meshfill plot broekn with bg=0

[doutriaux1]

the following script creates two different png depending if we use bg = True or not

bg=True leads to the correct output in bg mode, and incorrect in non bg mode

See attached pngs first is good and second is bad

import cdms2,vcs

bg=True

f=cdms2.open("data_portrait.nc")
s=f("variable_3")       
M=f("variable_115")

x=vcs.init()        
y=vcs.init()
x.scriptrun("mesh_bug.json")
x.setcolormap("bl_rd_12")
m=x.getmeshfill("portrait")
x.plot(s,M,m,bg=bg)
x.png("mesh_%i" % bg)
y.plot(s,M,m,bg=not bg)
y.png("mesh_%i" % (not bg))

now interestingly enough turning the bg to False leads to a core dump following this output:

radeon: Failed to allocate a buffer:
radeon:    size      : 1 bytes
radeon:    alignment : 1 bytes
radeon:    domains   : 2
radeon: Failed to allocate a buffer:
radeon:    size      : 1 bytes
radeon:    alignment : 1 bytes
radeon:    domains   : 2
Segmentation fault (core dumped)

[aashish24]

@doutriaux1 what machine is this? Is this remote or local?

[doutriaux1]

data and json and script can be found at: https://github.com/doutriaux1/VTKLearning/tree/master/uvcdat_bugs

@dlonie I will need your help on this one I think.

[doutriaux1]

it is on my local ubuntu. Confirmed by other users (@durack1) on RH6

[alliepiper]

Do you both have the same graphics card? The radeon errors make me wonder if this is hardware related...

[aashish24]

what driver you are using @doutriaux1 We should be able to re-produce this bug if it is not hardware specific. @dlonie can you try the script and the data?

[alliepiper]

Sure. It'll be a bit before I can get to it, but sometime today.

[durack1]

I don't think this has anything to do with hardware, as it happens both on a real and virtual machine..

[alliepiper]

Yep, it's not hardware, I can reproduce this on my nVidia system.

[doutriaux1]

@dlonie I don't think it's hardware related either happens on Ubunut/RH/Mac and a VM I would be really surprising if all these computers had the exact same graphic card (although I guess it is possible)

[durack1]

One of the concerning issues too, is that exporting to different formats yields completely different graphics.. So pngs seem to look fine, thought pdf and ps files have labels rotated differently and this difference is concerning..

[alliepiper]

That is a known issue unrelated to this one, see #503.

[durack1]

Yeah thanks @dlonie I was aware there was already an issue open on that.. Was just too lazy to link to it..

[alliepiper]

The problem seems to be triggered by the call to configureEvent() (since bg has no interactor, configureEvent never gets called). Still tracking down the specifics...

[alliepiper]

Removing the call to self.canvas.clear() in configureEvent (VTKPlots.py 173) fixes the issue. Is this call needed?

(note, I also had to add y.scriptrun("mesh_bug.json");y.setcolormap("bl_rd_12") to the test_bg.py script to get colors correct).

[doutriaux1]

let me look into this, I think if we don't do a clear then plots never get cleared in the GUI and they get drawn on top of each other. The real question I guess is why do we get half a plot when calling clear.

[alliepiper]

You're right, the clear does need to be there for the GUI. I'm guessing that the configureEvent is being called at a bad time, possibly while in the middle of building up the scene.

[doutriaux1]

I think this has been fixed by #873