Clarify that round() returns nonzero positive integers; mark round() vs ceil() as issue.

Author: fantasai

css3-background/Overview.html

@@ -22,8 +22,7 @@
 
    <h1>CSS Backgrounds and Borders Module Level 3</h1>
 
-   <h2 class="no-num no-toc" id=longstatus-date>Editor's Draft 07 October
-    2009</h2>
+   <h2 class="no-num no-toc" id=longstatus-date>Editor's Draft 7 October 2009</h2>
 
    <dl>
     <dt>This version:
@@ -1452,9 +1451,29 @@ <h3 id=the-background-size><span class=secno>3.9 </span>The &lsquo;<code
     <var>W</var> is the width of the background positioning area, then the
     rounded width <var>X'</var> = <var>W</var> / round(<var>W</var> /
     <var>X</var>) where round() is a function that returns the nearest
-    natural number.
+    natural number (integer greater than zero).
   </blockquote>
 
+  <p class=issue>This will round to the nearest whole number of tiles, which
+   keeps closer to the intended size and, in the case where one dimension is
+   fixed (e.g. in &lsquo;<code class=property><a
+   href="#border-image">border-image</a></code>&rsquo;), keeps the image
+   closer to the intended aspect ratio. This is almost certainly the best
+   solution for vector images and high-resolution raster images. However, if
+   the given image is a low-resolution raster image, it will require
+   interpolating pixels, which can look bad. See <a
+   href="http://www.bradclicks.com/cssplay/border-image/round-test1.html">Rounding
+   Extremes</a> for illustrations. The workaround is to specify a
+   higher-resolution image (e.g. by shrinking with &lsquo;<code
+   class=property><a
+   href="#background-size">background-size</a></code>&rsquo; or &lsquo;<code
+   class=property><a
+   href="#border-image-width">border-image-width</a></code>&rsquo;). Possible
+   spec solutions include introducing a separate keyword that always scales
+   down, or to change the algorithm so that we force scaling down when
+   interpolation would be required for scaling up. The CSSWG is looking for
+   feedback on this issue.
+
   <p>If &lsquo;<code class=property><a
    href="#background-repeat">background-repeat</a></code>&rsquo; is
    &lsquo;<code class=css>round</code>&rsquo; for one dimension only and if

css3-background/Overview.src.html

@@ -1009,9 +1009,25 @@ <h3 id="the-background-size">The 'background-size' property</h3>
   and <var>W</var> is the width of the background positioning area,
   then the rounded width
   <var>X'</var> = <var>W</var> / round(<var>W</var> / <var>X</var>)
-  where round() is a function that returns the nearest natural number.
+  where round() is a function that returns the nearest natural number
+  (integer greater than zero).
 </blockquote>
 
+<p class="issue">This will round to the nearest whole number of tiles,
+which keeps closer to the intended size and, in the case where one
+dimension is fixed (e.g. in 'border-image'), keeps the image closer to
+the intended aspect ratio. This is almost certainly the best solution
+for vector images and high-resolution raster images. However, if the
+given image is a low-resolution raster image, it will require interpolating
+pixels, which can look bad. See
+<a href="http://www.bradclicks.com/cssplay/border-image/round-test1.html">Rounding Extremes</a>
+for illustrations. The workaround is to specify a higher-resolution image
+(e.g. by shrinking with 'background-size' or 'border-image-width').
+Possible spec solutions include introducing a separate keyword that always
+scales down, or to change the algorithm so that we force scaling down when
+interpolation would be required for scaling up. The CSSWG is looking for
+feedback on this issue.</p>
+
 <p>If 'background-repeat' is ''round'' for one dimension only and if
 'background-size' is ''auto'' for the other dimension, then there is a third
 step: that other dimension is scaled so that the original aspect ratio