X11: select one monitor with UU_X11_MULTISCREEN=1 (#98)

AGENTS.md

@@ -119,6 +119,7 @@ viewer does not reconnect them automatically.
 | `~/.config/uuremote/config.json` | user_id, device_id, client_id, system_id, machine_guid, token, csrf; optional `sign_key`, `ffmpeg_root`. `$UU_CONFIG` overrides the path |
 | `~/.config/uuremote/host_settings.json` | persisted 安全 settings: auto_lock, keep_mute, mute_permanent; optional `video_encoder` (`auto`/`cpu`/`vaapi`/`nvenc`) and `video_codec` (`auto`/`h264`/`h265`), written by `set-encoder` / `set-codec` (0600, merge write). Non-`auto` beats `UU_VIDEO_ENCODER` / `UU_VIDEO_CODEC`. `set-encoder` and `set-codec` both send `SIGUSR2` to the pid in `host.pid` (same binary). Live desktop streams restart the encoder (`EncoderTask::restart`, sequence kept, first frame IDR) without dropping the peer. Codec preference is logged and applies on the next desktop stream. `SIGUSR2` does not re-probe. Hand edits: `kill -USR2 $(cat ~/.config/uuremote/host.pid)` |
 | `UU_DESKTOP_BACKEND` | `auto` (default) / `x11` / `wayland` / `portal` / `kms` (experimental feature) / `none` (headless: no display; desktop offers ignored; tool peers get a nominal 1920x1080 screen + Allow permissions, which the port-mapping viewer requires) |
+| `UU_X11_MULTISCREEN=1`, `UU_X11_OUTPUT` | X11: advertise each RandR monitor (`xrandr --listmonitors`, read once at start; primary or `UU_X11_OUTPUT` is id 0) through the same screen-switch path as `UU_WAYLAND_MULTISCREEN` (`wayland::outputs`/`select`). x11grab reads `display+x,y` at the monitor size, XTest/uinput absolute input adds the monitor offset, the in-process engine and `desktop_mode` resize are skipped (分辨率 scales). Needs two or more monitors |
 | `UU_DESKTOP_RESIZE=0` | X11: the viewer's 分辨率 choice scales the encoded stream instead of changing the display mode. Default: `src/desktop_mode.rs` switches the primary output via xrandr (adds a CVT-RB mode when missing, re-applies once if mutter resets it), restores the original mode for resolution_type 3 and 2 s after the last desktop stream ends |
 | `UU_NVENC_RGB=0` / `=1` | NVENC 4:2:0 without scaling takes packed BGRA/BGR0 directly (no CPU swscale). Default: on, except when a connected DRM connector belongs to an nvidia-bound card (desktop composited on the encoding GPU; #16 measured 82.6 vs 98 fps at 4K). `0` forces CPU conversion, `1` forces the direct path |
 | `UU_INPUT_BACKEND` | `auto` (uinput on Wayland/portal when `/dev/uinput` is accessible, udev `uaccess` rule in README) / `uinput` / `native`. X11 keeps XTest (held-key autorepeat handling) |

README.en.md

@@ -104,19 +104,21 @@ failure falls back to SHM. The virgl VM verified fallback only: successful impor
 mapping and hardware acceleration remain unverified. This is not zero-copy VAAPI;
 `UU_VIDEO_ENCODER=vaapi` still uploads CPU frames through FFmpeg.
 
-### Hyprland monitor switching (experimental)
+### Monitor switching (experimental)
 
 Native Wayland absolute pointer motion now targets the captured output by default, fixing multi-monitor click offsets. uinput may still handle keyboard, buttons and relative motion.
 
 Set `UU_WAYLAND_MULTISCREEN=1` to select physical monitors from the phone client. `UU_WAYLAND_OUTPUT` chooses the initial output. Capture and absolute pointer targeting switch together across different resolutions, scales and desktop positions.
 
+X11 sessions send all monitors as one combined picture by default. With `UU_X11_MULTISCREEN=1` the host lists the monitors from `xrandr --listmonitors`; the viewer selects one, only that region is captured, and absolute pointer positions are offset to that monitor. The primary monitor is the default; `UU_X11_OUTPUT` picks another (e.g. `HDMI-1`). The layout is read when the host starts; restart the service after changing it. In this mode the viewer's resolution setting scales the stream instead of changing the display mode. Experimental: unit tests and Xvfb virtual monitors only, not yet accepted with the official clients.
+
 Switching reuses one video track and pauses input until the new source sends a frame. Capture/encoder failure or a first-frame timeout requests restoration of the previous source. The inventory uses `hyprctl` and supports unrotated outputs; restart the host after monitor hotplug or layout changes. Phone-side switching and clicks were tested on Omarchy / Hyprland with 2560×1440 at 125% and 1920×1080 at 133.33%.
 
