On 11/7/19 7:48 pm, Prasanta Sadhukhan wrote:
On 08-Nov-19 9:07 AM, Sergey Bylokhov wrote:
On 11/7/19 7:32 pm, Prasanta Sadhukhan wrote:
Hi Sergey,
OK. Since the test pass in local ubuntu18.04 system and fail only in mach5
linux system, I thought it's a test issue. Do you know what window manager our
mach5 linux system is running?
I guess it is gnome3, BTW I am able to reproduce it as well on Ubuntu 19.10,
after a number of iterations(~50) and can confirm that when it fails the size
of the window is changed(it became smaller than initially).
OK. I will also try with more iterations on 18.04. But, one thing is test is
calling removeNotify,addNotify which are not supposed to be called directly
from programs. Shouldn't we use allowable APIs maybe setVisible(false),
setVisible(true)?
I guess in this particular test it is not so important which API to use, the
only difference is in the timings.
Regards
Prasanta
Regards
Prasanta
On 08-Nov-19 2:42 AM, Sergey Bylokhov wrote:
Hi, Prasanta.
I think that the test passed after the fix is because it does not have enough
time to update
the location of the Frame by the native callback.
Before the fix we have this sequence of calls:
1 Sets the bounds of the frame(this bounds are cached in the frame and
returned by getX/getY)
2 Make the frame visible
3 Wait while the callback from the native change the location of the frame
cached at step 1
4 Check the coordinates
5 Destroy/Recreate the peer
6 goto step 3
After the fix:
1 Sets the bounds of the frame(this bounds are cached in the frame and
returned by getX/getY)
2 Make the frame visible
3 Check the coordinates cached at step 1
4 Destroy/Recreate the peer
5 goto step 3
So actually the new test did not check location of the frame but check the
values cached in the Frame object.
On 11/6/19 3:57 am, Prasanta Sadhukhan wrote:
Hi All,
Please review a fix for an issue where it is seen the frame location is
sometimes wrong in mac on mach5 headful nightly run.
It seems to be a timing issue as it shows 10 frames by calling
frame.removeNotify(),frame.addNotify() repeatedly, which by the way are not
supposed to be called by programs directly.
Proposed fix is to make not to make thread sleep every time
removeNotify,addNotify is called so that there is no delay. I have ran mach5
job (in JBS) for 3 consecutive runs on all 3 platforms and they pass.
Bug: https://bugs.openjdk.java.net/browse/JDK-8233703
webrev: http://cr.openjdk.java.net/~psadhukhan/8233703/webrev.0/
Regards
Prasanta
--
Best regards, Sergey.
