Fix set_monitor_split never actually reaching srd monitors

Live-tested right after shipping it and caught immediately: srd dispatch
set output split returned ok, but srd monitors kept reporting the whole,
unsplit output. WindowManager::monitors is a passive cache, only
refreshed when a backend re-queries and calls set_monitors again - the
IPC handler mutated the split map directly but never triggered that
requery, unlike set_output_position's own drain site, which already
pushes a "just go recompute" event after applying.

Makes it a proper queued cross-boundary request instead, the same shape
as every other backend-owned effect on this socket: WindowManager::
request_monitor_split/drain_monitor_split_requests, dispatch queues
instead of mutating, the udev backend's poll drains it, applies via
set_monitor_split, and pushes the same recompute event. srd.monitor.
split's Lua config-time path is untouched - it runs before the very
first startup query, so it never had this problem.

6 files changed, 94 insertions, 26 deletions

diff --git a/crates/core/src/manager/mod.rs b/crates/core/src/manager/mod.rs
index bb22533..c9c77b2 100644
--- a/crates/core/src/manager/mod.rs
+++ b/crates/core/src/manager/mod.rs
@@ -103,6 +103,22 @@ pub struct WindowManager {
     /// [`MonitorSplit`]'s own doc comment for what this deliberately does
     /// and does not give a client (no new `wl_output`).
     monitor_splits: HashMap<String, MonitorSplit>,
+    /// Same cross-boundary-request pattern as `output_position_requests`
+    /// above - an IPC `set_monitor_split` dispatch (the live CLI/IPC path
+    /// for `srd.monitor.split`) mutating `monitor_splits` directly is not
+    /// enough on its own: `monitors` above is a passive cache, only
+    /// refreshed when a backend re-queries and calls `set_monitors` again
+    /// (a real hotplug, or another queued request's own drain site pushing
+    /// the same "just go recompute" `MonitorAdded` event - see `output_
+    /// position_requests`' own drain site for the exact precedent). A
+    /// direct mutation with nothing to trigger that requery left `srd
+    /// monitors` reporting the pre-split layout indefinitely, live-
+    /// reproduced the first time this was tried: `{"ok":true}` came back,
+    /// but the very next `srd monitors` still showed one whole, unsplit
+    /// output. Queued here instead so the backend's own drain site can
+    /// apply the split *and* push that same recompute signal, exactly like
+    /// `output_position_requests` already does.
+    monitor_split_requests: Vec<(String, u32, bool)>,
     /// `srd.monitor.scale(name, factor)` requests, by connector name --
     /// read once by a backend when it brings a head up (startup, hotplug,
     /// or re-enable), so a physically large, low-DPI monitor can run
@@ -454,6 +470,7 @@ impl WindowManager {
             output_enable_requests: Vec::new(),
             disabled_monitors: HashMap::new(),
             monitor_splits: HashMap::new(),
+            monitor_split_requests: Vec::new(),
             monitor_scales: HashMap::new(),
             lock_requested: false,
             capture_requests: Vec::new(),
diff --git a/crates/core/src/manager/monitors.rs b/crates/core/src/manager/monitors.rs
index bff3bc9..c79f76f 100644
--- a/crates/core/src/manager/monitors.rs
+++ b/crates/core/src/manager/monitors.rs
@@ -233,6 +233,24 @@ impl WindowManager {
         self.monitor_splits.get(name).copied()
     }
 
