1 files changed, 238 insertions, 1 deletions
diff --git a/crates/core/src/monitor.rs b/crates/core/src/monitor.rs
index 01f54cb..6ea053f 100644
--- a/crates/core/src/monitor.rs
+++ b/crates/core/src/monitor.rs
@@ -42,10 +42,247 @@ pub struct Monitor {
     pub name: String,
     pub refresh_rate_mhz: u32,
     pub primary: bool,
+    /// `true` when this entry is one part of a real output divided by
+    /// `srd.monitor.split` - not a second `wl_output`, not a second
+    /// physical connector. A display-arrangement UI reads this to tell a
+    /// split part apart from a genuinely separate monitor, so it does not
+    /// offer to move or extend a physical arrangement onto something that
+    /// is not a real, independent output. `false` for an ordinary,
+    /// undivided output.
+    pub split: bool,
+    /// This output's real scale factor (automatic, from `srdwm_core::
+    /// monitor::auto_scale_for`, or an explicit `srd.monitor.scale`
+    /// override) - `1.0` for an unscaled output. Every other field on
+    /// this struct (`geometry`, `full_geometry`, `maximize_geometry`) is
+    /// in *physical* pixels, not the logical points a Wayland client
+    /// itself sees; a caller that needs to convert between the two (a
+    /// display-arrangement UI chaining outputs by their reported size,
+    /// for instance) multiplies logical by this to get physical, or
+    /// divides physical by this to get logical. Requested directly by the
+    /// AGS peer session after a real bug (`srd dispatch set output
+    /// position` and this compositor's own physical-pixel bookkeeping
+    /// silently disagreeing with a client's logical one at any scale
+    /// other than `1.0`) traced back to exactly this missing piece of
+    /// information.
+    pub scale: f64,
 }
 
 impl Monitor {
     pub fn new(id: MonitorId, name: impl Into<String>, geometry: Rect) -> Self {
-        Self { id, name: name.into(), geometry, full_geometry: geometry, maximize_geometry: geometry, refresh_rate_mhz: 60_000, primary: false }
+        Self { id, name: name.into(), geometry, full_geometry: geometry, maximize_geometry: geometry, refresh_rate_mhz: 60_000, primary: false, split: false, scale: 1.0 }
+    }
+}
+
+/// A connector a backend has administratively disabled (`srd dispatch set
+/// output enabled <name> false`) but that's still physically connected --
+/// purely informational, reported by the backend via `WindowManager::
+/// set_disabled_monitor` for `srd monitors`/the `monitors` subscribe event
+/// to list (so a display-settings UI can offer to turn it back on by
+/// name), and deliberately never fed into `WindowManager::monitors()` or
+/// any real placement/tiling logic, which continues to see only genuinely
+/// live outputs exactly as before this existed. Geometry is a last-known
+/// snapshot from the moment it was disabled - stale by construction, and
+/// meant to be: a caller wanting to reposition it correctly re-queries
+/// once it's actually re-enabled, not from this.
+#[derive(Debug, Clone)]
+pub struct DisabledMonitor {
+    pub geometry: Rect,
+    pub full_geometry: Rect,
+    pub primary: bool,
+}
+
+/// A `srd.monitor.split(name, parts, direction)` config-time request:
+/// divide one real output into `parts` equal (within a pixel) logical
+/// [`Monitor`] entries, so placement/tiling can treat them as separate
+/// screens without any DRM/`wl_output` involvement - see `split_rect`'s
+/// own doc comment for the actual division, and the udev platform's
+/// `monitors()` for where this turns into real `Monitor` entries.
+///
+/// Deliberately just a division of one real output's rectangle for
+/// placement purposes, not a second `wl_output` global - a client
+/// fullscreening or querying `wl_output.enter`/scale for a specific
+/// sub-region still sees it as part of the one real output. See the
+/// "different monitors mode in one" plan for why that's an accepted,
+/// explicitly-flagged limitation of this first version.
+#[derive(Debug, Clone, Copy)]
+pub struct MonitorSplit {
+    pub parts: u32,
+    /// `false` (the default): side-by-side columns, splitting width.
