Allow open/interact without plotting anything

[chaosphere2112]

Currently, the canvas.open() call doesn't do anything. In developing and testing my GUI parts embedded in VCS, it's really quite handy to have a fully-prepared VCS canvas without actually plotting anything (or plotting something and then calling clear(), which gives the same end result). This PR makes it so you can open the canvas prior to plotting. As a caveat of that, if you call x.open(); x.interact(), it'll now enter interact mode. This is a little goofy, but given the fact that we have some GUI elements that allow for some user interaction with a blank canvas (markers and text buttons, top right), it's not completely ridiculous. I'm happy to scrap the ability to interact() with a blank canvas; it's not particularly useful for me.

All of the tests pass for me with this PR, though there is one test that will hang until manually quit (test_vcs_interact_no_plot.py). If we make it so interact() will fail when there are no plots, that test can remain the same; if we don't, this test will either have to change or be eliminated.

@doutriaux1 @aashish24 @dlonie Opinions?

[doutriaux1]

@chaosphere2112 do you mean your changes make it fail, right? I have no strong opinion either way whereas we can call x.interact () on a close canvas or not. Just let make sure we update the tests and let the users know about any change.

[chaosphere2112]

They make it fail, though it is not flagged by ctest as failing.

[alliepiper]

The test will need to return a non-zero exit status for CTest to pick up the failure. Can you update the test so this will happen?

The changes sounds good. I think being able to call interact on an open, empty canvas makes sense for the reason you mentioned (GUI widgets).

[chaosphere2112]

@dlonie The only stumbling block I'm seeing is that if we're going into interactive mode, there's no obvious way to exit it automatically in a test. The only things that come to mind are threads, but I think it would be pretty non-deterministic as to when the test would actually come to an end, because in my experience VTK tends to hog the GIL. Are there any VTK tests for Start that we could emulate?

[alliepiper]

@chaosphere2112 I've done similar things in the past by starting a timer in the eventloop or adding an observer for a signal to close things down.

Does UVCDAT have some sort of signal/slot or event/observer mechanism? If there's some way to register a callback function to be executed when the interaction starts, we could shut it down from there. Otherwise, look at what is managing the UVCDAT interaction event loop and see if there's a way to post a custom event/callback to it.

[chaosphere2112]

Oh, yeah, that would be easy; just signed up for the StartEvent event on the interactor before starting it.
Here's the test results:
https://open.cdash.org/buildSummary.php?buildid=3782918

[alliepiper]

Nice! Looks good. Is this ready to go in?

[chaosphere2112]

I think it's ready; @doutriaux1 Mind running the test and making sure it works in ubuntu?

[doutriaux1]

works for me at:
https://open.cdash.org/buildSummary.php?buildid=3783146