From http://www.jwz.org/xscreensaver/xscreensaver-5.40.tar.gz

[xscreensaver] / jwxyz / jwxyz.m

/* xscreensaver, Copyright (c) 1991-2018 Jamie Zawinski <jwz@jwz.org>
 *
 * Permission to use, copy, modify, distribute, and sell this software and its
 * documentation for any purpose is hereby granted without fee, provided that
 * the above copyright notice appear in all copies and that both that
 * copyright notice and this permission notice appear in supporting
 * documentation.  No representations are made about the suitability of this
 * software for any purpose.  It is provided "as is" without express or 
 * implied warranty.
 */

/* JWXYZ Is Not Xlib.

   But it's a bunch of function definitions that bear some resemblance to
   Xlib and that do Cocoa-ish things that bear some resemblance to the
   things that Xlib might have done.

   This is the original version of jwxyz for MacOS and iOS.
   The version used by Android is in jwxyz-gl.c and jwxyz-common.c.
   Both versions depend on jwxyz-cocoa.m.
 */

#ifdef JWXYZ_QUARTZ // entire file

#import <stdlib.h>
#import <stdint.h>
#import <wchar.h>

#ifdef USE_IPHONE
# import <UIKit/UIKit.h>
# import <UIKit/UIScreen.h>
# import <QuartzCore/QuartzCore.h>
# define NSView  UIView
# define NSRect  CGRect
# define NSPoint CGPoint
# define NSSize  CGSize
# define NSColor UIColor
# define NSImage UIImage
# define NSEvent UIEvent
# define NSFont  UIFont
# define NSGlyph CGGlyph
# define NSWindow UIWindow
# define NSMakeSize   CGSizeMake
# define NSBezierPath UIBezierPath
# define colorWithDeviceRed colorWithRed
#else
# import <Cocoa/Cocoa.h>
#endif

#import <CoreText/CTFont.h>
#import <CoreText/CTLine.h>

#import "jwxyzI.h"
#import "jwxyz-cocoa.h"
#import "jwxyz-timers.h"
#import "yarandom.h"
#import "utf8wc.h"
#import "xft.h"


struct jwxyz_Display {
  const struct jwxyz_vtbl *vtbl; // Must come first.

  Window main_window;
  CGBitmapInfo bitmap_info;
  Visual visual;
  struct jwxyz_sources_data *timers_data;

# ifndef USE_IPHONE
  CGDirectDisplayID cgdpy;  /* ...of the one and only Window, main_window.
                               This can change if the window is dragged to
                               a different screen. */
# endif

  CGColorSpaceRef colorspace;  /* Color space of this screen.  We tag all of
                                  our images with this to avoid translation
                                  when rendering. */

  unsigned long window_background;
};

struct jwxyz_GC {
  XGCValues gcv;
  unsigned int depth;
  CGImageRef clip_mask;  // CGImage copy of the Pixmap in gcv.clip_mask
};


// 8/16/24/32bpp -> 32bpp image conversion.
// Any of RGBA, BGRA, ABGR, or ARGB can be represented by a rotate of 0/8/16/24
// bits and an optional byte order swap.

// This type encodes such a conversion.
typedef unsigned convert_mode_t;

// It's rotate, then swap.
// A rotation here shifts bytes forward in memory. On x86/ARM, that's a left
// rotate, and on PowerPC, a rightward rotation.
static const convert_mode_t CONVERT_MODE_ROTATE_MASK = 0x3;
static const convert_mode_t CONVERT_MODE_SWAP = 0x4;


#if defined __LITTLE_ENDIAN__
# define PAD(r, g, b, a) ((r) | ((g) << 8) | ((b) << 16) | ((a) << 24))
#elif defined __BIG_ENDIAN__
# define PAD(r, g, b, a) (((r) << 24) | ((g) << 16) | ((b) << 8) | (a))
#else
# error "Can't determine system endianness."
#endif


// Converts an array of pixels ('src') from one format to another, placing the
// result in 'dest', according to the pixel conversion mode 'mode'.
static void
convert_row (uint32_t *dest, const void *src, size_t count,
             convert_mode_t mode, size_t src_bpp)
{
  Assert (src_bpp == 8 || src_bpp == 24 || src_bpp == 16 || src_bpp == 32,
          "weird bpp");

  // This works OK iff src == dest or src and dest do not overlap.

  if (!mode && src_bpp == 32) {
    if (src != dest)
      memcpy (dest, src, count * 4);
    return;
  }

  // This is correct, but not fast.
  convert_mode_t rot = (mode & CONVERT_MODE_ROTATE_MASK) * 8;
  convert_mode_t flip = mode & CONVERT_MODE_SWAP;

  src_bpp /= 8;

  uint32_t *dest_end = dest + count;
  while (dest != dest_end) {
    uint32_t x;

    const uint8_t *src8 = (const uint8_t *)src;
    switch (src_bpp) {
      case 4:
        x = *(const uint32_t *)src;
        break;
      case 3:
        x = PAD(src8[0], src8[1], src8[2], 0xff);
        break;
      case 2:
        x = PAD(src8[0], src8[0], src8[0], src8[1]);
        break;
      case 1:
        x = PAD(src8[0], src8[0], src8[0], 0xff);
        break;
    }

    src = (const uint8_t *)src + src_bpp;

    /* The naive (i.e. ubiquitous) portable implementation of bitwise rotation,
       for 32-bit integers, is:

       (x << rot) | (x >> (32 - rot))

       This works nearly everywhere. Compilers on x86 wil generally recognize
       the idiom and convert it to a ROL instruction. But there's a problem
       here: according to the C specification, bit shifts greater than or equal
       to the length of the integer are undefined. And if rot = 0:
       1. (x << 0) | (x >> (32 - 0))
       2. (x << 0) | (x >> 32)
       3. (x << 0) | (Undefined!)

       Still, when the compiler converts this to a ROL on x86, everything works
       as intended. But, there are two additional problems when Clang does
       compile-time constant expression evaluation with the (x >> 32)
       expression:
       1. Instead of evaluating it to something reasonable (either 0, like a
          human would intuitively expect, or x, like x86 would with SHR), Clang
          seems to pull a value out of nowhere, like -1, or some other random
          number.
       2. Clang's warning for this, -Wshift-count-overflow, only works when the
          shift count is a literal constant, as opposed to an arbitrary
          expression that is optimized down to a constant.
       Put together, this means that the assertions in
       jwxyz_quartz_make_display with convert_px break with the above naive
       rotation, but only for a release build.

       http://blog.regehr.org/archives/1063
       http://llvm.org/bugs/show_bug.cgi?id=17332
       As described in those links, there is a solution here: Masking the
       undefined shift with '& 31' as below makes the experesion well-defined
       again. And LLVM is set to pick up on this safe version of the idiom and
       use a rotation instruction on architectures (like x86) that support it,
       just like it does with the unsafe version.

       Too bad LLVM doesn't want to pick up on that particular optimization
       here. Oh well. At least this code usually isn't critical w.r.t.
       performance.
     */

# if defined __LITTLE_ENDIAN__
    x = (x << rot) | (x >> ((32 - rot) & 31));
# elif defined __BIG_ENDIAN__
    x = (x >> rot) | (x << ((32 - rot) & 31));
# endif

    if (flip)
      x = __builtin_bswap32(x); // LLVM/GCC built-in function.