+    /// Queues a live `srd dispatch set output split` request - see
+    /// `monitor_split_requests`' own doc comment for why this can't just
+    /// call `set_monitor_split` directly from the IPC dispatch handler.
+    /// Same "replace, don't accumulate" per-name semantics as `request_
+    /// output_position`.
+    pub fn request_monitor_split(&mut self, name: String, parts: u32, rows: bool) {
+        self.monitor_split_requests.retain(|(existing, _, _)| *existing != name);
+        self.monitor_split_requests.push((name, parts, rows));
+    }
+
+    /// [`Self::drain_output_position_requests`]'s counterpart for split
+    /// requests - the backend applies each via `set_monitor_split` and
+    /// pushes its own "just go recompute" event afterward, same as that
+    /// function's own drain site.
+    pub fn drain_monitor_split_requests(&mut self) -> Vec<(String, u32, bool)> {
+        std::mem::take(&mut self.monitor_split_requests)
+    }
+
     /// `srd.monitor.scale(name, factor)` - a backend applies this the
     /// next time it brings connector `name`'s head up (startup, hotplug,
     /// or re-enable). `factor <= 0.0` clears any existing override rather
diff --git a/crates/platform/src/ipc/dispatch.rs b/crates/platform/src/ipc/dispatch.rs
index 665cb59..918b7e0 100644
--- a/crates/platform/src/ipc/dispatch.rs
+++ b/crates/platform/src/ipc/dispatch.rs
@@ -322,17 +322,17 @@ pub(crate) fn handle_request(line: &[u8], wm: &std::rc::Rc<std::cell::RefCell<Wi
         // "rows":<bool, optional, default false>}` - the live CLI/IPC path
         // for `srd.monitor.split(name, parts, direction)` (`crates/config/
         // src/engine/general.rs`'s own `fn_monitor_split`), which until now
-        // only ever ran once at config load. `WindowManager::
-        // set_monitor_split` just mutates `monitor_splits`, and every
-        // backend's own `monitors()` already reads that map fresh on every
-        // single call (see the udev platform's own `monitors()`) - so,
-        // unlike `set_output_position`/`set_output_enabled` above, this
-        // needs no queue-and-drain at all: the very next `monitors()` query
-        // already reflects it. `parts` <= 1 clears an existing split, same
-        // as the Lua function. Same "resolve id to a name first" fallback
-        // `set_output_enabled` above already uses, since a caller working
-        // from a numeric id shouldn't have to look the name up itself
-        // first just to turn around and split it.
+        // only ever ran once at config load. Queued via `request_monitor_
+        // split`, same cross-boundary "core has no way to trigger its own
+        // requery" reasoning as `set_output_position` above - see
+        // `WindowManager::monitor_split_requests`'s own doc comment for the
+        // real, live-reproduced staleness bug that came from calling
+        // `set_monitor_split` directly here on a first attempt. `parts` <=
+        // 1 clears an existing split, same as the Lua function. Same
+        // "resolve id to a name first" fallback `set_output_enabled` above
+        // already uses, since a caller working from a numeric id shouldn't
+        // have to look the name up itself first just to turn around and
+        // split it.
         "set_monitor_split" => {
             let name = match req.get("name").and_then(|v| v.as_str()) {
                 Some(name) => Some(name.to_string()),
@@ -343,7 +343,7 @@ pub(crate) fn handle_request(line: &[u8], wm: &std::rc::Rc<std::cell::RefCell<Wi
                 return (err("missing parts"), false);
             };
             let rows = req.get("rows").and_then(|v| v.as_bool()).unwrap_or(false);
-            wm.borrow_mut().set_monitor_split(name, parts as u32, rows);
+            wm.borrow_mut().request_monitor_split(name, parts as u32, rows);
             (ok(), true)
         }
         // `{"cmd":"capture_workspace","id":<workspace id>,"path":<string>,
diff --git a/crates/platform/src/ipc/tests.rs b/crates/platform/src/ipc/tests.rs
index 591be56..109a9ec 100644
--- a/crates/platform/src/ipc/tests.rs
+++ b/crates/platform/src/ipc/tests.rs
@@ -426,10 +426,15 @@ fn set_output_enabled_with_neither_name_nor_a_resolvable_id_errors() {
 }
 
