[cairo] Bug: more grief in 1.9.8

[cu]

 I downloaded from git the updated cairo-image-surface.c, recompiled and
indeed the memory corruption (and a few other issues) went away.

 However, here is a new one. Switching on ANTIALIAS_NONE causes this
polygon to be improperly filled (note that both should be the same - I
am filling the same outline). If you take out the last set of X/Y
(marked with "comment this out") then in ANTIALIAS_NONE mode the fill is
not being performed at all.

I appreciate the work you guys do, no question here. That said, FWIW, my
experience with 1.9.8 so far has not been too great. I realize it is a
snapshot, but I don't recall having so many issues with a snapshot
previously, especially since 1.9.8 is a release candidate and supposed
to be fixing a number of issues.

----------------------- Cut Here -----------------------

#include <stdio.h>
#include <stdlib.h>
#include <stdint.h>
#include <string.h>
#include <math.h>
#include <cairo.h>

struct xy_st {
   double x;
   double y;
};

struct xy_st arrXY[] = {
 { 627.016212, 221.749777 },
 { 756.120787, 221.749777 },
 { 756.120787, 557.602766 },
 { 626.952721, 557.602766 },
 { 626.548456, 493.315729 },  // comment this out for another one
 };

int main()
{
   int i, num;
   cairo_surface_t   *surf;
   cairo_t  *cr;
   double w, h, x, y, offt_x, offt_y;

   w = 1000;
   h = 800;
   surf = cairo_image_surface_create(CAIRO_FORMAT_ARGB32, w, h);

   cr = cairo_create(surf);

   num = sizeof(arrXY) / sizeof(arrXY[0]);
   for (i = 0; i < num; ++i) {
      x = arrXY[i].x;
      y = arrXY[i].y;

      if (i == 0)
         cairo_move_to(cr, x, y);
      else
         cairo_line_to(cr, x, y);
   }
   cairo_close_path(cr);

   cairo_set_source_rgba(cr, 1, 0, 0, 1);
   cairo_fill_preserve(cr);

   cairo_set_antialias(cr, CAIRO_ANTIALIAS_NONE);
   cairo_set_source_rgba(cr, 0, 1, 0, 1);
   cairo_fill(cr);

   cairo_destroy(cr);

   cairo_surface_write_to_png(surf, "surf.png");

   return 0;
}