    *dest = x;
    ++dest;
  }
}


// Converts a single pixel.
static uint32_t
convert_px (uint32_t px, convert_mode_t mode)
{
  convert_row (&px, &px, 1, mode, 32);
  return px;
}


// This returns the inverse conversion mode, such that:
// pixel
//   == convert_px(convert_px(pixel, mode), convert_mode_invert(mode))
//   == convert_px(convert_px(pixel, convert_mode_invert(mode)), mode)
static convert_mode_t
convert_mode_invert (convert_mode_t mode)
{
  // swap(0); rot(n) == rot(n); swap(0)
  // swap(1); rot(n) == rot(-n); swap(1)
  return mode & CONVERT_MODE_SWAP ? mode : CONVERT_MODE_ROTATE_MASK & -mode;
}


// This combines two conversions into one, such that:
// convert_px(convert_px(pixel, mode0), mode1)
//   == convert_px(pixel, convert_mode_merge(mode0, mode1))
static convert_mode_t
convert_mode_merge (convert_mode_t m0, convert_mode_t m1)
{
  // rot(r0); swap(s0); rot(r1); swap(s1)
  // rot(r0); rot(s0 ? -r1 : r1); swap(s0); swap(s1)
  // rot(r0 + (s0 ? -r1 : r1)); swap(s0 + s1)
  return
    ((m0 + (m0 & CONVERT_MODE_SWAP ? -m1 : m1)) & CONVERT_MODE_ROTATE_MASK) |
    ((m0 ^ m1) & CONVERT_MODE_SWAP);
}


// This returns a conversion mode that converts an arbitrary 32-bit format
// specified by bitmap_info to RGBA.
static convert_mode_t
convert_mode_to_rgba (CGBitmapInfo bitmap_info)
{
  // Former default: kCGImageAlphaNoneSkipFirst | kCGBitmapByteOrder32Little
  // i.e. BGRA
  // red   = 0x00FF0000;
  // green = 0x0000FF00;
  // blue  = 0x000000FF;

  // RGBA: kCGImageAlphaNoneSkipLast | kCGBitmapByteOrder32Big

  CGImageAlphaInfo alpha_info =
    (CGImageAlphaInfo)(bitmap_info & kCGBitmapAlphaInfoMask);

  Assert (! (bitmap_info & kCGBitmapFloatComponents),
          "kCGBitmapFloatComponents unsupported");
  Assert (alpha_info != kCGImageAlphaOnly, "kCGImageAlphaOnly not supported");

  convert_mode_t rot = alpha_info == kCGImageAlphaFirst ||
                       alpha_info == kCGImageAlphaPremultipliedFirst ||
                       alpha_info == kCGImageAlphaNoneSkipFirst ?
                       3 : 0;

  CGBitmapInfo byte_order = bitmap_info & kCGBitmapByteOrderMask;

  Assert (byte_order == kCGBitmapByteOrder32Little ||
          byte_order == kCGBitmapByteOrder32Big,
          "byte order not supported");

  convert_mode_t swap = byte_order == kCGBitmapByteOrder32Little ?
                        CONVERT_MODE_SWAP : 0;
  if (swap)
    rot = CONVERT_MODE_ROTATE_MASK & -rot;
  return swap | rot;
}


union color_bytes
{
  uint32_t pixel;
  uint8_t bytes[4];
};


static void
query_color_float (Display *dpy, unsigned long pixel, CGFloat *rgba)
{
  JWXYZ_QUERY_COLOR (dpy, pixel, (CGFloat)1, rgba);
}


extern const struct jwxyz_vtbl quartz_vtbl;

Display *
jwxyz_quartz_make_display (Window w)
{
  CGContextRef cgc = w->cgc;

  Display *d = (Display *) calloc (1, sizeof(*d));
  d->vtbl = &quartz_vtbl;

  d->bitmap_info = CGBitmapContextGetBitmapInfo (cgc);

# if 0
  // Tests for the image conversion modes.
  {
    const uint32_t key = 0x04030201;
#  ifdef __LITTLE_ENDIAN__
    assert (convert_px (key, 0) == key);
    assert (convert_px (key, 1) == 0x03020104);
    assert (convert_px (key, 3) == 0x01040302);
    assert (convert_px (key, 4) == 0x01020304);
    assert (convert_px (key, 5) == 0x04010203);
#  endif
    for (unsigned i = 0; i != 8; ++i) {
      assert (convert_px(convert_px(key, i), convert_mode_invert(i)) == key);
      assert (convert_mode_invert(convert_mode_invert(i)) == i);
    }

    for (unsigned i = 0; i != 8; ++i) {
      for (unsigned j = 0; j != 8; ++j)
        assert (convert_px(convert_px(key, i), j) ==
                convert_px(key, convert_mode_merge(i, j)));
    }
  }
# endif

  Visual *v = &d->visual;
  v->class      = TrueColor;

  union color_bytes color;
  convert_mode_t mode =
    convert_mode_invert (convert_mode_to_rgba (d->bitmap_info));
  unsigned long masks[4];
  for (unsigned i = 0; i != 4; ++i) {
    color.pixel = 0;
    color.bytes[i] = 0xff;
    masks[i] = convert_px (color.pixel, mode);
  }
  v->red_mask   = masks[0];
  v->green_mask = masks[1];
  v->blue_mask  = masks[2];
  v->alpha_mask = masks[3];

  CGBitmapInfo byte_order = d->bitmap_info & kCGBitmapByteOrderMask;
  Assert ( ! (d->bitmap_info & kCGBitmapFloatComponents) &&
          (byte_order == kCGBitmapByteOrder32Little ||
           byte_order == kCGBitmapByteOrder32Big),
          "invalid bits per channel");
  
  d->timers_data = jwxyz_sources_init (XtDisplayToApplicationContext (d));

  d->window_background = BlackPixel(d,0);

  d->main_window = w;

  Assert (cgc, "no CGContext");
  return d;
}

void
jwxyz_quartz_free_display (Display *dpy)
{
  jwxyz_sources_free (dpy->timers_data);
  
  free (dpy);
}


/* Call this after any modification to the bits on a Pixmap or Window.
   Most Pixmaps are used frequently as sources and infrequently as
   destinations, so it pays to cache the data as a CGImage as needed.
 */
void
invalidate_drawable_cache (Drawable d)
{
  if (d && d->cgi) {
    CGImageRelease (d->cgi);
    d->cgi = 0;
  }
}


/* Call this when the View changes size or position.
 */
void
jwxyz_window_resized (Display *dpy)
{
  Window w = dpy->main_window;

# ifndef USE_IPHONE
  // Figure out which screen the window is currently on.
  {
    int wx, wy;
    XTranslateCoordinates (dpy, w, NULL, 0, 0, &wx, &wy, NULL);
    CGPoint p;
    p.x = wx;
    p.y = wy;
    CGDisplayCount n;
    dpy->cgdpy = 0;
    CGGetDisplaysWithPoint (p, 1, &dpy->cgdpy, &n);
    // Auuugh!
    if (! dpy->cgdpy) {
      p.x = p.y = 0;
      CGGetDisplaysWithPoint (p, 1, &dpy->cgdpy, &n);
    }
    Assert (dpy->cgdpy, "unable to find CGDisplay");
  }
# endif // USE_IPHONE

/*
  {
    // Figure out this screen's colorspace, and use that for every CGImage.
    //
    CMProfileRef profile = 0;
    CMGetProfileByAVID ((CMDisplayIDType) dpy->cgdpy, &profile);
    Assert (profile, "unable to find colorspace profile");
    dpy->colorspace = CGColorSpaceCreateWithPlatformColorSpace (profile);
    Assert (dpy->colorspace, "unable to find colorspace");
  }
 */

