x0 >= x+1)
      STBTT_assert(x1 >= x+1);
   else
      STBTT_assert(x1 >= x && x1 <= x+1);

   if (x0 <= x && x1 <= x)
      scanline[x] += e->direction * (y1-y0);
   else if (x0 >= x+1 && x1 >= x+1)
      ;
   else {
      STBTT_assert(x0 >= x && x0 <= x+1 && x1 >= x && x1 <= x+1);
      scanline[x] += e->direction * (y1-y0) * (1-((x0-x)+(x1-x))/2); // coverage = 1 - average x position
   }
}

static float stbtt__sized_trapezoid_area(float height, float top_width, float bottom_width)
{
   STBTT_assert(top_width >= 0);
   STBTT_assert(bottom_width >= 0);
   return (top_width + bottom_width) / 2.0f * height;
}

static float stbtt__position_trapezoid_area(float height, float tx0, float tx1, float bx0, float bx1)
{
   return stbtt__sized_trapezoid_area(height, tx1 - tx0, bx1 - bx0);
}

static float stbtt__sized_triangle_area(float height, float width)
{
   return height * width / 2;
}

static void stbtt__fill_active_edges_new(float *scanline, float *scanline_fill, int len, stbtt__active_edge *e, float y_top)
{
   float y_bottom = y_top+1;

   while (e) {
      // brute force every pixel

      // compute intersection points with top & bottom
      STBTT_assert(e->ey >= y_top);

      if (e->fdx == 0) {
         float x0 = e->fx;
         if (x0 < len) {
            if (x0 >= 0) {
               stbtt__handle_clipped_edge(scanline,(int) x0,e, x0,y_top, x0,y_bottom);
               stbtt__handle_clipped_edge(scanline_fill-1,(int) x0+1,e, x0,y_top, x0,y_bottom);
            } else {
               stbtt__handle_clipped_edge(scanline_fill-1,0,e, x0,y_top, x0,y_bottom);
            }
         }
      } else {
         float x0 = e->fx;
         float dx = e->fdx;
         float xb = x0 + dx;
         float x_top, x_bottom;
         float sy0,sy1;
         float dy = e->fdy;
         STBTT_assert(e->sy <= y_bottom && e->ey >= y_top);

         // compute endpoints of line segment clipped to this scanline (if the
         // line segment starts on this scanline. x0 is the intersection of the
         // line with y_top, but that may be off the line segment.
         if (e->sy > y_top) {
            x_top = x0 + dx * (e->sy - y_top);
            sy0 = e->sy;
         } else {
            x_top = x0;
            sy0 = y_top;
         }
         if (e->ey < y_bottom) {
            x_bottom = x0 + dx * (e->ey - y_top);
            sy1 = e->ey;
         } else {
            x_bottom = xb;
            sy1 = y_bottom;
         }

         if (x_top >= 0 && x_bottom >= 0 && x_top < len && x_bottom < len) {
            // from here on, we don't have to range check x values

            if ((int) x_top == (int) x_bottom) {
               float height;
               // simple case, only spans one pixel
               int x = (int) x_top;
               height = (sy1 - sy0) * e->direction;
               STBTT_assert(x >= 0 && x < len);
               scanline[x]      += stbtt__position_trapezoid_area(height, x_top, x+1.0f, x_bottom, x+1.0f);
               scanline_fill[x] += height; // everything right of this pixel is filled
            } else {
               int x,x1,x2;
               float y_crossing, y_final, step, sign, area;
               // covers 2+ pixels
               if (x_top > x_bottom) {
                  // flip scanline vertically; signed area is the same
                  float t;
                  sy0 = y_bottom - (sy0 - y_top);
                  sy1 = y_bottom - (sy1 - y_top);
                  t = sy0, sy0 = sy1, sy1 = t;
                  t = x_bottom, x_bottom = x_top, x_top = t;
                  dx = -dx;
                  dy = -dy;
                  t = x0, x0 = xb, xb = t;
               }
               STBTT_assert(dy >= 0);
               STBTT_assert(dx >= 0);

               x1 = (int) x_top;
               x2 = (int) x_bottom;
               // compute intersection with y axis at x1+1
               y_crossing = y_top + dy * (x1+1 - x0);

               // compute intersection with y axis at x2
               y_final = y_top + dy * (x2 - x0);

               //           x1    x_top                            x2    x_bottom
               //     y_top  +------|-----+------------+------------+--------|---+------------+
               //            |            |            |            |            |            |
               //            |            |            |            |            |            |
               //       sy0  |      Txxxxx|............|............|............|............|
               // y_crossing |            *xxxxx.......|............|............|............|
               //            |            |     xxxxx..|............|............|............|
               //            |            |     /-   xx*xxxx........|............|............|
               //            |            | dy <       |    xxxxxx..|............|............|