DPI for everyone (#548)

CHANGELOG.md

@@ -1,6 +1,17 @@
 # Unreleased
 
 - **Breaking:** Removed `VirtualKeyCode::LMenu` and `VirtualKeyCode::RMenu`; Windows now generates `VirtualKeyCode::LAlt` and `VirtualKeyCode::RAlt` instead.
+- On X11, exiting fullscreen no longer leaves the window in the monitor's top left corner.
+- **Breaking:** `Window::hidpi_factor` has been renamed to `Window::get_hidpi_factor` for better consistency. `WindowEvent::HiDPIFactorChanged` has been renamed to `WindowEvent::HiDpiFactorChanged`. DPI factors are always represented as `f64` instead of `f32` now.
+- The Windows backend is now DPI aware. `WindowEvent::HiDpiFactorChanged` is implemented, and `MonitorId::get_hidpi_factor` and `Window::hidpi_factor` return accurate values.
+- Implemented `WindowEvent::HiDpiFactorChanged` on X11.
+- On macOS, `Window::set_cursor_position` is now relative to the client area.
+- On macOS, setting the maximum and minimum dimensions now applies to the client area dimensions rather than to the window dimensions.
+- On iOS, `MonitorId::get_dimensions` has been implemented and both `MonitorId::get_hidpi_factor` and `Window::get_hidpi_factor` return accurate values.
+- On Emscripten, `MonitorId::get_hidpi_factor` now returns the same value as `Window::get_hidpi_factor` (it previously would always return 1.0).
+- **Breaking:** The entire API for sizes, positions, etc. has changed. In the majority of cases, winit produces and consumes positions and sizes as `LogicalPosition` and `LogicalSize`, respectively. The notable exception is `MonitorId` methods, which deal in `PhysicalPosition` and `PhysicalSize`. See the documentation for specifics and explanations of the types. Additionally, winit automatically conserves logical size when the DPI factor changes.
+- **Breaking:** All deprecated methods have been removed. For `Window::platform_display` and `Window::platform_window`, switch to the appropriate platform-specific `WindowExt` methods. For `Window::get_inner_size_points` and `Window::get_inner_size_pixels`, use the `LogicalSize` returned by `Window::get_inner_size` and convert as needed.
+- HiDPI support for Wayland.
 
 # Version 0.15.1 (2018-06-13)
 

Cargo.toml

@@ -43,6 +43,7 @@ features = [
     "objbase",
     "processthreadsapi",
     "shellapi",
+    "shellscalingapi",
     "shobjidl_core",
     "unknwnbase",
     "windowsx",

examples/min_max_size.rs

@@ -1,14 +1,16 @@
 extern crate winit;
 
+use winit::dpi::LogicalSize;
+
 fn main() {
     let mut events_loop = winit::EventsLoop::new();
 
     let window = winit::WindowBuilder::new()
         .build(&events_loop)
         .unwrap();
 
-    window.set_min_dimensions(Some((400, 200)));
-    window.set_max_dimensions(Some((800, 400)));
+    window.set_min_dimensions(Some(LogicalSize::new(400.0, 200.0)));
+    window.set_max_dimensions(Some(LogicalSize::new(800.0, 400.0)));
 
     events_loop.run_forever(|event| {
         println!("{:?}", event);

examples/resizable.rs

@@ -7,7 +7,7 @@ fn main() {
 
     let window = winit::WindowBuilder::new()
         .with_title("Hit space to toggle resizability.")
-        .with_dimensions(400, 200)
+        .with_dimensions((400, 200).into())
         .with_resizable(resizable)
         .build(&events_loop)
         .unwrap();