 ## Not implemented / not supported
 
 | Feature | Status |
 |---|---|
-| Switching between multiple monitors | Experimental on native Hyprland Wayland; see above |
+| Switching between multiple monitors | Experimental on native Hyprland Wayland and on X11 (`UU_X11_MULTISCREEN=1`); see above |
 | True color (4:4:4) | Supported (v0.1.15): the default FFmpeg pipeline with NVENC / libx264 encodes H.264 High 4:4:4; `inproc-av` native NVENC encodes H.264 High 4:4:4 / HEVC FRExt; VAAPI stays 4:2:0 and reports that to the client |
 | HDR | Not implemented |
 | Super screen / virtual display | Not implemented |
@@ -339,6 +341,8 @@ Common environment variables:
 | `UU_NVENC_RGB=0` / `=1` | NVENC encodes BGRA directly (saves CPU). Off by default when a monitor is connected to the NVIDIA card (slower when the desktop is composited on the same GPU); `0` forces off, `1` forces on |
 | `UU_WAYLAND_MULTISCREEN=1` | Enable Hyprland physical monitor switching (off by default) |
 | `UU_WAYLAND_OUTPUT` | Initial native Wayland output name, e.g. `HDMI-A-1` |
+| `UU_X11_MULTISCREEN=1` | Select and capture X11 monitors separately (experimental, off by default, needs `xrandr`) |
+| `UU_X11_OUTPUT` | Initial X11 monitor name (`xrandr --listmonitors`); default: the primary |
 | `UU_WAYLAND_NO_VIRTUAL_KEYBOARD=1` | Native Wayland skips the virtual keyboard (e.g. Denial reserves it for its input method); keys go through uinput. Without wlr-virtual-pointer the pointer also uses uinput (#72) |
 | `UU_INPUT_BACKEND` | `auto` (default: uinput on Wayland when accessible) / `uinput` / `native` |
 | `UU_TERMINAL_SHORTCUTS=0` | Disable the Ctrl+V paste rewrite in terminal windows |

README.md

@@ -90,19 +90,21 @@ GNOME Portal 的控制通道不再等待授权窗口或第一帧画面：先通
 原生 Wayland 实验选项：`UU_WAYLAND_PIPELINE=0` 可关闭流水线做对照。`UU_WAYLAND_DMABUF=1` 显式尝试 dma-buf→GBM CPU 映射（默认关闭，需要运行时 libgbm；可用 `UU_WAYLAND_DMABUF_DEVICE` 指定 render node，默认 `/dev/dri/renderD128`）。仅支持显式线性修饰符、单平面 8-bit RGB，失败回退 SHM；virgl VM 中只验证了回退，成功导入/映射和硬件加速均未验证。这不是零拷贝 VAAPI 路径；选择 `UU_VIDEO_ENCODER=vaapi` 时仍由 FFmpeg 上传 CPU 帧。
 首次连接时，桌面会弹窗请求「远程桌面 / 屏幕共享」授权，同意后会记住授权。
 
-### Hyprland 多显示器切换（实验性）
+### 多显示器切换（实验性）
 
 原生 Wayland 的绝对鼠标定位默认绑定正在采集的输出，修复双屏桌面下的点击偏移。uinput 可继续用于键盘、按键和相对移动。
 
 设置 `UU_WAYLAND_MULTISCREEN=1` 后，手机端可选择并切换物理显示器；`UU_WAYLAND_OUTPUT` 指定默认屏幕。画面与绝对鼠标定位同步切换，支持不同分辨率、缩放和桌面位置。
 
+X11 会话默认把所有显示器拼成一张画面。设置 `UU_X11_MULTISCREEN=1` 后，主机按 `xrandr --listmonitors` 列出各个显示器，控制端可选择其中一块，只采集该区域，绝对鼠标坐标加上该屏幕在桌面中的偏移；默认屏幕为主显示器，`UU_X11_OUTPUT` 可指定其他屏幕（如 `HDMI-1`）。显示器布局在主机启动时读取，改动后需重启服务。此模式下控制端的「分辨率」改为缩放画面，不再修改显示模式。实验性：目前只有单元测试与 Xvfb 虚拟显示器实测，尚未经官方客户端实机验收。
+
 切屏复用同一条视频轨道，过渡期间暂停输入，发送新屏首帧后恢复；采集或编码失败、等待首帧超时会请求恢复原屏。当前通过 `hyprctl` 枚举未旋转的显示器，显示器插拔或布局变化后需重启服务。已在 Omarchy / Hyprland 上用官方手机客户端实测 2560×1440（125%）与 1920×1080（133.33%）双屏切换及点击。
 
