Roadmap
Work that is being built now, and work that is planned. None of it is in the engine yet. Each item moves into the rest of these docs the day it merges.
Nothing on this page is available yet. Names, calls and behaviour may change before they land. Do not build against them.
In progress
Nothing is being built right now. The next items come from the planned list below.
Planned
Not started yet.
| Work | What it adds | Status |
|---|---|---|
| Render policy modes | One setting for how the engine renders: minimal, immediate, retained and auto. Minimal runs on the CPU only. | Planned |
| Memory optimization, pass 2 | Goes with the render policy modes. It covers:
| Planned |
| Memory vs speed optimization mode | A runtime choice between keeping assets resident for speed and freeing memory to recompute on demand. In memory mode the engine drops cached textures, decoded animation frames, generated images, CPU copies and offscreen demo assets, and regenerates each one when it is needed again. It goes with the render policy modes, minimal in particular, and memory pass 2. It builds on techniques proven on an earlier CPU renderer, which ran the eq app in about 1.8 MB:
| Planned |
| Platform native menus | Menus the operating system draws, modelled on Electron's Menu API. It covers:
| Planned |
| Hot reload in dev builds | The app view as a reloadable module while state lives in the host, so popovers, scroll, focus and text edits survive, plus asset, shader and deck reload. The target is under 300 ms per edit, and release builds stay free of it. | Planned |
| Deck on the layout engine | The deck language lays out slides with a subset of the layout engine, and the baker links the C++ engine instead of its own solver. | Planned |
| Live component previews | Live component previews in these docs through a wasm build of the engine, mounting the same demos the gallery shows. | Planned |
| First real RTX 2050 Vulkan run | The whole engine measured on the event GPU, the transparent window route included. | Planned |
| Platform code behind engine services | The eq and graph apps move their own platform code behind engine services. | Planned |
| More overlay layers | Lifting the overlay layer limit. Today past 14 open overlays the rest share the last layer. | Planned |
| Vector effects (elba) | Group effects in the vector engine. | Planned |
Known gaps
- elba declares linear blending, a GPU backend, float precision, and path trim, stroke and offset, but they are not built yet: linear and float fall back, the GPU choice renders on the CPU, and the three path calls return
elba_not_implemented. See the vector engine. - A semi transparent polyline that folds back within its own stroke width still draws the overlap twice. Joined segments otherwise cover each pixel once. See the draw list.
- Transparent windows are measured on the Metal and CPU routes. The Vulkan route on the RTX 2050 is not verified yet. See transparent windows.
ui::search_boxhas no live mode yet: results follow Enter, not every keystroke. See text input.- The gallery capture of the image button comes out blank, a pill with no label, so it has no preview picture on these docs.
- A popover taller than the view draws nothing. See popovers.
- Switching between menubar menus shows both panels for one frame. See menus.
- The gallery grid runs all 100 demos live, so it holds about 240 MB on the Windows CPU backend. See the gallery.
- Icon atlases are loaded more than once. A fix is coming.
- The boot splash bar does not scale with DPI yet. Everything else follows the display scale.