OpenGL Widget: Proof of concept

[corranwebster]

This is a first cut at seeing what an OpenGLView widget might look like. Goals for this are:

• to put something out there so people can try it out
• to work out how to get an OpenGL view up and running on each platform and do something trivial
• to start experimenting with the interface to see what that would look like as part of the design process

Currently supports:

• Android
• Cocoa
• Gtk
• iOS
• Qt
• Web
• Winforms

Ref #51

PR Checklist:

• All new features have been tested
• All new features have been documented
• I have read the CONTRIBUTING.md file
• I will abide by the code of conduct

[freakboy3742]

Flagging that this file needs updating

[freakboy3742]

If we have an OpenGL example app that doesn't draw a teapot, I'm going to be very disappointed... 😝

More seriously - I presume this is "first light" for the API, rather than anything close to a final API?

[corranwebster]

Yes, very much - this is in the 'investigation' phase - the focus so far is on "what does it involve to get an OpenGL view on various platforms", which I think I've mostly got a handle on, so I'm now moving on to "what are the differences/commonalities between the platform OpenGL implementations."

I've been making notes and recording thoughts in #51.

Edit: I'm at the "how do I draw a cross-platform triangle" stage. But yes, there should be a teapot at some point.

[corranwebster]

I've run into a serious problem:

• testbed tests do not run in the main thread
• OpenGL widgets on both MacOS and Qt do not like this (segfaults on both platforms in tests). The Qt error explicitly says "Cannot make QOpenGLContext current in a different thread"

I don't think we can test OpenGL stuff using testbed.

[freakboy3742]

I've run into a serious problem:

• testbed tests do not run in the main thread
• OpenGL widgets on both MacOS and Qt do not like this (segfaults on both platforms in tests). The Qt error explicitly says "Cannot make QOpenGLContext current in a different thread"

Hrm... the test suite executes in a separate thread, but the actual app should be entirely on the main thread, and the way pytest-asyncio operates, all the test cases should be "run_coroutine_threadsafe" on the main event loop...

[corranwebster]

Hrm... the test suite executes in a separate thread, but the actual app should be entirely on the main thread, and the way pytest-asyncio operates, all the test cases should be "run_coroutine_threadsafe" on the main event loop...

A little bit more fiddling, and it's the fixtures which are the problem. In particular, the widget fixtures create the widget off of the main thread and then it seems like they get passed to the main thread as parameters of the actual test.

By avoiding passing the widget in as a fixture I can stop the segfaults. Still getting test failures on cocoa, but not hard crashes.

Edit: and if I make the widget fixture async that seems to fix the problem as well - I guess an async fixture defers the creation and gets run on the main thread. It all feels a bit fragile, but hopefully it'll work now.

[corranwebster]

I'm going to close this in favour of #4279. There are things in this branch which may eventually make their way into Toga, but I think it needs to be done in more organised steps.

[johnzhou721]

Hrm... the test suite executes in a separate thread, but the actual app should be entirely on the main thread, and the way pytest-asyncio operates, all the test cases should be "run_coroutine_threadsafe" on the main event loop...

A little bit more fiddling, and it's the fixtures which are the problem. In particular, the widget fixtures create the widget off of the main thread and then it seems like they get passed to the main thread as parameters of the actual test.

By avoiding passing the widget in as a fixture I can stop the segfaults. Still getting test failures on cocoa, but not hard crashes.

Edit: and if I make the widget fixture async that seems to fix the problem as well - I guess an async fixture defers the creation and gets run on the main thread. It all feels a bit fragile, but hopefully it'll work now.

Yes this was my advice on the Qt tree view issue. async functions are ran — the way it happens is it’s created as a task, which is explicitly forwarded by us to main thread in our testing setup.