  // WTF?  It's faster if we *do not* use the screen's colorspace!
  //
  dpy->colorspace = CGColorSpaceCreateDeviceRGB();

  invalidate_drawable_cache (w);
}


void
jwxyz_flush_context (Display *dpy)
{
  // CGContextSynchronize is another possibility.
  CGContextFlush(dpy->main_window->cgc);
}

static jwxyz_sources_data *
display_sources_data (Display *dpy)
{
  return dpy->timers_data;
}


static Window
root (Display *dpy)
{
  return dpy->main_window;
}

static Visual *
visual (Display *dpy)
{
  return &dpy->visual;
}


void
set_color (Display *dpy, CGContextRef cgc, unsigned long argb,
           unsigned int depth, Bool alpha_allowed_p, Bool fill_p)
{
  jwxyz_validate_pixel (dpy, argb, depth, alpha_allowed_p);
  if (depth == 1) {
    if (fill_p)
      CGContextSetGrayFillColor   (cgc, (argb ? 1.0 : 0.0), 1.0);
    else
      CGContextSetGrayStrokeColor (cgc, (argb ? 1.0 : 0.0), 1.0);
  } else {
    CGFloat rgba[4];
    query_color_float (dpy, argb, rgba);
    if (fill_p)
      CGContextSetRGBFillColor   (cgc, rgba[0], rgba[1], rgba[2], rgba[3]);
    else
      CGContextSetRGBStrokeColor (cgc, rgba[0], rgba[1], rgba[2], rgba[3]);
  }
}

static void
set_line_mode (CGContextRef cgc, XGCValues *gcv)
{
  CGContextSetLineWidth (cgc, gcv->line_width ? gcv->line_width : 1);
  CGContextSetLineJoin  (cgc,
                         gcv->join_style == JoinMiter ? kCGLineJoinMiter :
                         gcv->join_style == JoinRound ? kCGLineJoinRound :
                         kCGLineJoinBevel);
  CGContextSetLineCap   (cgc, 
                         gcv->cap_style == CapNotLast ? kCGLineCapButt  :
                         gcv->cap_style == CapButt    ? kCGLineCapButt  :
                         gcv->cap_style == CapRound   ? kCGLineCapRound :
                         kCGLineCapSquare);
}

static void
set_clip_mask (Drawable d, GC gc)
{
  Assert (!!gc->gcv.clip_mask == !!gc->clip_mask, "GC clip mask mixup");

  Pixmap p = gc->gcv.clip_mask;
  if (!p) return;
  Assert (p->type == PIXMAP, "not a pixmap");

  XRectangle wr = d->frame;
  CGRect to;
  to.origin.x    = wr.x + gc->gcv.clip_x_origin;
  to.origin.y    = wr.y + wr.height - gc->gcv.clip_y_origin
                    - p->frame.height;
  to.size.width  = p->frame.width;
  to.size.height = p->frame.height;

  CGContextClipToMask (d->cgc, to, gc->clip_mask);
}


/* Pushes a GC context; sets BlendMode and ClipMask.
 */
void
push_gc (Drawable d, GC gc)
{
  CGContextRef cgc = d->cgc;
  CGContextSaveGState (cgc);

  switch (gc->gcv.function) {
    case GXset:
    case GXclear:
    case GXcopy:/*CGContextSetBlendMode (cgc, kCGBlendModeNormal);*/   break;
    case GXxor:   CGContextSetBlendMode (cgc, kCGBlendModeDifference); break;
    case GXor:    CGContextSetBlendMode (cgc, kCGBlendModeLighten);    break;
    case GXand:   CGContextSetBlendMode (cgc, kCGBlendModeDarken);     break;
    default: Assert(0, "unknown gcv function"); break;
  }

  if (gc->gcv.clip_mask)
    set_clip_mask (d, gc);
}


/* Pushes a GC context; sets BlendMode, ClipMask, Fill, and Stroke colors.
 */
void
push_color_gc (Display *dpy, Drawable d, GC gc, unsigned long color,
               Bool antialias_p, Bool fill_p)
{
  push_gc (d, gc);

  int depth = gc->depth;
  switch (gc->gcv.function) {
    case GXset:   color = (depth == 1 ? 1 : WhitePixel(dpy,0)); break;
    case GXclear: color = (depth == 1 ? 0 : BlackPixel(dpy,0)); break;
  }

  CGContextRef cgc = d->cgc;
  set_color (dpy, cgc, color, depth, gc->gcv.alpha_allowed_p, fill_p);
  CGContextSetShouldAntialias (cgc, antialias_p);
}


/* Pushes a GC context; sets Fill and Stroke colors to the foreground color.
 */
static void
push_fg_gc (Display *dpy, Drawable d, GC gc, Bool fill_p)
{
  push_color_gc (dpy, d, gc, gc->gcv.foreground, gc->gcv.antialias_p, fill_p);
}

static Bool
bitmap_context_p (Drawable d)
{
  return True;
}



/* You've got to be fucking kidding me!

   It is *way* faster to draw points by creating and drawing a 1x1 CGImage
   with repeated calls to CGContextDrawImage than it is to make a single
   call to CGContextFillRects() with a list of 1x1 rectangles!

   I still wouldn't call it *fast*, however...
 */
#define XDRAWPOINTS_IMAGES

/* Update, 2012: Kurt Revis <krevis@snoize.com> points out that diddling
   the bitmap data directly is faster.  This only works on Pixmaps, though,
   not Windows.  (Fortunately, on iOS, the Window is really a Pixmap.)
 */
#define XDRAWPOINTS_CGDATA

static int
DrawPoints (Display *dpy, Drawable d, GC gc,
            XPoint *points, int count, int mode)
{
  int i;
  XRectangle wr = d->frame;

# ifdef XDRAWPOINTS_CGDATA

  if (bitmap_context_p (d))
  {
    CGContextRef cgc = d->cgc;
    void *data = CGBitmapContextGetData (cgc);
    size_t bpr = CGBitmapContextGetBytesPerRow (cgc);
    size_t w = CGBitmapContextGetWidth (cgc);
    size_t h = CGBitmapContextGetHeight (cgc);

    Assert (data, "no bitmap data in Drawable");

    unsigned long argb = gc->gcv.foreground;
    jwxyz_validate_pixel (dpy, argb, gc->depth, gc->gcv.alpha_allowed_p);
    if (gc->depth == 1)
      argb = (gc->gcv.foreground ? WhitePixel(dpy,0) : BlackPixel(dpy,0));

    CGFloat x0 = wr.x;
    CGFloat y0 = wr.y; // Y axis is refreshingly not flipped.

    // It's uglier, but faster, to hoist the conditional out of the loop.
    if (mode == CoordModePrevious) {
      CGFloat x = x0, y = y0;
      for (i = 0; i < count; i++, points++) {
        x += points->x;
        y += points->y;

        if (x >= 0 && x < w && y >= 0 && y < h) {
          unsigned int *p = (unsigned int *)
            ((char *) data + (size_t) y * bpr + (size_t) x * 4);
          *p = (unsigned int) argb;
        }
      }
    } else {
      for (i = 0; i < count; i++, points++) {
        CGFloat x = x0 + points->x;
        CGFloat y = y0 + points->y;

        if (x >= 0 && x < w && y >= 0 && y < h) {
          unsigned int *p = (unsigned int *)
            ((char *) data + (size_t) y * bpr + (size_t) x * 4);
          *p = (unsigned int) argb;
        }
      }
    }

  } else        /* d->type == WINDOW */

# endif /* XDRAWPOINTS_CGDATA */
  {
    push_fg_gc (dpy, d, gc, YES);

# ifdef XDRAWPOINTS_IMAGES

    unsigned long argb = gc->gcv.foreground;
    jwxyz_validate_pixel (dpy, argb, gc->depth, gc->gcv.alpha_allowed_p);
    if (gc->depth == 1)
      argb = (gc->gcv.foreground ? WhitePixel(dpy,0) : BlackPixel(dpy,0));

    CGDataProviderRef prov = CGDataProviderCreateWithData (NULL, &argb, 4,
                                                           NULL);
    CGImageRef cgi = CGImageCreate (1, 1,
                                    8, 32, 4,
                                    dpy->colorspace, 
                                    /* Host-ordered, since we're using the
                                       address of an int as the color data. */
                                    dpy->bitmap_info,
                                    prov, 
                                    NULL,  /* decode[] */
                                    NO, /* interpolate */
                                    kCGRenderingIntentDefault);
    CGDataProviderRelease (prov);

