How should one deal with coordinate spaces with non-affine transforms?

[coderedart]

So, looking at the API, it seems like nodes are expected to be axis-aligned rects with position + size. How should GUI toolkits expose:

• rotated widgets - where a widget + its children are rotated, breaking the axis alignment of widget rectangles (bounding boxes?). eg: 45 degree rotation, making the widget go from left-bottom to right-top.
• 3D widget - viewport-like widgets often have controls in 3D space, where widgets are not rectangle-shaped (an interactive sphere) or exist in awkward orientations.
• non-rect widgets - Let's say a profile picture widget, which is shaped like circle.

In all these cases, accesskit_consumer's node_at_point fn may get confused, because the bounds are often weird shapes from the 2D window perspective.

Off-topic, but is there an example that showcases how to use the API for a widget like text (or text-input). Is a single paragraph widget expected to be broken into one or more accesskit-nodes with TextRun role like:

• each line of text is a separate TextRun node
• each line of text with different formatting (eg: foreground color or strikethrough) splits into multiple TextRun nodes?

If so, how should we deal with focus? Do we focus these individual TextRun nodes or do we just maintain focus on the parent Label/TextEdit widget?

[mwcampbell]

Thanks for working with AccessKit. I'll try to answer your questions.

For widgets whose bounding shape is a rotated rectangle, an affine transform should work fine. If you apply the transform to the top-level rotated widget, it will apply to its children too. Please let me know if I'm missing something, or if you've found a problem with this approach.

We don't have a good solution for 3D or non-rectangular widgets. Currently the best you can do is set the bounds to the tightest rectangle that encloses the widget.

You have the right idea about text input. The de-facto reference implementation of how to do this is currently in parley.

[coderedart]

Thanks for the quick response. As the Affine struct only had scale and translate constructors, I just assumed that rotation was not supported.