Update README.md

README.md

@@ -149,7 +149,7 @@ I had been enjoying the ergonomics of SwiftUI, but SwiftUI simply can't handle b
 
 Not so long ago, before programmable shaders, it was necessary to cache parts of a UI in textures (CoreAnimation for example does this) to get good performance. Now we have extremely fast GPUs and such caching is not necessary to achieve good performance. In fact if enough is animating, lots of texture caching can hinder performance, since the caches need to be updated so often. Plus, the textures consume a fair amount of memory, and when you have an unbounded node-graph like Audulus, that memory usage would be unbounded. And what resolution do you pick for those textures?
 
-So rui starts from the assumption that 2D UI graphics (not general vector graphics!) are a trivial workload for a GPU. If you consider how advanced games are now, doing realtime path-tracing, this seems intuitively correct, but Audulus more-or-less proves it. So that means we can do away with the texture caching, and we really might not even need damage regions either. I'm also skeptical of the need for parallel encoding or caching parts of the scene for 2D UI graphics, since, again, it's just a trivial GPU workload.
+So rui starts from the assumption that 2D UI graphics (not general vector graphics!) are a trivial workload for a GPU. If you consider how advanced games are now, doing realtime global illumination and such, this seems intuitively correct, but Audulus more-or-less proves it. So that means we can do away with the texture caching, and we really might not even need damage regions either. I'm also skeptical of the need for parallel encoding or caching parts of the scene for 2D UI graphics, since, again, it's just a trivial GPU workload.
 
 Layout, on the other hand, can't easily be offloaded to GPU free-performance land. It's necessary to cache layout information and try not to recompute it all the time. So rui caches layout and only recomputes it when the state changes (unlike a typical immediate mode UI which computes layout on the fly and is constrained to very simple layouts). For Audulus, this isn't quite enough, since some view-local state will be changing all the time as things are animating (Audulus solves this by only recomputing layout when the central document state changes). Perhaps this is where proponents of DOM-ish things (some other OOP-ish tree of widgets) would jump in and make their case, but I'm skeptical that's really necessary. Think of what actually needs to be (re)computed: a layout box for each (ephemeral) View, does this really require a separate tree of objects? Time will tell!