    CGContextRef cgc = d->cgc;
    CGRect rect;
    rect.size.width = rect.size.height = 1;
    for (i = 0; i < count; i++) {
      if (i > 0 && mode == CoordModePrevious) {
        rect.origin.x += points->x;
        rect.origin.x -= points->y;
      } else {
        rect.origin.x = wr.x + points->x;
        rect.origin.y = wr.y + wr.height - points->y - 1;
      }

      //Assert(CGImageGetColorSpace (cgi) == dpy->colorspace,"bad colorspace");
      CGContextDrawImage (cgc, rect, cgi);
      points++;
    }

    CGImageRelease (cgi);

# else /* ! XDRAWPOINTS_IMAGES */

    CGRect *rects = (CGRect *) malloc (count * sizeof(CGRect));
    CGRect *r = rects;
  
    for (i = 0; i < count; i++) {
      r->size.width = r->size.height = 1;
      if (i > 0 && mode == CoordModePrevious) {
        r->origin.x = r[-1].origin.x + points->x;
        r->origin.y = r[-1].origin.x - points->y;
      } else {
        r->origin.x = wr.origin.x + points->x;
        r->origin.y = wr.origin.y + wr.size.height - points->y;
      }
      points++;
      r++;
    }

    CGContextFillRects (d->cgc, rects, count);
    free (rects);

# endif /* ! XDRAWPOINTS_IMAGES */

    pop_gc (d, gc);
  }

  invalidate_drawable_cache (d);

  return 0;
}


CGPoint
map_point (Drawable d, int x, int y)
{
  const XRectangle *wr = &d->frame;
  CGPoint p;
  p.x = wr->x + x;
  p.y = wr->y + wr->height - y;
  return p;
}


static void
adjust_point_for_line (GC gc, CGPoint *p)
{
  // Here's the authoritative discussion on how X draws lines:
  // http://www.x.org/releases/current/doc/xproto/x11protocol.html#requests:CreateGC:line-width
  if (gc->gcv.line_width <= 1) {
    /* Thin lines are "drawn using an unspecified, device-dependent
       algorithm", but seriously though, Bresenham's algorithm. Bresenham's
       algorithm runs to and from pixel centers.

       There's a few screenhacks (Maze, at the very least) that set line_width
       to 1 when it probably should be set to 0, so it's line_width <= 1
       instead of < 1.
     */
    p->x += 0.5;
    p->y -= 0.5;
  } else {
    /* Thick lines OTOH run from the upper-left corners of pixels. This means
       that a horizontal thick line of width 1 straddles two scan lines.
       Aliasing requires one of these scan lines be chosen; the following
       nudges the point so that the right choice is made. */
    p->y -= 1e-3;
  }
}


static CGPoint
point_for_line (Drawable d, GC gc, int x, int y)
{
  CGPoint result = map_point (d, x, y);
  adjust_point_for_line (gc, &result);
  return result;
}


static int
DrawLines (Display *dpy, Drawable d, GC gc, XPoint *points, int count,
            int mode)
{
  int i;
  CGPoint p;
  push_fg_gc (dpy, d, gc, NO);

  CGContextRef cgc = d->cgc;

  set_line_mode (cgc, &gc->gcv);
  
  // if the first and last points coincide, use closepath to get
  // the proper line-joining.
  BOOL closed_p = (points[0].x == points[count-1].x &&
                   points[0].y == points[count-1].y);
  if (closed_p) count--;
  
  p = point_for_line(d, gc, points->x, points->y);
  points++;
  CGContextBeginPath (cgc);
  CGContextMoveToPoint (cgc, p.x, p.y);
  for (i = 1; i < count; i++) {
    if (mode == CoordModePrevious) {
      p.x += points->x;
      p.y -= points->y;
    } else {
      p = point_for_line(d, gc, points->x, points->y);
    }
    CGContextAddLineToPoint (cgc, p.x, p.y);
    points++;
  }
  if (closed_p) CGContextClosePath (cgc);
  CGContextStrokePath (cgc);
  pop_gc (d, gc);
  invalidate_drawable_cache (d);
  return 0;
}


static int
DrawSegments (Display *dpy, Drawable d, GC gc, XSegment *segments, int count)
{
  int i;

  CGContextRef cgc = d->cgc;

  push_fg_gc (dpy, d, gc, NO);
  set_line_mode (cgc, &gc->gcv);
  CGContextBeginPath (cgc);
  for (i = 0; i < count; i++) {
    // when drawing a zero-length line, obey line-width and cap-style.
    if (segments->x1 == segments->x2 && segments->y1 == segments->y2) {
      int w = gc->gcv.line_width;
      int x1 = segments->x1 - w/2;
      int y1 = segments->y1 - w/2;
      if (gc->gcv.line_width > 1 && gc->gcv.cap_style == CapRound)
        XFillArc (dpy, d, gc, x1, y1, w, w, 0, 360*64);
      else {
        if (!w)
          w = 1; // Actually show zero-length lines.
        XFillRectangle (dpy, d, gc, x1, y1, w, w);
      }
    } else {
      CGPoint p = point_for_line (d, gc, segments->x1, segments->y1);
      CGContextMoveToPoint (cgc, p.x, p.y);
      p = point_for_line (d, gc, segments->x2, segments->y2);
      CGContextAddLineToPoint (cgc, p.x, p.y);
    }

    segments++;
  }
  CGContextStrokePath (cgc);
  pop_gc (d, gc);
  invalidate_drawable_cache (d);
  return 0;
}


static int
ClearWindow (Display *dpy, Window win)
{
  Assert (win && win->type == WINDOW, "not a window");
  XRectangle wr = win->frame;
  return XClearArea (dpy, win, 0, 0, wr.width, wr.height, 0);
}

static unsigned long *
window_background (Display *dpy)
{
  return &dpy->window_background;
}

static void
fill_rects (Display *dpy, Drawable d, GC gc,
            const XRectangle *rectangles, unsigned long nrectangles,
            unsigned long pixel)
{
  Assert (!gc || gc->depth == jwxyz_drawable_depth (d), "depth mismatch");

  CGContextRef cgc = d->cgc;

  Bool fast_fill_p =
    bitmap_context_p (d) &&
    (!gc || (gc->gcv.function == GXcopy &&
             !gc->gcv.alpha_allowed_p &&
             !gc->gcv.clip_mask));

  if (!fast_fill_p) {
    if (gc)
      push_color_gc (dpy, d, gc, pixel, gc->gcv.antialias_p, YES);
    else
      set_color (dpy, d->cgc, pixel, jwxyz_drawable_depth (d), NO, YES);
  }

  for (unsigned i = 0; i != nrectangles; ++i) {

    int x = rectangles[i].x;
    int y = rectangles[i].y;
    unsigned long width = rectangles[i].width;
    unsigned long height = rectangles[i].height;

    if (fast_fill_p) {
      long   // negative_int > unsigned_int == 1
        dw = CGBitmapContextGetWidth (cgc),
        dh = CGBitmapContextGetHeight (cgc);

      if (x >= dw || y >= dh)
        continue;

      if (x < 0) {
        width += x;
        x = 0;
      }

      if (y < 0) {
        height += y;
        y = 0;
      }

      if (width <= 0 || height <= 0)
        continue;

      unsigned long max_width = dw - x;
      if (width > max_width)
        width = max_width;
      unsigned long max_height = dh - y;
      if (height > max_height)
        height = max_height;

      if (jwxyz_drawable_depth (d) == 1)
        pixel = pixel ? WhitePixel(dpy, 0) : BlackPixel(dpy, 0);

      size_t dst_bytes_per_row = CGBitmapContextGetBytesPerRow (d->cgc);
      void *dst = SEEK_XY (CGBitmapContextGetData (d->cgc),
                           dst_bytes_per_row, x, y);