+    /// `true`: stacked rows, splitting height.
+    pub rows: bool,
+}
+
+/// Divides `rect` into `parts` equal (within one pixel) pieces along one
+/// axis, returning piece number `index` (`0..parts`). `rows` chooses which
+/// axis: stacked rows (splitting height) when `true`, side-by-side columns
+/// (splitting width) when `false`.
+///
+/// Any remainder from an uneven division is spread one pixel at a time
+/// across the first `remainder` pieces, rather than dumped entirely onto
+/// the last one - so a 1919px-wide monitor split into 2 columns yields
+/// 960/959, not a lopsided 959/960 vs. a naive 959/960-plus-slack-on-one-
+/// side that would leave one part visibly wider for no reason tied to the
+/// actual pixel count.
+///
+/// `index >= parts` or `parts == 0` returns `rect` unchanged - callers
+/// are expected to only iterate `0..parts.max(1)`, this is just a safe
+/// fallback rather than a panic for a config-driven value.
+pub fn split_rect(rect: Rect, index: u32, parts: u32, rows: bool) -> Rect {
+    if parts <= 1 || index >= parts {
+        return rect;
+    }
+    let total = if rows { rect.height } else { rect.width };
+    let other = if rows { rect.width } else { rect.height };
+    let base = total / parts;
+    let remainder = total % parts;
+    let size_for = |i: u32| base + if i < remainder { 1 } else { 0 };
+    let offset: u32 = (0..index).map(size_for).sum();
+    let size = size_for(index);
+    if rows {
+        Rect::new(rect.x, rect.y + offset as i32, other, size)
+    } else {
+        Rect::new(rect.x + offset as i32, rect.y, size, other)
+    }
+}
+
+/// The pixel density (in real, physical-size terms) srdwm treats as
+/// needing no scale correction at all. `92`, close to the classic desktop
+/// "96 DPI" constant - lowered from an initial `109` (roughly a 24"
+/// 1920x1080 or 27" 2560x1440 monitor) after live testing on a real 1080p
+/// monitor at ~78 PPI: `109` produced a `0.71` scale there, reported as
+/// too aggressive a shrink; `92` produces `~0.85`, still a real reduction
+/// but closer to what actually reads as "more space", not "suddenly tiny
+/// text".
+const REFERENCE_PPI: f64 = 92.0;
+
+/// Automatically derives an output scale from real EDID physical size and
+/// native resolution, with no monitor name or fixed size bucket involved
+/// anywhere - a large panel with low pixel density (a big monitor at the
+/// same resolution as a much smaller one, the concrete case this exists
+/// for) gets scaled down smoothly in proportion to how far its real PPI
+/// falls below [`REFERENCE_PPI`], clamped to `0.5` so a pathologically
+/// large/low-res panel doesn't shrink text into illegibility. Deliberately
+/// never scales *above* `1.0` on its own - a high-density panel already
+/// benefits from more detail, not less, and plenty of people want native
+/// crispness there; `srd.monitor.scale` remains the explicit, manual way
+/// to opt into upscaling a specific connector.
+///
+/// `physical_mm` of `(0, 0)` (no EDID physical-size descriptor at all --
+/// some VMs/adapters report this) returns `1.0` rather than guessing from
+/// nothing.
+pub fn auto_scale_for(physical_mm: (i32, i32), resolution_px: (i32, i32)) -> f64 {
+    let (pw, ph) = physical_mm;
+    if pw <= 0 || ph <= 0 {
+        return 1.0;
+    }
+    let diagonal_mm = ((pw as f64).powi(2) + (ph as f64).powi(2)).sqrt();
+    let diagonal_in = diagonal_mm / 25.4;
+    let (rw, rh) = resolution_px;
+    let diagonal_px = ((rw as f64).powi(2) + (rh as f64).powi(2)).sqrt();
+    let ppi = diagonal_px / diagonal_in;
+    if ppi >= REFERENCE_PPI {
+        1.0
+    } else {
+        (ppi / REFERENCE_PPI).clamp(0.5, 1.0)
+    }
+}
+
+#[cfg(test)]
+mod auto_scale_tests {
+    use super::*;
+
+    #[test]
+    fn a_15_inch_1080p_laptop_panel_needs_no_correction() {
+        // 340mm x 190mm, ~143 PPI - comfortably above the reference, and
+        // the concrete real-hardware case this must not regress: this
+        // laptop's own panel was already correct at 1.0.
