Today I changed the performance test that measures summary table
scrolling speed.

Previously the test did a single jump of 25 units, followed by
widget.Update(). Current test scrolls 1 unit 25 times, each time
followed by wx.Yield().

The motivation for this change was that even though the table scrolling
performance has improved a lot recently, our test was not noticing it.
Partly this is because the first unit scroll speed did not change
significantly, and possibly because we were doing the table update
rather than the yield.

The added benefit of scrolling more than one unit is that if you run the
test while you are watching it you can see how smooth the scrolling
feels like. I also experimented on a very fast box by artificially
introducing slowdowns to see what time would be acceptable (=smooth)
scrolling performance. In my experiments I determined that 1 seconds
feels smooth, but 1.5 seconds does not. Therefore I set the target for
Preview as 1 second.

It seems our reference Intel Mac and Linux platforms are clearly under
the 1 s limit. I think there is a bug on PPC Mac that makes the test
work incorrectly and under report the times (was there before this
change as well). Also I am suspicious of the Windows time, since it now
reports the test passes faster even though it should be slower. I'll be
monitoring the tests.

There is still one bug that we'd like to fix regarding this, and it is
specific to Windows. The problem is that on Windows we paint too much.
We have several paint problems in the app so it is possible that by
fixing this we could fix other performance issues caused by excessive
painting.

Of course, since the test changed it does not make sense to compare the
new numbers to what the old test was measuring.

-- 
  Heikki Toivonen



Attachment: signature.asc
Description: OpenPGP digital signature

_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

Open Source Applications Foundation "chandler-dev" mailing list
http://lists.osafoundation.org/mailman/listinfo/chandler-dev

Reply via email to