      Assert(sizeof(wchar_t) == 4, "somebody changed the ABI");
      while (height) {
        // Would be nice if Apple used SSE/NEON in wmemset. Maybe someday.
        wmemset (dst, (wchar_t) pixel, width);
        --height;
        dst = (char *) dst + dst_bytes_per_row;
      }

    } else {
      CGRect r;
      r.origin = map_point (d, x, y);
      r.origin.y -= height;
      r.size.width = width;
      r.size.height = height;
      CGContextFillRect (cgc, r);
    }
  }

  if (!fast_fill_p && gc)
    pop_gc (d, gc);
  invalidate_drawable_cache (d);
}


static int
FillPolygon (Display *dpy, Drawable d, GC gc,
             XPoint *points, int npoints, int shape, int mode)
{
  XRectangle wr = d->frame;
  int i;
  push_fg_gc (dpy, d, gc, YES);
  CGContextRef cgc = d->cgc;
  CGContextBeginPath (cgc);
  float x = 0, y = 0;
  for (i = 0; i < npoints; i++) {
    if (i > 0 && mode == CoordModePrevious) {
      x += points[i].x;
      y -= points[i].y;
    } else {
      x = wr.x + points[i].x;
      y = wr.y + wr.height - points[i].y;
    }
        
    if (i == 0)
      CGContextMoveToPoint (cgc, x, y);
    else
      CGContextAddLineToPoint (cgc, x, y);
  }
  CGContextClosePath (cgc);
  if (gc->gcv.fill_rule == EvenOddRule)
    CGContextEOFillPath (cgc);
  else
    CGContextFillPath (cgc);
  pop_gc (d, gc);
  invalidate_drawable_cache (d);
  return 0;
}

#define radians(DEG) ((DEG) * M_PI / 180.0)
#define degrees(RAD) ((RAD) * 180.0 / M_PI)

static int
draw_arc (Display *dpy, Drawable d, GC gc, int x, int y,
                unsigned int width, unsigned int height,
                int angle1, int angle2, Bool fill_p)
{
  XRectangle wr = d->frame;
  CGRect bound;
  bound.origin.x = wr.x + x;
  bound.origin.y = wr.y + wr.height - y - (int)height;
  bound.size.width = width;
  bound.size.height = height;
  
  CGPoint ctr;
  ctr.x = bound.origin.x + bound.size.width /2;
  ctr.y = bound.origin.y + bound.size.height/2;
  
  float r1 = radians (angle1/64.0);
  float r2 = radians (angle2/64.0) + r1;
  BOOL clockwise = angle2 < 0;
  BOOL closed_p = (angle2 >= 360*64 || angle2 <= -360*64);
  
  push_fg_gc (dpy, d, gc, fill_p);

  CGContextRef cgc = d->cgc;
  CGContextBeginPath (cgc);
  
  CGContextSaveGState(cgc);
  CGContextTranslateCTM (cgc, ctr.x, ctr.y);
  CGContextScaleCTM (cgc, width/2.0, height/2.0);
  if (fill_p)
    CGContextMoveToPoint (cgc, 0, 0);

  CGContextAddArc (cgc, 0.0, 0.0, 1, r1, r2, clockwise);
  CGContextRestoreGState (cgc);  // restore before stroke, for line width

  if (closed_p)
    CGContextClosePath (cgc); // for proper line joining
  
  if (fill_p) {
    CGContextFillPath (cgc);
  } else {
    set_line_mode (cgc, &gc->gcv);
    CGContextStrokePath (cgc);
  }

  pop_gc (d, gc);
  invalidate_drawable_cache (d);
  return 0;
}


static XGCValues *
gc_gcv (GC gc)
{
  return &gc->gcv;
}


static unsigned int
gc_depth (GC gc)
{
  return gc->depth;
}


static GC
CreateGC (Display *dpy, Drawable d, unsigned long mask, XGCValues *xgcv)
{
  struct jwxyz_GC *gc = (struct jwxyz_GC *) calloc (1, sizeof(*gc));
  gc->depth = jwxyz_drawable_depth (d);

  jwxyz_gcv_defaults (dpy, &gc->gcv, gc->depth);
  XChangeGC (dpy, gc, mask, xgcv);
  return gc;
}


static int
FreeGC (Display *dpy, GC gc)
{
  if (gc->gcv.font)
    XUnloadFont (dpy, gc->gcv.font);

  Assert (!!gc->gcv.clip_mask == !!gc->clip_mask, "GC clip mask mixup");

  if (gc->gcv.clip_mask) {
    XFreePixmap (dpy, gc->gcv.clip_mask);
    CGImageRelease (gc->clip_mask);
  }
  free (gc);
  return 0;
}


static void
flipbits (unsigned const char *in, unsigned char *out, int length)
{
  static const unsigned char table[256] = {
    0x00, 0x80, 0x40, 0xC0, 0x20, 0xA0, 0x60, 0xE0, 
    0x10, 0x90, 0x50, 0xD0, 0x30, 0xB0, 0x70, 0xF0, 
    0x08, 0x88, 0x48, 0xC8, 0x28, 0xA8, 0x68, 0xE8, 
    0x18, 0x98, 0x58, 0xD8, 0x38, 0xB8, 0x78, 0xF8, 
    0x04, 0x84, 0x44, 0xC4, 0x24, 0xA4, 0x64, 0xE4, 
    0x14, 0x94, 0x54, 0xD4, 0x34, 0xB4, 0x74, 0xF4, 
    0x0C, 0x8C, 0x4C, 0xCC, 0x2C, 0xAC, 0x6C, 0xEC, 
    0x1C, 0x9C, 0x5C, 0xDC, 0x3C, 0xBC, 0x7C, 0xFC, 
    0x02, 0x82, 0x42, 0xC2, 0x22, 0xA2, 0x62, 0xE2, 
    0x12, 0x92, 0x52, 0xD2, 0x32, 0xB2, 0x72, 0xF2, 
    0x0A, 0x8A, 0x4A, 0xCA, 0x2A, 0xAA, 0x6A, 0xEA, 
    0x1A, 0x9A, 0x5A, 0xDA, 0x3A, 0xBA, 0x7A, 0xFA, 
    0x06, 0x86, 0x46, 0xC6, 0x26, 0xA6, 0x66, 0xE6, 
    0x16, 0x96, 0x56, 0xD6, 0x36, 0xB6, 0x76, 0xF6, 
    0x0E, 0x8E, 0x4E, 0xCE, 0x2E, 0xAE, 0x6E, 0xEE, 
    0x1E, 0x9E, 0x5E, 0xDE, 0x3E, 0xBE, 0x7E, 0xFE, 
    0x01, 0x81, 0x41, 0xC1, 0x21, 0xA1, 0x61, 0xE1, 
    0x11, 0x91, 0x51, 0xD1, 0x31, 0xB1, 0x71, 0xF1, 
    0x09, 0x89, 0x49, 0xC9, 0x29, 0xA9, 0x69, 0xE9, 
    0x19, 0x99, 0x59, 0xD9, 0x39, 0xB9, 0x79, 0xF9, 
    0x05, 0x85, 0x45, 0xC5, 0x25, 0xA5, 0x65, 0xE5, 
    0x15, 0x95, 0x55, 0xD5, 0x35, 0xB5, 0x75, 0xF5, 
    0x0D, 0x8D, 0x4D, 0xCD, 0x2D, 0xAD, 0x6D, 0xED, 
    0x1D, 0x9D, 0x5D, 0xDD, 0x3D, 0xBD, 0x7D, 0xFD, 
    0x03, 0x83, 0x43, 0xC3, 0x23, 0xA3, 0x63, 0xE3, 
    0x13, 0x93, 0x53, 0xD3, 0x33, 0xB3, 0x73, 0xF3, 
    0x0B, 0x8B, 0x4B, 0xCB, 0x2B, 0xAB, 0x6B, 0xEB, 
    0x1B, 0x9B, 0x5B, 0xDB, 0x3B, 0xBB, 0x7B, 0xFB, 
    0x07, 0x87, 0x47, 0xC7, 0x27, 0xA7, 0x67, 0xE7, 
    0x17, 0x97, 0x57, 0xD7, 0x37, 0xB7, 0x77, 0xF7, 
    0x0F, 0x8F, 0x4F, 0xCF, 0x2F, 0xAF, 0x6F, 0xEF, 
    0x1F, 0x9F, 0x5F, 0xDF, 0x3F, 0xBF, 0x7F, 0xFF
  };
  while (length-- > 0)
    *out++ = table[*in++];
}


static int
PutImage (Display *dpy, Drawable d, GC gc, XImage *ximage,
          int src_x, int src_y, int dest_x, int dest_y,
          unsigned int w, unsigned int h)
{
  XRectangle wr = d->frame;

