vectorcraft/docs/hardware-parity.md

# Hardware parity

> **Last reviewed:** 2026-10-10 · **Last updated:** 2026-10-30 · **Change:** patch (a second finger cancels the tool press the first finger began, and a finger landing while another is down presses nothing; the pinch only navigates, #4198) · **Target:** Adobe Illustrator 2026 (30.x)

What hardware Illustrator uses, per platform, against VectorCraft. Illustrator's side is from its public
documentation (GPU Performance, touch workspace, pen pressure pages); ours from the code
(`crates/ui-egui/src/canvas.rs`, `touch.rs`, `graphics.rs`, `crates/render`). Part of
[target-app-parity.md](target-app-parity.md).

## Summary

**~35% (estimated), 25–45 h.** VectorCraft rasterizes the canvas on the CPU (multithreaded `vello_cpu`) and uses
the GPU only to composite, so any GPU that can show a window works, at the cost of GPU-accelerated preview
on huge documents. VectorCraft reads pen pressure on Windows, on the web, on macOS, and on Linux under X11 and XWayland, and it
reads the pen's eraser end on macOS, on Linux under X11 and XWayland, and on the web when it presses.
reads the pen's eraser end on Windows, on macOS and on Linux under X11 and XWayland.

| Feature | Illustrator | VectorCraft macOS | Windows | Linux | Web | Status / hours |
|---|---|---|---|---|---|---|
| GPU-accelerated canvas (GPU Performance, GPU Preview, GPU anti-aliasing) | Metal, DirectX | CPU raster, GPU composite (wgpu/Metal) | CPU raster, GPU composite (DX12, OpenGL fallback) | CPU raster, GPU composite (Vulkan/GL) | WebGPU, WebGL2 fallback | partial; a GPU renderer spike is pending (M5): 15–25 h |
| Multithreaded rendering off the UI thread | partial | yes | yes | yes | single-threaded | beyond Illustrator |
| Hybrid-graphics GPU choice | automatic | power-saving GPU by default, Preferences › Graphics Processor | same; DX12 then OpenGL; no Vulkan unless asked (#545, #806) | the desktop's GPU | browser | done |
| Restart on another GPU when the first fails to present | — | yes (#502, #964) | yes | yes | restart on lost device (#369) | done |
| Pen pressure | yes (Wacom, Windows Ink, macOS tablet events) | yes, read from AppKit's tablet events before the window gets them (winit reports none), each move of a stroke with its own pressure, mouse event coalescing off (#852; not yet tried on a Mac) | yes (pointer force via `WM_POINTER`) | under X11 and XWayland (Preferences › Performance › Linux Display Server, #1151, #2028), read from XInput2 raw events (#764); **no** on native Wayland, where winit reports none and the pen may not reach the window (#491) | browser pointer events, with coalesced events; Apple Pencil on iPad (hover, palm rejection: fingers navigate after a pen contact until reload) | partial: 2–4 h for Wayland (winit 0.31), plus a check on a Mac |
| Pen tilt, bearing, barrel rotation (6D Art Pen) | yes, drives brushes | no | no | no | no | missing: 4–8 h after the input layer (#372) |
| Pen eraser end | yes | the eraser end (AppKit's tablet proximity events) selects the Eraser tool and the tip brings back the tool used before, as on Linux (#852; not yet tried on a Mac) | no | under X11 and XWayland, the eraser end selects the Eraser tool and the tip brings back the tool used before (#764); **no** on native Wayland | a press of the eraser end (Pointer Events eraser button, `button` 5) selects the Eraser tool, and that press erases; a press of the tip brings back the tool used before; Pointer Events report no end while the pen hovers | partial |
| Pen eraser end | yes | the eraser end (AppKit's tablet proximity events) selects the Eraser tool and the tip brings back the tool used before, as on Linux (#852; not yet tried on a Mac) | with Windows Ink, an inverted pen or its eraser button (`PEN_FLAG_INVERTED`, `PEN_FLAG_ERASER` in the pen's pointer messages) selects the Eraser tool and the tip brings back the tool used before, as on Linux (not yet tried with a pen) | under X11 and XWayland, the eraser end selects the Eraser tool and the tip brings back the tool used before (#764); **no** on native Wayland | no | partial |
| Touch gestures (pinch zoom, two-finger pan, rotate view, Touch workspace) | yes (Windows touch, Surface) | trackpad pinch, scroll and rotate | pinch, pan (a second finger cancels the first finger's tool press, #4198); two-finger tap undo, three-finger redo (#585) | same where winit reports touch | eframe until the first pen contact, then Pointer Events (pen and fingers told apart); long press opens the context menu | partial: touch workspace and touch-screen rotation missing, 3–5 h |
| Trackpad (smooth scroll, pinch) | yes | yes | precision touchpads | yes | yes | done |
| HiDPI / Retina, per-monitor scale | yes | yes | yes | yes | yes | done; a window dragged between monitors of different scale can still grow on Windows (#1459, open); the custom title bar's window on Windows stays truly borderless so the pointer lines up with what is drawn (#1791, #1758; awaiting the reporter's confirmation) |
| Multiple monitors: panels and document windows on another screen | yes | no (one OS window) | no | no | — | missing: 6–10 h (with multiple windows, gap G7) |
| Display colour profile (proof and display through the monitor's ICC) | yes | no | no | no | — | missing: 3–5 h |
| 3D GPU rendering (Materials, ray tracing) | yes | — | — | — | — | blocked on 3D (gap G3) |