 ## 尚未实现 / 不支持
 
 | 功能 | 状态 |
 |---|---|
-| 多显示器切换 | Hyprland 原生 Wayland 后端实验性支持，见上文 |
+| 多显示器切换 | Hyprland 原生 Wayland 后端与 X11（`UU_X11_MULTISCREEN=1`）实验性支持，见上文 |
 | 真彩（4:4:4） | 支持（v0.1.15）：NVENC / libx264 的默认 FFmpeg 管线输出 H.264 High 4:4:4；`inproc-av` 原生 NVENC 输出 H.264 High 4:4:4 / HEVC FRExt；VAAPI 管线仍为 4:2:0，并如实回报给客户端 |
 | HDR | 未实现 |
 | 超级屏 / 虚拟显示器 | 未实现 |
@@ -307,6 +309,8 @@ systemctl --user restart uuremote-native.service
 | `UU_NVENC_RGB=0` / `=1` | NVENC 直接编码 BGRA（省 CPU）。显示器接在 NVIDIA 卡上时默认关闭（同卡合成时直吃反而更慢），`0` 强制关，`1` 强制开 |
 | `UU_WAYLAND_MULTISCREEN=1` | 启用 Hyprland 物理显示器切换（默认关闭） |
 | `UU_WAYLAND_OUTPUT` | 原生 Wayland 默认采集输出名称，例如 `HDMI-A-1` |
+| `UU_X11_MULTISCREEN=1` | X11 下按显示器分别选择和采集（实验性，默认关闭，需要 `xrandr`） |
+| `UU_X11_OUTPUT` | X11 默认采集的显示器名称（`xrandr --listmonitors`），默认主显示器 |
 | `UU_WAYLAND_NO_VIRTUAL_KEYBOARD=1` | 原生 Wayland 不创建虚拟键盘（如 Denial 只把它给自己的输入法），键盘改走 uinput。合成器没有 wlr-virtual-pointer 时鼠标也自动改走 uinput（#72） |
 | `UU_INPUT_BACKEND` | `auto`（默认，Wayland 下可用时用 uinput）/ `uinput` / `native` |
 | `UU_TERMINAL_SHORTCUTS=0` | 关闭终端窗口里的 Ctrl+V 粘贴转换 |

src/capture.rs

@@ -56,6 +56,12 @@ pub fn x11_size() -> Option<(u32, u32)> {
 }
 
 /// The root size from the last refresh, without an X round trip (pointer mapping).
+/// Display the X11 engine opened, once a capture has initialized it.
+pub fn x11_display() -> Option<String> {
+    let name = unsafe { x11_display_name() };
+    if name.is_null() { None } else { Some(unsafe { std::ffi::CStr::from_ptr(name) }.to_string_lossy().into_owned()) }
+}
+
 pub fn x11_last_size() -> Option<(u32, u32)> {
     let (mut w, mut h): (libc::c_int, libc::c_int) = (0, 0);
     if unsafe { x11_get_dimensions(&mut w, &mut h) } != 0 || w <= 0 || h <= 0 { return None; }