src/dpi.rs

@@ -0,0 +1,307 @@
+
+//! DPI is important, so read the docs for this module if you don't want to be confused.
+//!
+//! Originally, `winit` dealt entirely in physical pixels (excluding unintentional inconsistencies), but now all
+//! window-related functions both produce and consume logical pixels. Monitor-related functions still use physical
+//! pixels, as do any context-related functions in `glutin`.
+//!
+//! If you've never heard of these terms before, then you're not alone, and this documentation will explain the
+//! concepts.
+//!
+//! Modern screens have a defined physical resolution, most commonly 1920x1080. Indepedent of that is the amount of
+//! space the screen occupies, which is to say, the height and width in millimeters. The relationship between these two
+//! measurements is the *pixel density*. Mobile screens require a high pixel density, as they're held close to the
+//! eyes. Larger displays also require a higher pixel density, hence the growing presence of 1440p and 4K displays.
+//!
+//! So, this presents a problem. Let's say we want to render a square 100px button. It will occupy 100x100 of the
+//! screen's pixels, which in many cases, seems perfectly fine. However, because this size doesn't account for the
+//! screen's dimensions or pixel density, the button's size can vary quite a bit. On a 4K display, it would be unusably
+//! small.
+//!
+//! That's a description of what happens when the button is 100x100 *physical* pixels. Instead, let's try using 100x100
+//! *logical* pixels. To map logical pixels to physical pixels, we simply multiply by the DPI factor. On a "typical"
+//! desktop display, the DPI factor will be 1.0, so 100x100 logical pixels equates to 100x100 physical pixels. However,
+//! a 1440p display may have a DPI factor of 1.25, so the button is rendered as 125x125 physical pixels. Ideally, the
+//! button now has approximately the same perceived size across varying displays.
+//!
+//! Failure to account for the DPI factor can create a badly degraded user experience. Most notably, it can make users
+//! feel like they have bad eyesight, which will potentially cause them to think about growing elderly, resulting in
+//! them entering an existential panic. Once users enter that state, they will no longer be focused on your application.
+//!
+//! There are two ways to get the DPI factor: either by calling
+//! [`MonitorId::get_hidpi_factor`](../struct.MonitorId.html#method.get_hidpi_factor), or
+//! [`Window::get_hidpi_factor`](../struct.Window.html#method.get_hidpi_factor). You'll almost always use the latter,
+//! which is basically equivalent to `window.get_current_monitor().get_hidpi_factor()` anyway.
+//!
+//! Here's an overview of what sort of DPI factors you can expect, and where they come from:
+//! - **Windows:** On Windows 8 and 10, per-monitor scaling is readily configured by users from the display settings.
+//! While users are free to select any option they want, they're only given a selection of "nice" DPI factors, i.e.
+//! 1.0, 1.25, 1.5... on Windows 7, the DPI factor is global and changing it requires logging out.
+//! - **macOS:** The buzzword is "retina displays", which have a DPI factor of 2.0. Otherwise, the DPI factor is 1.0.
+//! Intermediate DPI factors are never used, thus 1440p displays/etc. aren't properly supported. It's possible for any
+//! display to use that 2.0 DPI factor, given the use of the command line.
+//! - **X11:** On X11, we calcuate the DPI factor based on the millimeter dimensions provided by XRandR. This can
+//! result in a wide range of possible values, including some interesting ones like 1.0833333333333333. This can be
+//! overridden using the `WINIT_HIDPI_FACTOR` environment variable, though that's not recommended.
+//! - **Wayland:** On Wayland, DPI factors are very much at the discretion of the user.
+//! - **iOS:** DPI factors are both constant and device-specific on iOS.