 #[test]
-fn set_monitor_split_accepts_a_name_directly_and_applies_immediately() {
-    // Unlike `set_output_position`/`set_output_enabled`, this one is a
-    // plain `WindowManager` mutation with nothing to drain - the very
-    // next `monitor_split` read already reflects it.
+fn set_monitor_split_accepts_a_name_directly_and_queues_a_request() {
+    // Queued, not applied directly - see `WindowManager::monitor_split_
+    // requests`'s own doc comment for why a direct mutation here left
+    // `srd monitors` reporting the stale, unsplit layout indefinitely
+    // (nothing re-triggers `monitors`' own passive cache). Only the
+    // backend that owns real output hardware can call `set_monitor_split`
+    // and push the matching recompute event, so this test - run from the
+    // platform crate alone, with no such backend - can only observe the
+    // queued request, not the applied state.
     let dir = tempfile::tempdir().unwrap();
     let mut server = IpcServer::bind_in(dir.path(), "test").unwrap();
     let wm = Rc::new(RefCell::new(WindowManager::new()));
@@ -441,9 +446,7 @@ fn set_monitor_split_accepts_a_name_directly_and_applies_immediately() {
     server.poll(&wm);
     let _ = read_line(&mut reader);
 
-    let split = wm.borrow().monitor_split("eDP-1").unwrap();
-    assert_eq!(split.parts, 2);
-    assert!(!split.rows);
+    assert_eq!(wm.borrow_mut().drain_monitor_split_requests(), vec![("eDP-1".to_string(), 2, false)]);
 }
 
 #[test]
@@ -459,26 +462,35 @@ fn set_monitor_split_resolves_an_id_to_its_name() {
     server.poll(&wm);
     let _ = read_line(&mut reader);
 
-    let split = wm.borrow().monitor_split("HDMI-A-1").unwrap();
-    assert_eq!(split.parts, 3);
-    assert!(split.rows);
+    assert_eq!(wm.borrow_mut().drain_monitor_split_requests(), vec![("HDMI-A-1".to_string(), 3, true)]);
 }
 
 #[test]
-fn set_monitor_split_with_one_part_clears_an_existing_split() {
+fn set_monitor_split_replaces_a_still_pending_request_for_the_same_name() {
+    // Same "last write wins per name" semantics as `request_output_
+    // position` - only the latest requested split for a given output
+    // matters if several arrive before the backend's next drain.
     let dir = tempfile::tempdir().unwrap();
     let mut server = IpcServer::bind_in(dir.path(), "test").unwrap();
     let wm = Rc::new(RefCell::new(WindowManager::new()));
     wm.borrow_mut().set_monitors(vec![srdwm_core::Monitor::new(0, "eDP-1", srdwm_core::Rect::new(0, 0, 1920, 1080))]);
-    wm.borrow_mut().set_monitor_split("eDP-1".to_string(), 2, false);
 
     let mut client = UnixStream::connect(&server.path).unwrap();
     let mut reader = std::io::BufReader::new(client.try_clone().unwrap());
-    client.write_all(b"{\"cmd\":\"set_monitor_split\",\"name\":\"eDP-1\",\"parts\":1}\n").unwrap();
+    client.write_all(b"{\"cmd\":\"set_monitor_split\",\"name\":\"eDP-1\",\"parts\":2,\"rows\":false}\n").unwrap();
     server.poll(&wm);
     let _ = read_line(&mut reader);
 
-    assert!(wm.borrow().monitor_split("eDP-1").is_none());
+    // A fresh connection - each `srd dispatch` is its own one-shot socket
+    // connection in practice, not a second write on an already-answered
+    // one (which the server closes after replying).
+    let mut client2 = UnixStream::connect(&server.path).unwrap();
+    let mut reader2 = std::io::BufReader::new(client2.try_clone().unwrap());
+    client2.write_all(b"{\"cmd\":\"set_monitor_split\",\"name\":\"eDP-1\",\"parts\":1}\n").unwrap();
+    server.poll(&wm);
+    let _ = read_line(&mut reader2);
+
+    assert_eq!(wm.borrow_mut().drain_monitor_split_requests(), vec![("eDP-1".to_string(), 1, false)]);
 }
 