src/input.rs

@@ -334,9 +334,12 @@ impl InputInjector {
             let result=crate::wayland::move_pointer(x,y,false);
             if result!=-2 { return result; }
         }
+        // X11 multi-screen: the point is inside the selected monitor; XTest and uinput span the root.
+        let placement=if relative || self.backend!=crate::desktop_backend::DesktopBackend::X11 {None} else {crate::wayland::x11_placement()};
+        let (x,y)=placement.map_or((x,y),|(ox,oy)|(x+ox,y+oy));
         if let Some(uinput)=&self.uinput {
             if relative { return Self::uinput_result(uinput.move_relative(x,y)); }
-            let (w,h)=*self.screen.lock().unwrap();
+            let (w,h)=placement.and_then(|_|crate::capture::x11_last_size()).unwrap_or(*self.screen.lock().unwrap());
             if w > 1 && h > 1 {
                 let scale = |v: i32, extent: u32| (i64::from(v) * i64::from(crate::uinput::ABS_MAX) / i64::from(extent - 1)) as i32;
                 return Self::uinput_result(uinput.move_absolute(scale(x,w),scale(y,h)));
@@ -700,3 +703,28 @@ mod output_pointer_tests {
         }
     }
 }
+
+#[cfg(all(test,target_os="linux"))]
+mod x11_monitor_pointer_tests {
+    use super::*;
+    #[test]
+    #[ignore="requires an X11 display with two or more RandR monitors and UU_X11_MULTISCREEN=1; moves pointer only"]
+    fn absolute_pointer_lands_on_the_selected_x11_monitor() {
+        let _capture=crate::capture::ScreenCapture::new().unwrap();
+        let input=InputInjector::new();
+        assert_eq!(input.backend,crate::desktop_backend::DesktopBackend::X11);
+        let outputs=crate::wayland::outputs();assert!(crate::wayland::x11_multiscreen() && outputs.len()>=2,"needs two X11 monitors");
+        for (id,output) in outputs.iter().enumerate().chain(outputs.iter().enumerate().rev()) {
+            crate::wayland::select(id).unwrap();
+            for (x,y) in [(0.0,0.0),(0.5,0.5),(0.999,0.999),(0.1,0.9)] {
+                let size=(output.width,output.height);
+                let pixels=crate::input_mapping::normalized_pointer(x,y,size.0,size.1).unwrap();
+                input.mouse_absolute_normalized(x,y,size);input.flush_pointer();
+                let (mut ax,mut ay,mut mask)=(0,0,0);
+                assert_eq!(unsafe {x11_pointer_state(&mut ax,&mut ay,&mut mask)},0);
+                assert_eq!((ax,ay),(output.x+pixels.0,output.y+pixels.1),"monitor={} normalized=({x},{y})",output.name);
+            }
+        }
+        crate::wayland::select(0).unwrap();
+    }
+}

src/video.rs

@@ -814,18 +814,20 @@ async fn read_encoder_with(
     settings: VideoSettings, #[cfg_attr(not(feature = "inproc-av"), allow(unused_variables))] idr: Arc<AtomicBool>, budget: Arc<Semaphore>,
     selection: crate::video_select::Selection,
 )->Result<(),String>{
-    let selected_output=if display.starts_with("wayland:") {crate::wayland::source(settings.source_id)} else {None};
+    let selected_output=crate::wayland::capture_source(&display,portal.is_some(),settings.source_id);
     let size=selected_output.map_or(size,|s|(s.width,s.height));
     let kms=display.starts_with("kms:");
     let wayland=kms || display.starts_with("wayland:");
+    // X11 multi-screen: x11grab reads only the selected monitor's region of the root.
+    let x11_region=selected_output.filter(|_|!wayland);
     let (permits_tx,mut permits_rx)=mpsc::channel::<OwnedSemaphorePermit>(4);
     let kms_capture=if kms {
         let capture=tokio::task::spawn_blocking(crate::kms::Capture::open).await.map_err(|_|"KMS worker failed")??;
         if (capture.width,capture.height)!=size { return Err("KMS mode changed; restart host to refresh desktop geometry".into()); }
         Some(Arc::new(capture))
     } else { None };
     #[cfg(feature = "inproc-av")]
-    if allow_inproc && native_inproc(&display, portal.is_some(), encoder) {
+    if allow_inproc && x11_region.is_none() && native_inproc(&display, portal.is_some(), encoder) {
         let max_kbps = settings.max_kbps().unwrap_or(30_000) as i64;
         let config = crate::inproc_encoder::Config {
             display: display.clone(), input: size, output: settings.output, fps: u32::from(settings.fps),
@@ -859,8 +861,9 @@ async fn read_encoder_with(
             "-framerate",&fps,"-i","pipe:0","-an"]);
     } else {
         // The viewer draws its local pointer; do not bake a second into video.
+        let input=x11_region.map_or_else(||display.clone(),|o|format!("{display}+{},{}",o.x,o.y));
         command.args(["-f","x11grab","-draw_mouse","0","-framerate",&fps,
-            "-video_size",&size,"-i",&display,"-an"]);
+            "-video_size",&size,"-i",&input,"-an"]);
     }
     // 其他分辨率: scale the captured desktop (the host display mode is left untouched).
     command.args(scale_options(&settings, encoder, Some(source_size)));