+        assert_eq!(auto_scale_for((340, 190), (1920, 1080)), 1.0);
+    }
+
+    #[test]
+    fn a_physically_large_1080p_monitor_scales_down() {
+        // 600mm x 400mm at the same 1920x1080 as the laptop above --
+        // ~78 PPI, well under the reference. The concrete case this whole
+        // function exists for: reported live as "too big, should utilize
+        // greater real estate" on exactly this monitor.
+        let s = auto_scale_for((600, 400), (1920, 1080));
+        assert!(s < 1.0 && s > 0.5, "expected a real scale-down, got {s}");
+    }
+
+    #[test]
+    fn a_high_density_panel_is_never_auto_upscaled() {
+        // A small, very high-resolution panel (e.g. a 13" 4K) - far above
+        // the reference PPI. Must clamp at 1.0, not scale past it.
+        assert_eq!(auto_scale_for((290, 170), (3840, 2160)), 1.0);
+    }
+
+    #[test]
+    fn an_extreme_low_density_panel_clamps_at_half_scale() {
+        let s = auto_scale_for((2000, 1200), (1024, 768));
+        assert_eq!(s, 0.5);
+    }
+
+    #[test]
+    fn missing_physical_size_does_not_guess() {
+        assert_eq!(auto_scale_for((0, 0), (1920, 1080)), 1.0);
+    }
+}
+
+#[cfg(test)]
+mod split_tests {
+    use super::*;
+
+    #[test]
+    fn single_part_returns_the_whole_rect_unchanged() {
+        let r = Rect::new(0, 0, 1920, 1080);
+        assert_eq!(split_rect(r, 0, 1, false), r);
+    }
+
+    #[test]
+    fn even_columns_split_width_with_no_gap_or_overlap() {
+        let r = Rect::new(100, 0, 1920, 1080);
+        let a = split_rect(r, 0, 2, false);
+        let b = split_rect(r, 1, 2, false);
+        assert_eq!(a, Rect::new(100, 0, 960, 1080));
+        assert_eq!(b, Rect::new(1060, 0, 960, 1080));
+        assert_eq!(a.right(), b.x, "no gap or overlap between adjacent parts");
+    }
+
+    #[test]
+    fn uneven_columns_spread_the_remainder_one_pixel_at_a_time() {
+        let r = Rect::new(0, 0, 1919, 1080);
+        let a = split_rect(r, 0, 2, false);
+        let b = split_rect(r, 1, 2, false);
+        assert_eq!(a.width, 960);
+        assert_eq!(b.width, 959);
+        assert_eq!(a.width + b.width, r.width);
+        assert_eq!(a.right(), b.x);
+    }
+
+    #[test]
+    fn rows_split_height_and_leave_width_untouched() {
+        let r = Rect::new(0, 50, 1920, 1080);
+        let a = split_rect(r, 0, 2, true);
+        let b = split_rect(r, 1, 2, true);
+        assert_eq!(a, Rect::new(0, 50, 1920, 540));
+        assert_eq!(b, Rect::new(0, 590, 1920, 540));
+        assert_eq!(a.bottom(), b.y);
+    }
+
+    #[test]
+    fn three_parts_covers_the_whole_rect_exactly() {
+        let r = Rect::new(0, 0, 1000, 500);
+        let parts: Vec<Rect> = (0..3).map(|i| split_rect(r, i, 3, false)).collect();
+        let total_width: u32 = parts.iter().map(|p| p.width).sum();
+        assert_eq!(total_width, r.width);
+        for w in parts.windows(2) {
+            assert_eq!(w[0].right(), w[1].x);
+        }
+    }
+
+    #[test]
+    fn out_of_range_index_returns_the_whole_rect_unchanged() {
+        let r = Rect::new(0, 0, 1920, 1080);
+        assert_eq!(split_rect(r, 5, 2, false), r);
     }
 }