+//! - **Android:** This feature isn't yet implemented on Android, so the DPI factor will always be returned as 1.0.
+//!
+//! The window's logical size is conserved across DPI changes, resulting in the physical size changing instead. This
+//! may be surprising on X11, but is quite standard elsewhere. Physical size changes produce a
+//! [`Resized`](../enum.WindowEvent.html#variant.Resized) event, even on platforms where no resize actually occurs,
+//! such as macOS and Wayland. As a result, it's not necessary to separately handle
+//! [`HiDpiFactorChanged`](../enum.WindowEvent.html#variant.HiDpiFactorChanged) if you're only listening for size.
+//!
+//! Your GPU has no awareness of the concept of logical pixels, and unless you like wasting pixel density, your
+//! framebuffer's size should be in physical pixels.
+
+/// Checks that the DPI factor is a normal positive `f64`.
+///
+/// All functions that take a DPI factor assert that this will return `true`. If you're sourcing DPI factors from
+/// anywhere other than winit, it's recommended to validate them using this function before passing them to winit;
+/// otherwise, you risk panics.
+#[inline]
+pub fn validate_hidpi_factor(dpi_factor: f64) -> bool {
+    dpi_factor.is_sign_positive() && dpi_factor.is_normal()
+}
+
+/// A position represented in logical pixels.
+///
+/// The position is stored as floats, so please be careful. Casting floats to integers truncates the fractional part,
+/// which can cause noticable issues. To help with that, an `Into<(i32, i32)>` implementation is provided which
+/// does the rounding for you.
+#[derive(Debug, Copy, Clone, PartialEq)]
+pub struct LogicalPosition {
+    pub x: f64,
+    pub y: f64,
+}
+
+impl LogicalPosition {
+    #[inline]
+    pub fn new(x: f64, y: f64) -> Self {
+        LogicalPosition { x, y }
+    }
+
+    #[inline]
+    pub fn from_physical<T: Into<PhysicalPosition>>(physical: T, dpi_factor: f64) -> Self {
+        physical.into().to_logical(dpi_factor)
+    }
+
+    #[inline]
+    pub fn to_physical(&self, dpi_factor: f64) -> PhysicalPosition {
+        assert!(validate_hidpi_factor(dpi_factor));
+        let x = self.x * dpi_factor;
+        let y = self.y * dpi_factor;
+        PhysicalPosition::new(x, y)
+    }
+}
+
+impl From<(f64, f64)> for LogicalPosition {
+    #[inline]
+    fn from((x, y): (f64, f64)) -> Self {
+        Self::new(x, y)
+    }
+}
+
+impl From<(i32, i32)> for LogicalPosition {
+    #[inline]
+    fn from((x, y): (i32, i32)) -> Self {
+        Self::new(x as f64, y as f64)
+    }
+}
+
+impl Into<(f64, f64)> for LogicalPosition {
+    #[inline]
+    fn into(self) -> (f64, f64) {
+        (self.x, self.y)
+    }
+}
+
+impl Into<(i32, i32)> for LogicalPosition {
+    /// Note that this rounds instead of truncating.
+    #[inline]
+    fn into(self) -> (i32, i32) {
+        (self.x.round() as _, self.y.round() as _)
+    }
+}
+
+/// A position represented in physical pixels.
+///
+/// The position is stored as floats, so please be careful. Casting floats to integers truncates the fractional part,
+/// which can cause noticable issues. To help with that, an `Into<(i32, i32)>` implementation is provided which
+/// does the rounding for you.
+#[derive(Debug, Copy, Clone, PartialEq)]
+pub struct PhysicalPosition {
+    pub x: f64,
+    pub y: f64,
+}
+
+impl PhysicalPosition {
+    #[inline]
+    pub fn new(x: f64, y: f64) -> Self {
+        PhysicalPosition { x, y }
+    }
+
+    #[inline]
+    pub fn from_logical<T: Into<LogicalPosition>>(logical: T, dpi_factor: f64) -> Self {
+        logical.into().to_physical(dpi_factor)
+    }
+
+    #[inline]
+    pub fn to_logical(&self, dpi_factor: f64) -> LogicalPosition {
+        assert!(validate_hidpi_factor(dpi_factor));
+        let x = self.x / dpi_factor;
+        let y = self.y / dpi_factor;
+        LogicalPosition::new(x, y)
+    }
+}
+
+impl From<(f64, f64)> for PhysicalPosition {
+    #[inline]
