Hi Anton,
the fix looks good to me.
BTW you can use GetBooleanAction instead of GetPropertyAction.
Thanks,
Alexander.
On 07/01/2014 07:22 PM, Anton V. Tarasov wrote:
Hello,
Please review a fix:
bug: https://bugs.openjdk.java.net/browse/JDK-8048887
webrev: http://cr.openjdk.java.net/~ant/JDK-8048887/webrev.0
The problem is that SortingFocusTraversalPolicy uses ROW_TOLERANCE
conception to alter a disposition of components in a focus cycle for
the sake of visually more appropriate traversal order. This however
breaks the transitivity rule, which in case of using the default
tim-sort algo leads to an exception.
We've got a number of complaints filed against this issue, so I
eventually agreed to use reflection as suggested in JDK-7075600.
I've tested performance in the following way. In a jframe I was adding
up to 10.000 components a) at random location (with frame's null
Layout) so that to completely untie traversal order from container
order (hard case for the sort), and b) sequentially, so that to have
the traversal order pre-sorted. Then I counted time spent in the
sorting procedure. The results of the legacy merge-sort and the
default tim-sort were pretty close. So, the fallback to the legacy
sorting algo shouldn't hit performance for Swing.
Just for the case of any unpredictable issue (and for the testing
purpose), I left an ability to switch to the default tim-sort via
setting the (undocumented) "swing.legacySortingFTPEnabled" property to
false.
Regards,
Anton.