## Revision history

| Date | Change | Summary |
|---|---|---|
| 2026-10-30 | patch | Touch (#4198): a second finger landing cancels the tool press the first finger began, also in the same frame (the Pencil's stroke and a Selection move are rolled back, nothing enters the undo history), and a finger landing while another is down presses nothing, so a pinch or two-finger pan only navigates. Headless canvas test `a_second_finger_cancels_the_tool_gesture_the_first_began`; not tried on a touch screen; score unchanged (estimated) |
| 2026-10-24 | minor | macOS: an AppKit local event monitor reads each mouse and tablet event's pen pressure and the pen's end before the window gets it and writes them into the canvas's pen feed (#852): each pointer move of a stroke keeps its own pressure, mouse event coalescing is off, and the eraser end selects the Eraser tool. The mapping is unit-tested on every platform and the macOS code type-checks for `aarch64-apple-darwin`; not yet tried on a Mac with a tablet, so the score is unchanged (estimated) |
| 2026-10-15 | patch | Web review: preserve pen buttons and modifiers; leave fingers to eframe before the first pen contact; remove persistent pen policy. Canvas regression tests cover intermediate moves and long-touch cancellation on native and web input. Physical-device QA pending. |
| 2026-10-14 | minor | Tool cursors are painted into the window on Wayland (#1652; not run on Linux, #1645 stays open until confirmed); On Windows the custom title bar's window stays truly borderless, so the pointer lines up with what is drawn on systems where the hidden caption offset it (#1791, #1758; awaiting the reporter's confirmation); neither is confirmed on the hardware that reported it, so nothing is marked done |
| 2026-10-13 | minor | Web: the shell reads pen and finger Pointer Events itself (pen pressure with coalesced events, Apple Pencil hover, palm rejection: after a pen contact fingers only navigate until reload; long press opens the context menu); score unchanged (estimated) |
| 2026-10-13 | minor | A window whose surface is larger than 8192 px (stretched over several monitors, or grown by a drag between monitors of different scale) is configured up to the graphics adapter's limit instead of closing the app (#1495, part of #1459); the window growing during that drag on Windows is still open; not reproduced on a mixed-scale setup |
| 2026-10-11 | minor | Preferences › Performance › Linux Display Server › X11 (XWayland) opens the window through XWayland on a Wayland session, where pens and dropped files reach the app (#1151); score unchanged (estimated) |
| 2026-10-10 | major | First checklist from the canvas, touch and graphics code and the open hardware issues |