Libui: Master issue for Windows redraw performance woes

[andlabs]

Currently libui on Windows fakes transparency by drawing the parent control onto each child control as a background each time, resulting in awful performance. Not doing this with the current implementation of uiControl and control hierarchies results in controls indiscriminately having a default background color. I have two options: flatten the control hierarchy or split the work between WM_ERASEBKGND and WM_CTLCOLOR*. @lxn figured a lot of this out.

This issue will combine all existing issues on the matter.

This issue includes and continues the discussion of andlabs/ui#184

[andlabs]

Huh, that was fewer issues than I thought there were...

[kdar]

I tried libui a year ago on win10 64bit and had flickering issues. Seems to still be the case.

[andlabs]

Yes, that is what we are trying to resolve here. The question is how, and then I need to actually implement the fixes. lxn had a suggestion in the issue I linked in the first post, but I have yet to try it.

[msink]

So now - after 0.4 - most annoying is this issue.

Sometimes window opening takes 5 and more seconds, as in StapleButter/melonDS#220

[andlabs]

Yes, and addressing this issue is what's in store for Alpha 5's changes. I still need to figure out what that simpler solution lxn used was first, but there are several other advantages to just flattening the control hierarchy.

[msink]

If it should be almost fully rewritten anyway - maybe not invent a wheel and use Yoga as layout engine?
Is it possible?

[tom-corwin]

@msink coming from a C# perspective, the C# bindings for yoga are very hacky, requiring (at least) 2 extra libraries on top of the actual Yoga library, plus a whole separate framework (Xamarin) for it to be used on macOS and Linux platforms. I think it would be more or a hinder than a help for alot of the bindings, though I can't speak for the others.

[msink]

Well - I mean not adding it as dependency, but include in libui as library.
Or at least implement similar interface and borrow some code.
Because there are WYSIWYG editors for it - would be great to reuse existing tools and infrastructure.

[tom-corwin]

Well - I mean not adding it as dependency, but include in libui as library.

My apologies. I get your point now :)

Or at least implement similar interface and borrow some code.
Because there are WYSIWYG editors for it - would be great to reuse existing tools and infrastructure.

I agree that it would be great to reuse existing tools and code to accomplish this.

[mischnic]

This should get a very high priority.
I noticed this again during development in the tester. Moving (or even worse resizing) the window with page 1 open gives me the feeling that my PC is potato.

[mleise]

I didn't realize performance could be an issue with a toolkit using native widgets. Best of luck with the solution! I see WM_PRINTCLIENT was already tried and @lxn seems confident that a more performant solution can be implemented.