  Assert (gc, "no GC");
  Assert ((w < 65535), "improbably large width");
  Assert ((h < 65535), "improbably large height");
  Assert ((src_x  < 65535 && src_x  > -65535), "improbably large src_x");
  Assert ((src_y  < 65535 && src_y  > -65535), "improbably large src_y");
  Assert ((dest_x < 65535 && dest_x > -65535), "improbably large dest_x");
  Assert ((dest_y < 65535 && dest_y > -65535), "improbably large dest_y");

  // Clip width and height to the bounds of the Drawable
  //
  if (dest_x + (int)w > wr.width) {
    if (dest_x > wr.width)
      return 0;
    w = wr.width - dest_x;
  }
  if (dest_y + (int)h > wr.height) {
    if (dest_y > wr.height)
      return 0;
    h = wr.height - dest_y;
  }
  if (w <= 0 || h <= 0)
    return 0;

  // Clip width and height to the bounds of the XImage
  //
  if (src_x + w > ximage->width) {
    if (src_x > ximage->width)
      return 0;
    w = ximage->width - src_x;
  }
  if (src_y + h > ximage->height) {
    if (src_y > ximage->height)
      return 0;
    h = ximage->height - src_y;
  }
  if (w <= 0 || h <= 0)
    return 0;

  CGContextRef cgc = d->cgc;

  if (jwxyz_dumb_drawing_mode(dpy, d, gc, dest_x, dest_y, w, h))
    return 0;

  int bpl = ximage->bytes_per_line;
  int bpp = ximage->bits_per_pixel;
  int bsize = bpl * h;
  char *data = ximage->data;

  CGRect r;
  r.origin.x = wr.x + dest_x;
  r.origin.y = wr.y + wr.height - dest_y - (int)h;
  r.size.width = w;
  r.size.height = h;

  if (bpp == 32) {

    /* Take advantage of the fact that it's ok for (bpl != w * bpp)
       to create a CGImage from a sub-rectagle of the XImage.
     */
    data += (src_y * bpl) + (src_x * 4);
    CGDataProviderRef prov = 
      CGDataProviderCreateWithData (NULL, data, bsize, NULL);

    CGImageRef cgi = CGImageCreate (w, h,
                                    bpp/4, bpp, bpl,
                                    dpy->colorspace, 
                                    dpy->bitmap_info,
                                    prov, 
                                    NULL,  /* decode[] */
                                    NO, /* interpolate */
                                    kCGRenderingIntentDefault);
    CGDataProviderRelease (prov);
    //Assert (CGImageGetColorSpace (cgi) == dpy->colorspace, "bad colorspace");
    CGContextDrawImage (cgc, r, cgi);
    CGImageRelease (cgi);

  } else {   // (bpp == 1)

    /* To draw a 1bpp image, we use it as a mask and fill two rectangles.

       #### However, the bit order within a byte in a 1bpp XImage is
            the wrong way around from what Quartz expects, so first we
            have to copy the data to reverse it.  Shit!  Maybe it
            would be worthwhile to go through the hacks and #ifdef
            each one that diddles 1bpp XImage->data directly...
     */
    Assert ((src_x % 8) == 0,
            "XPutImage with non-byte-aligned 1bpp X offset not implemented");

    data += (src_y * bpl) + (src_x / 8);   // move to x,y within the data
    unsigned char *flipped = (unsigned char *) malloc (bsize);

    flipbits ((unsigned char *) data, flipped, bsize);

    CGDataProviderRef prov = 
      CGDataProviderCreateWithData (NULL, flipped, bsize, NULL);
    CGImageRef mask = CGImageMaskCreate (w, h, 
                                         1, bpp, bpl,
                                         prov,
                                         NULL,  /* decode[] */
                                         NO); /* interpolate */
    push_fg_gc (dpy, d, gc, YES);

    CGContextFillRect (cgc, r);                         // foreground color
    CGContextClipToMask (cgc, r, mask);
    set_color (dpy, cgc, gc->gcv.background, gc->depth, NO, YES);
    CGContextFillRect (cgc, r);                         // background color
    pop_gc (d, gc);

    free (flipped);
    CGDataProviderRelease (prov);
    CGImageRelease (mask);
  }

  invalidate_drawable_cache (d);

  return 0;
}


static XImage *
GetSubImage (Display *dpy, Drawable d, int x, int y,
             unsigned int width, unsigned int height,
             unsigned long plane_mask, int format,
             XImage *image, int dest_x, int dest_y)
{
  const unsigned char *data = 0;
  size_t depth, ibpp, ibpl;
  convert_mode_t mode;

  Assert ((width  < 65535), "improbably large width");
  Assert ((height < 65535), "improbably large height");
  Assert ((x < 65535 && x > -65535), "improbably large x");
  Assert ((y < 65535 && y > -65535), "improbably large y");

  CGContextRef cgc = d->cgc;

  {
    depth = jwxyz_drawable_depth (d);
    mode = convert_mode_to_rgba (dpy->bitmap_info);
    ibpp = CGBitmapContextGetBitsPerPixel (cgc);
    ibpl = CGBitmapContextGetBytesPerRow (cgc);
    data = CGBitmapContextGetData (cgc);
    Assert (data, "CGBitmapContextGetData failed");
  }
  
  // data points at (x,y) with ibpl rowstride.  ignore x,y from now on.
  data += (y * ibpl) + (x * (ibpp/8));
  
  format = (depth == 1 ? XYPixmap : ZPixmap);
  
  int obpl = image->bytes_per_line;
  
  /* both PPC and Intel use word-ordered ARGB frame buffers, which
     means that on Intel it is BGRA when viewed by bytes (And BGR
     when using 24bpp packing).

     BUT! Intel-64 stores alpha at the other end! 32bit=RGBA, 64bit=ARGB.
     The NSAlphaFirstBitmapFormat bit in bitmapFormat seems to be the
     indicator of this latest kink.
   */
  int xx, yy;
  if (depth == 1) {
    const unsigned char *iline = data;
    for (yy = 0; yy < height; yy++) {

      const unsigned char *iline2 = iline;
      for (xx = 0; xx < width; xx++) {

        iline2++;                     // ignore R  or  A  or  A  or  B
        iline2++;                     // ignore G  or  B  or  R  or  G
        unsigned char r = *iline2++;  // use    B  or  G  or  G  or  R
        if (ibpp == 32) iline2++;     // ignore A  or  R  or  B  or  A

