This has come up in a real PDF (unfortunately I no longer remember which one it
was) where image->n + image->a was 0 which caused a crash in renderimage()
because we do:
tileheight = 512 * 1024 / (image->w * (image->n + image->a));

which is division by 0.

A fix is below. I believe it's better to return fz_okay in this case instead of
throwing an error because that way we'll show the remaining of the page (as
opposed to probably not rendering the page at all).

$ darcs diff -u
diff -rN -u old-mupdf-official/fitzdraw/render.c
new-mupdf-official/fitzdraw/render.c
--- old-mupdf-official/fitzdraw/render.c        2009-04-12 22:43:00.546875000 -0700
+++ new-mupdf-official/fitzdraw/render.c        2009-04-12 22:43:00.796875000 -0700
@@ -413,6 +413,9 @@
         if (image->w == 0 || image->h == 0)
                 return fz_okay;

+       if (image->n + image->a == 0)
+               return fz_okay;
+
        calcimagescale(ctm, image->w, image->h, &dx, &dy);

        /* try to fit tile into a typical L2 cachce */