+    fn from((x, y): (f64, f64)) -> Self {
+        Self::new(x, y)
+    }
+}
+
+impl From<(i32, i32)> for PhysicalPosition {
+    #[inline]
+    fn from((x, y): (i32, i32)) -> Self {
+        Self::new(x as f64, y as f64)
+    }
+}
+
+impl Into<(f64, f64)> for PhysicalPosition {
+    #[inline]
+    fn into(self) -> (f64, f64) {
+        (self.x, self.y)
+    }
+}
+
+impl Into<(i32, i32)> for PhysicalPosition {
+    /// Note that this rounds instead of truncating.
+    #[inline]
+    fn into(self) -> (i32, i32) {
+        (self.x.round() as _, self.y.round() as _)
+    }
+}
+
+/// A size represented in logical pixels.
+///
+/// The size is stored as floats, so please be careful. Casting floats to integers truncates the fractional part,
+/// which can cause noticable issues. To help with that, an `Into<(u32, u32)>` implementation is provided which
+/// does the rounding for you.
+#[derive(Debug, Copy, Clone, PartialEq)]
+pub struct LogicalSize {
+    pub width: f64,
+    pub height: f64,
+}
+
+impl LogicalSize {
+    #[inline]
+    pub fn new(width: f64, height: f64) -> Self {
+        LogicalSize { width, height }
+    }
+
+    #[inline]
+    pub fn from_physical<T: Into<PhysicalSize>>(physical: T, dpi_factor: f64) -> Self {
+        physical.into().to_logical(dpi_factor)
+    }
+
+    #[inline]
+    pub fn to_physical(&self, dpi_factor: f64) -> PhysicalSize {
+        assert!(validate_hidpi_factor(dpi_factor));
+        let width = self.width * dpi_factor;
+        let height = self.height * dpi_factor;
+        PhysicalSize::new(width, height)
+    }
+}
+
+impl From<(f64, f64)> for LogicalSize {
+    #[inline]
+    fn from((width, height): (f64, f64)) -> Self {
+        Self::new(width, height)
+    }
+}
+
+impl From<(u32, u32)> for LogicalSize {
+    #[inline]
+    fn from((width, height): (u32, u32)) -> Self {
+        Self::new(width as f64, height as f64)
+    }
+}
+
+impl Into<(f64, f64)> for LogicalSize {
+    #[inline]
+    fn into(self) -> (f64, f64) {
+        (self.width, self.height)
+    }
+}
+
+impl Into<(u32, u32)> for LogicalSize {
+    /// Note that this rounds instead of truncating.
+    #[inline]
+    fn into(self) -> (u32, u32) {
+        (self.width.round() as _, self.height.round() as _)
+    }
+}
+
+/// A size represented in physical pixels.
+///
+/// The size is stored as floats, so please be careful. Casting floats to integers truncates the fractional part,
+/// which can cause noticable issues. To help with that, an `Into<(u32, u32)>` implementation is provided which
+/// does the rounding for you.
+#[derive(Debug, Copy, Clone, PartialEq)]
+pub struct PhysicalSize {
+    pub width: f64,
+    pub height: f64,
+}
+
+impl PhysicalSize {
+    #[inline]
+    pub fn new(width: f64, height: f64) -> Self {
+        PhysicalSize { width, height }
+    }
+
+    #[inline]
+    pub fn from_logical<T: Into<LogicalSize>>(logical: T, dpi_factor: f64) -> Self {
+        logical.into().to_physical(dpi_factor)
+    }
+
+    #[inline]
+    pub fn to_logical(&self, dpi_factor: f64) -> LogicalSize {
+        assert!(validate_hidpi_factor(dpi_factor));
+        let width = self.width / dpi_factor;
+        let height = self.height / dpi_factor;
+        LogicalSize::new(width, height)
+    }
+}
+
+impl From<(f64, f64)> for PhysicalSize {
+    #[inline]
+    fn from((width, height): (f64, f64)) -> Self {
+        Self::new(width, height)
+    }
+}
+
+impl From<(u32, u32)> for PhysicalSize {
+    #[inline]
+    fn from((width, height): (u32, u32)) -> Self {
+        Self::new(width as f64, height as f64)
+    }
+}
+
+impl Into<(f64, f64)> for PhysicalSize {
+    #[inline]
+    fn into(self) -> (f64, f64) {
+        (self.width, self.height)
+    }
+}
+
+impl Into<(u32, u32)> for PhysicalSize {
+    /// Note that this rounds instead of truncating.
+    #[inline]
+    fn into(self) -> (u32, u32) {
+        (self.width.round() as _, self.height.round() as _)
+    }
+}