        XPutPixel (image, xx + dest_x, yy + dest_y, (r ? 1 : 0));
      }
      iline += ibpl;
    }
  } else {
    const unsigned char *iline = data;
    unsigned char *oline = (unsigned char *) image->data + dest_y * obpl +
                             dest_x * 4;

    mode = convert_mode_merge (mode,
             convert_mode_invert (
               convert_mode_to_rgba (dpy->bitmap_info)));

    for (yy = 0; yy < height; yy++) {

      convert_row ((uint32_t *)oline, iline, width, mode, ibpp);

      oline += obpl;
      iline += ibpl;
    }
  }

  return image;
}



/* Returns a transformation matrix to do rotation as per the provided
   EXIF "Orientation" value.
 */
static CGAffineTransform
exif_rotate (int rot, CGSize rect)
{
  CGAffineTransform trans = CGAffineTransformIdentity;
  switch (rot) {
  case 2:               // flip horizontal
    trans = CGAffineTransformMakeTranslation (rect.width, 0);
    trans = CGAffineTransformScale (trans, -1, 1);
    break;

  case 3:               // rotate 180
    trans = CGAffineTransformMakeTranslation (rect.width, rect.height);
    trans = CGAffineTransformRotate (trans, M_PI);
    break;

  case 4:               // flip vertical
    trans = CGAffineTransformMakeTranslation (0, rect.height);
    trans = CGAffineTransformScale (trans, 1, -1);
    break;

  case 5:               // transpose (UL-to-LR axis)
    trans = CGAffineTransformMakeTranslation (rect.height, rect.width);
    trans = CGAffineTransformScale (trans, -1, 1);
    trans = CGAffineTransformRotate (trans, 3 * M_PI / 2);
    break;

  case 6:               // rotate 90
    trans = CGAffineTransformMakeTranslation (0, rect.width);
    trans = CGAffineTransformRotate (trans, 3 * M_PI / 2);
    break;

  case 7:               // transverse (UR-to-LL axis)
    trans = CGAffineTransformMakeScale (-1, 1);
    trans = CGAffineTransformRotate (trans, M_PI / 2);
    break;

  case 8:               // rotate 270
    trans = CGAffineTransformMakeTranslation (rect.height, 0);
    trans = CGAffineTransformRotate (trans, M_PI / 2);
    break;

  default: 
    break;
  }

  return trans;
}


void
jwxyz_draw_NSImage_or_CGImage (Display *dpy, Drawable d, 
                                Bool nsimg_p, void *img_arg,
                               XRectangle *geom_ret, int exif_rotation)
{
  CGImageRef cgi;
# ifndef USE_IPHONE
  CGImageSourceRef cgsrc;
# endif // USE_IPHONE
  NSSize imgr;

  CGContextRef cgc = d->cgc;

  if (nsimg_p) {

    NSImage *nsimg = (NSImage *) img_arg;
    imgr = [nsimg size];

# ifndef USE_IPHONE
    // convert the NSImage to a CGImage via the toll-free-bridging 
    // of NSData and CFData...
    //
    NSData *nsdata = [NSBitmapImageRep
                       TIFFRepresentationOfImageRepsInArray:
                         [nsimg representations]];
    CFDataRef cfdata = (CFDataRef) nsdata;
    cgsrc = CGImageSourceCreateWithData (cfdata, NULL);
    cgi = CGImageSourceCreateImageAtIndex (cgsrc, 0, NULL);
# else  // USE_IPHONE
    cgi = nsimg.CGImage;
# endif // USE_IPHONE

  } else {
    cgi = (CGImageRef) img_arg;
    imgr.width  = CGImageGetWidth (cgi);
    imgr.height = CGImageGetHeight (cgi);
  }

  Bool rot_p = (exif_rotation >= 5);

  if (rot_p)
    imgr = NSMakeSize (imgr.height, imgr.width);

  XRectangle winr = d->frame;
  float rw = winr.width  / imgr.width;
  float rh = winr.height / imgr.height;
  float r = (rw < rh ? rw : rh);

  /* If the window is a goofy aspect ratio, take a middle slice of
     the image instead. */
  if (winr.width > winr.height * 5 ||
      winr.width > winr.width * 5) {
    r *= (winr.width > winr.height
         ? winr.width  / (double) winr.height
         : winr.height / (double) winr.width);
    // NSLog (@"weird aspect: scaling by %.1f\n", r);
  }

  CGRect dst, dst2;
  dst.size.width  = imgr.width  * r;
  dst.size.height = imgr.height * r;
  dst.origin.x = (winr.width  - dst.size.width)  / 2;
  dst.origin.y = (winr.height - dst.size.height) / 2;

  dst2.origin.x = dst2.origin.y = 0;
  if (rot_p) {
    dst2.size.width = dst.size.height; 
    dst2.size.height = dst.size.width;
  } else {
    dst2.size = dst.size;
  }

  // Clear the part not covered by the image to background or black.
  //
  if (d->type == WINDOW)
    XClearWindow (dpy, d);
  else {
    jwxyz_fill_rect (dpy, d, 0, 0, 0, winr.width, winr.height,
                     jwxyz_drawable_depth (d) == 1 ? 0 : BlackPixel(dpy,0));
  }

  CGAffineTransform trans = 
    exif_rotate (exif_rotation, rot_p ? dst2.size : dst.size);

  CGContextSaveGState (cgc);
  CGContextConcatCTM (cgc, 
                      CGAffineTransformMakeTranslation (dst.origin.x,
                                                        dst.origin.y));
  CGContextConcatCTM (cgc, trans);
  //Assert (CGImageGetColorSpace (cgi) == dpy->colorspace, "bad colorspace");
  CGContextDrawImage (cgc, dst2, cgi);
  CGContextRestoreGState (cgc);

# ifndef USE_IPHONE
  if (nsimg_p) {
    CFRelease (cgsrc);
    CGImageRelease (cgi);
  }
# endif // USE_IPHONE

  if (geom_ret) {
    geom_ret->x = dst.origin.x;
    geom_ret->y = dst.origin.y;
    geom_ret->width  = dst.size.width;
    geom_ret->height = dst.size.height;
  }

  invalidate_drawable_cache (d);
}


XImage *
jwxyz_png_to_ximage (Display *dpy, Visual *visual,
                     const unsigned char *png_data, unsigned long data_size)
{
  NSImage *img = [[NSImage alloc] initWithData:
                                    [NSData dataWithBytes:png_data
                                            length:data_size]];
  if (! img) return 0;
#ifndef USE_IPHONE
  NSBitmapImageRep *bm = [NSBitmapImageRep
                           imageRepWithData:
                             [NSBitmapImageRep
                               TIFFRepresentationOfImageRepsInArray:
                                 [img representations]]];
  if (!bm) {
    [img release];
    return 0;
  }
  int width   = [img size].width;
  int height  = [img size].height;
  size_t ibpp = [bm bitsPerPixel];
  size_t ibpl = [bm bytesPerRow];
  const unsigned char *data = [bm bitmapData];
  convert_mode_t mode = (([bm bitmapFormat] & NSAlphaFirstBitmapFormat)
                         ? CONVERT_MODE_ROTATE_MASK
                         : 0);
#else  // USE_IPHONE
  CGImageRef cgi = [img CGImage];
  if (!cgi) {
    [img release];
    return 0;
  }
  int width = CGImageGetWidth (cgi);
  int height = CGImageGetHeight (cgi);
  size_t ibpp = 32;
  size_t ibpl = ibpp/4 * width;
  unsigned char *data = (unsigned char *) calloc (ibpl, height);
  const CGBitmapInfo bitmap_info =
    kCGBitmapByteOrder32Big | kCGImageAlphaPremultipliedLast;
  CGContextRef cgc =
    CGBitmapContextCreate (data, width, height,
                           8, /* bits per component */
                           ibpl, dpy->colorspace,
                           bitmap_info);
  CGContextDrawImage (cgc, CGRectMake (0, 0, width, height), cgi);

  convert_mode_t mode = convert_mode_to_rgba (bitmap_info);

#endif // USE_IPHONE

  XImage *image = XCreateImage (dpy, visual, 32, ZPixmap, 0, 0,
                                width, height, 8, 0);
  image->data = (char *) malloc (image->height * image->bytes_per_line);

  // data points at (x,y) with ibpl rowstride.

