<div>
  By default, Hudson builds all the possible combinations of axes exhaustively. But sometimes this is too many,
  or may contain combinations that don't make sense.
  In such a situation, you can make the matrix sparse by filtering out combinations that you don't want through
  a Groovy expression that returns true or false.

  <p>
  When you specify a Groovy expression here, only the combinations that result in <b>true</b>
  will be built. In evaluating the expression, axes are exposed as variables (with their values set
  to the current combination evaluated).

  <h4>Filtering based on values</h4>
  <p>
  For example, let's say you are building on different operating systems for
  different compilers. Assume that your slave labels are <b>label=[linux,solaris]</b> and you have
  created an axis as <b>compiler=[gcc,cc].</b>

  Any of the following expressions will filter out <b>cc</b> builds on <b>linux</b>. Depending on
  how you think about this constraint, you'll probably find some more intuitive than others.

  <center>
    <table>
        <tr>
            <td>Read "if both linux and cc, it's invalid"</td>
            <td>
                <pre>!(label=="linux" && compiler=="cc")</pre>
            </td>
        </tr>
        <tr>
            <td>Read "for a combination to be valid, it has to be either on solaris or on gcc."</td>
            <td>
                <pre>label=="solaris" || compiler=="gcc"</pre>
            </td>
        </tr>
        <tr>
            <td>Read "if on Solaris, just do cc"</td>
            <td>
                <pre>(label=="solaris").implies(compiler=="cc")</pre>
            </td>
        </tr>
    </table>
  </center>

  <h4>Sparsening of the matrix</h4>
  <p>
  In addition to the specific filtering rules based on values, one can also use a special
  variable "<tt>index</tt>", which can be used to sparsen the matrix.

  <p>
  For example, <tt>index%2==0</tt> would cut the matrix size in half by removing one
  combination per every 2 combinations, in such a way that the coverage is still reasonable.
  Similarly, <tt>index%3!=0</tt> would cut the matrix size into 66% by throwing
  away every 1 out of 3 combinations.
 <p>
 It's possible to use Project cascading feature for this property. Please review <a href="http://wiki.hudson-ci.org/display/HUDSON/Project+cascading">
 this document</a>.
</div>