Apex

Apex is just a simple proof of concept to demonstrate how easily you can build your own UI Toolkit
from scratch. This code base is most likely full of bugs and design mistakes but it should help
you understand the basics of a UI Toolkit. It is meant as a learning and demonstration tool only.

Tested only in an emulator and only on API level 31.

Concepts

Widgets are called elements and are all instances of the Element class. Apex elements are not
intended to be subclassed, instead you build widgets by adding components to an element. Each
component has a single responsibility: layout, rendering, input event, or whatever else you want.
Components can be anything, and in the current codebase some are classes, some are interfaces, some
are enums.

For instance a Button is an Element with the following components:

  • A ButtonModel (text, click listener, etc.), the public API of a button
  • A RenderComponent, to render the button
  • A LayoutComponent, to compute its own size and position the text
  • A MotionInputComponent, to react the touch events and handle clicks
  • An InternalState, to track the pressed state of the button

Apex also offers Provider instances, which are roughly equivalent to Jetpack Compose’s composition
locals. The give access to global data throughout the tree: Resources, display density, the
current theme, etc. Any Element can inject new providers or override existing providers by using
the ProviderComponent component. MainActivity shows an example of using a ThemeProvider to
modify the current theme.

Exercises for the reader

If you’d like to play with this codebase a bit, here are a few things you could try:

  • Optimize components lookup. Right now, every lookup iterates over a flat list. It’s not a big deal
    since most elements will have a short list but this could be improved. Since it’s intended that an
    element can own multiple components of the same type, you’d probable a data structure that maps
    component types to a list (a linked hashmap for instance)
  • Optimize providers handling. Every layout/render/motion input phase currently re-applies the
    providers. It’s not very efficient. And the layout phase doesn’t correctly apply the providers at
    every level of the tree
  • Take the MotionInputComponent from Button and make it a generic, reusable API so you can
    perform clicks on the Image widget in MainActivity as well
  • Track changes in data models to re-layout/re-draw only when needed
  • Don’t relayout/redraw on every v-sync. It’s wasteful
  • Reduce memory allocations (esp. generated by the many RectF and SizeF instances, among
    other things)
  • Cleanup the inline/noinline/crossinline and reified generic mess in the various helper functions
  • Make this a multi-platform UI Toolkit! Remove Android-specific APIs (Canvas, Bitmap, etc.)
    and use your own abstractions. For rendering use skiko

Screenshot

Not super exciting, but here it is:

Apex demo: a photo centered on screen with two buttons below, Previous and Next

License

See LICENSE.

GitHub

View Github