  int obpl = image->bytes_per_line;
  const unsigned char *iline = data;
  unsigned char *oline = (unsigned char *) image->data;
  int yy;
  for (yy = 0; yy < height; yy++) {
    convert_row ((uint32_t *)oline, iline, width, mode, ibpp);
    oline += obpl;
    iline += ibpl;
  }

  [img release];

#ifndef USE_IPHONE
  // [bm release];
# else
  CGContextRelease (cgc);
  free (data);
# endif

  return image;
}


Pixmap
XCreatePixmap (Display *dpy, Drawable d,
               unsigned int width, unsigned int height, unsigned int depth)
{
  if (!dpy) abort();
  char *data = (char *) malloc (width * height * 4);
  if (! data) return 0;

  Pixmap p = (Pixmap) calloc (1, sizeof(*p));
  p->type = PIXMAP;
  p->frame.width       = width;
  p->frame.height      = height;
  p->pixmap.depth      = depth;
  p->pixmap.cgc_buffer = data;
  
  /* Quartz doesn't have a 1bpp image type.
     Used to use 8bpp gray images instead of 1bpp, but some Mac video cards
     don't support that!  So we always use 32bpp, regardless of depth. */

  p->cgc = CGBitmapContextCreate (data, width, height,
                                  8, /* bits per component */
                                  width * 4, /* bpl */
                                  dpy->colorspace,
                                  dpy->bitmap_info);
  Assert (p->cgc, "could not create CGBitmapContext");
  return p;
}


int
XFreePixmap (Display *d, Pixmap p)
{
  Assert (p && p->type == PIXMAP, "not a pixmap");
  invalidate_drawable_cache (p);
  CGContextRelease (p->cgc);
  if (p->pixmap.cgc_buffer)
    free (p->pixmap.cgc_buffer);
  free (p);
  return 0;
}


static Pixmap
copy_pixmap (Display *dpy, Pixmap p)
{
  if (!p) return 0;
  Assert (p->type == PIXMAP, "not a pixmap");

  Pixmap p2 = 0;

  Window root;
  int x, y;
  unsigned int width, height, border_width, depth;
  if (XGetGeometry (dpy, p, &root,
                    &x, &y, &width, &height, &border_width, &depth)) {
    XGCValues gcv;
    gcv.function = GXcopy;
    GC gc = XCreateGC (dpy, p, GCFunction, &gcv);
    if (gc) {
      p2 = XCreatePixmap (dpy, p, width, height, depth);
      if (p2)
        XCopyArea (dpy, p, p2, gc, 0, 0, width, height, 0, 0);
      XFreeGC (dpy, gc);
    }
  }

  Assert (p2, "could not copy pixmap");

  return p2;
}


// Returns the verbose Unicode name of this character, like "agrave" or
// "daggerdouble".  Used by fontglide debugMetrics.
//
char *
jwxyz_unicode_character_name (Display *dpy, Font fid, unsigned long uc)
{
  char *ret = 0;
  NSFont *nsfont = (NSFont *) jwxyz_native_font (fid);
  CTFontRef ctfont =
    CTFontCreateWithName ((CFStringRef) [nsfont fontName],
                          [nsfont pointSize],
                          NULL);
  Assert (ctfont, "no CTFontRef for UIFont");

  CGGlyph cgglyph;
  if (CTFontGetGlyphsForCharacters (ctfont, (UniChar *) &uc, &cgglyph, 1)) {
    CGFontRef cgfont = CTFontCopyGraphicsFont (ctfont, 0);
    NSString *name = (NSString *) CGFontCopyGlyphNameForGlyph(cgfont, cgglyph);
    ret = (name ? strdup ([name UTF8String]) : 0);
    CGFontRelease (cgfont);
    [name release];
  }

  CFRelease (ctfont);
  return ret;
}


static int
draw_string (Display *dpy, Drawable d, GC gc, int x, int y,
             const char *str, size_t len, int utf8_p)
{
  NSString *nsstr = nsstring_from (str, len, utf8_p);

  if (! nsstr) return 1;

  XRectangle wr = d->frame;
  CGContextRef cgc = d->cgc;

  unsigned long argb = gc->gcv.foreground;
  if (gc->depth == 1) argb = (argb ? WhitePixel(dpy,0) : BlackPixel(dpy,0));
  CGFloat rgba[4];
  query_color_float (dpy, argb, rgba);
  NSColor *fg = [NSColor colorWithDeviceRed:rgba[0]
                                      green:rgba[1]
                                       blue:rgba[2]
                                      alpha:rgba[3]];

  if (!gc->gcv.font) {
    Assert (0, "no font");
    return 1;
  }

  /* This crashes on iOS 5.1 because NSForegroundColorAttributeName,
      NSFontAttributeName, and NSAttributedString are only present on iOS 6
      and later.  We could resurrect the Quartz code from v5.29 and do a
      runtime conditional on that, but that would be a pain in the ass.
      Probably time to just make iOS 6 a requirement.
   */

  NSDictionary *attr =
    [NSDictionary dictionaryWithObjectsAndKeys:
                    (NSFont *) jwxyz_native_font (gc->gcv.font), NSFontAttributeName,
                    fg, NSForegroundColorAttributeName,
                  nil];

  // Don't understand why we have to do both set_color and
  // NSForegroundColorAttributeName, but we do.
  //
  set_color (dpy, cgc, argb, 32, NO, YES);

  NSAttributedString *astr = [[NSAttributedString alloc]
                               initWithString:nsstr
                                   attributes:attr];
  CTLineRef dl = CTLineCreateWithAttributedString (
                   (__bridge CFAttributedStringRef) astr);

  // Not sure why this is necessary, but xoff is positive when the first
  // character on the line has a negative lbearing.  Without this, the
  // string is rendered with the first ink at 0 instead of at lbearing.
  // I have not seen xoff be negative, so I'm not sure if that can happen.
  //
  // Test case: "Combining Double Tilde" U+0360 (\315\240) followed by
  // a letter.
  //
  CGFloat xoff = CTLineGetOffsetForStringIndex (dl, 0, NULL);
  Assert (xoff >= 0, "unexpected CTLineOffset");
  x -= xoff;

  CGContextSetTextPosition (cgc,
                            wr.x + x,
                            wr.y + wr.height - y);
  CGContextSetShouldAntialias (cgc, gc->gcv.antialias_p);

  CTLineDraw (dl, cgc);
  CFRelease (dl);
  [astr release];

  invalidate_drawable_cache (d);
  return 0;
}


static int
SetClipMask (Display *dpy, GC gc, Pixmap m)
{
  Assert (!!gc->gcv.clip_mask == !!gc->clip_mask, "GC clip mask mixup");

  if (gc->gcv.clip_mask) {
    XFreePixmap (dpy, gc->gcv.clip_mask);
    CGImageRelease (gc->clip_mask);
  }

  gc->gcv.clip_mask = copy_pixmap (dpy, m);
  if (gc->gcv.clip_mask)
    gc->clip_mask =
      CGBitmapContextCreateImage (gc->gcv.clip_mask->cgc);
  else
    gc->clip_mask = 0;

  return 0;
}

static int
SetClipOrigin (Display *dpy, GC gc, int x, int y)
{
  gc->gcv.clip_x_origin = x;
  gc->gcv.clip_y_origin = y;
  return 0;
}


const struct jwxyz_vtbl quartz_vtbl = {
  root,
  visual,
  display_sources_data,

  window_background,
  draw_arc,
  fill_rects,
  gc_gcv,
  gc_depth,
  draw_string,

  jwxyz_quartz_copy_area,

  DrawPoints,
  DrawSegments,
  CreateGC,
  FreeGC,
  ClearWindow,
  SetClipMask,
  SetClipOrigin,
  FillPolygon,
  DrawLines,
  PutImage,
  GetSubImage
};

#endif // JWXYZ_QUARTZ -- entire file