 #[test]
diff --git a/crates/wayland/src/udev/platform.rs b/crates/wayland/src/udev/platform.rs
index 4012e3c..37abf5d 100644
--- a/crates/wayland/src/udev/platform.rs
+++ b/crates/wayland/src/udev/platform.rs
@@ -603,6 +603,19 @@ impl Platform for UdevPlatform {
                 self.pending.borrow_mut().push(CoreEvent::MonitorAdded(srdwm_core::Monitor::new(0, "", srdwm_core::Rect::new(0, 0, 0, 0))));
             }
         }
+        // Applies any `srd dispatch set output split` IPC requests queued
+        // since the last poll - see `WindowManager::monitor_split_
+        // requests`'s own doc comment for why this needs the same "apply,
+        // then push a recompute event" shape `set_output_position`'s own
+        // drain just above uses, rather than `set_monitor_split` being
+        // called straight from the IPC dispatch handler.
+        let split_requests = self.state.wm.borrow_mut().drain_monitor_split_requests();
+        if !split_requests.is_empty() {
+            for (name, parts, rows) in split_requests {
+                self.state.wm.borrow_mut().set_monitor_split(name, parts, rows);
+            }
+            self.pending.borrow_mut().push(CoreEvent::MonitorAdded(srdwm_core::Monitor::new(0, "", srdwm_core::Rect::new(0, 0, 0, 0))));
+        }
         // Applies any `srd set_output_enabled` IPC requests queued since
         // the last poll - `disable_connector_by_name`/`enable_connector_
         // by_name` already push their own `MonitorRemoved`/`MonitorAdded`
diff --git a/docs/TODO.md b/docs/TODO.md
index f58f1df..67ff09a 100644
--- a/docs/TODO.md
+++ b/docs/TODO.md
@@ -13,6 +13,14 @@ that has the full story. Keep this list current as items close or open;
 update the source doc's own entry too, don't let this drift into a
 second stale copy the way `PANEL_SUPPORT_TODO.md` did.
 
+## Real bug in the same-day monitor-split live-exposure: `srd monitors` never reflected it (2026-08-27)
+
+Tried it live for the user right after shipping it: `srd dispatch set output split eDP-1 2 columns` returned `{"ok":true}`, but the very next `srd monitors` still showed one whole, unsplit output. Root cause: `WindowManager::monitors` is a passive cache, only refreshed when a backend re-queries and calls `set_monitors` again (a real hotplug, or another request's own drain site pushing a `MonitorAdded` "just go recompute" event - see `output_position_requests`' own drain site, which already does exactly this after applying a position). The IPC handler called `set_monitor_split` directly, mutating the split map correctly but never triggering that requery - the same class of bug `output_position_requests` was already built to avoid, just missed when this feature was added.
+
+Fixed by making it a proper queued cross-boundary request like every other backend-owned effect on this socket: new `WindowManager::request_monitor_split`/`drain_monitor_split_requests` (`monitor_split_requests`, replace-not-accumulate per name, same as `request_output_position`), IPC dispatch now queues instead of mutating directly, and the udev backend's own poll drains it, applies via `set_monitor_split`, and pushes the same recompute event `set_output_position`'s drain site does. `srd.monitor.split`'s Lua config-time path is untouched - it runs before the very first startup `monitors()` query, so it never had this staleness problem to begin with.
+
+Full workspace build/test/clippy clean (34 platform tests). Live-verified this time before calling it done, not just build-clean - exactly the mistake this entry itself is about.
+
 ## Fake-monitor incident, root-caused and fixed jointly with the AGS peer session (2026-08-27)
 
 Follow-up to the incident entry directly below. `dotfiles-1a` (AGS) confirmed from AGS's own session log (`~/.local/state/wm-session-*.log` - not `~/.cache/ags/ags.log`, which was stale) that the X-position jump was AGS's own remembered "extend-left" layout restore firing twice, once per fake monitor appearing (`arrange()` places the primary at 0,0 and walks the rest `left -= width`, then normalizes - correct behavior for a *real* second monitor; the only fault was a fake one being in the arrangeable set at all). Fixed on AGS's side: `readArrangeable()` now filters `!m.split && !m.virtual`. Their match was keyed on connector *name*, not index, ruling out the index-shift half of my own original theory.