core: optimize struct layouts, eliminate branches, and fix event loop dispatch

- Fix Wayland event loop lockout by properly decoupling read arming and adding poll-backed dispatching
- Support standalone display lifecycle without requiring caller to manage a loop
- Eliminate alignment holes across yasl_surface, yasl_display, seat, touch_point, and conn_drag using pahole
- Replace branching state/button decoders with compile-time locked LUTs and branchless bitwise operations
- Accelerate UTF-8 validation with 64-bit word scanning for ASCII runs

1 files changed, 99 insertions, 19 deletions

diff --git a/src/wayland/surface.c b/src/wayland/surface.c
index da334e4..5138201 100644
--- a/src/wayland/surface.c
+++ b/src/wayland/surface.c
@@ -83,6 +83,14 @@ static int display_ready(int fd, short revents, void *ud)
         else if (wl_display_read_events(c->wl) < 0)
             revents |= POLLERR;
     }
+    if (revents & POLLOUT) {
+        if (wl_display_flush(c->wl) >= 0) {
+            if (c->loop && c->watch > 0)
+                yasl_loop_set_fd_events(c->loop, c->watch, POLLIN);
+        } else if (errno != EAGAIN) {
+            display_lose(c);
+        }
+    }
     if (wl_display_dispatch_pending(c->wl) < 0 ||
         (revents & (POLLERR | POLLHUP | POLLNVAL)))
         display_lose(c);
@@ -90,35 +98,92 @@ static int display_ready(int fd, short revents, void *ud)
     return 0;
 }
 
-void yasl_display_dispatch(struct yasl_display *c)
-{
-    if (!c || c->lost || c->reading)
-        return;
-    display_enter(c);
-    if (wl_display_dispatch_pending(c->wl) < 0)
-        display_lose(c);
-    display_leave(c);
-}
-
-static void display_arm(struct yasl_display *c)
+static bool display_arm(struct yasl_display *c)
 {
+    bool had_pending = false;
     while (!c->reading && wl_display_prepare_read(c->wl) != 0) {
+        had_pending = true;
         if (wl_display_dispatch_pending(c->wl) < 0) {
             display_lose(c);
-            return;
+            return false;
         }
     }
     c->reading = true;
     short events = POLLIN;
     if (wl_display_flush(c->wl) < 0) {
         if (errno != EAGAIN) {
+            display_cancel_read(c);
             display_lose(c);
-            return;
+            return false;
         }
         events |= POLLOUT;
     }
-    if (c->loop)
+    if (c->loop && c->watch > 0)
         yasl_loop_set_fd_events(c->loop, c->watch, events);
+    return had_pending;
+}
+
+static bool display_has_immediate_work(const struct yasl_display *c)
+{
+    if (!c || c->lost) return false;
+    for (struct yasl_surface *w = c->surfaces; w; w = w->conn_next) {
+        if (w->closed) continue;
+        if (w->input.count > 0)
+            return true;
+        if (w->want_draw && !w->frame_cb && w->configured)
+            return true;
+    }
+    return false;
+}
+
+int yasl_display_dispatch_timeout(struct yasl_display *c, int timeout_ms)
+{
+    if (!c || c->lost)
+        return -1;
+
+    display_enter(c);
+
+    if (wl_display_dispatch_pending(c->wl) < 0) {
+        display_lose(c);
+        display_leave(c);
+        return -1;
+    }
+
+    bool had_pending = display_arm(c);
+    if (c->lost) {
+        display_leave(c);
+        return -1;
+    }
+
+    if (had_pending && timeout_ms != 0 && display_has_immediate_work(c)) {
+        timeout_ms = 0;
+    }
+
+    int n = 0;
+    if (c->loop) {
+        n = yasl_loop_step(c->loop, timeout_ms);
+    }
+
+    if (c->reading) {
+        display_cancel_read(c);
+    }
+
+    if (wl_display_dispatch_pending(c->wl) < 0) {
+        display_lose(c);
+        display_leave(c);
+        return -1;
+    }
+
+    display_leave(c);
+    return n;
+}
+
+void yasl_display_dispatch(struct yasl_display *c)
+{
+    if (!c || c->lost)
+        return;
+    int timeout = display_has_immediate_work(c) ? 0 : -1;
+    yasl_display_dispatch_timeout(c, timeout);
 }
 
 void yasl_display_flush(struct yasl_display *c)
@@ -130,8 +195,10 @@ void yasl_display_flush(struct yasl_display *c)
         next = w->conn_next;
         surface_draw_frame(w);
     }
-    if (!c->lost)
-        display_arm(c);
+    if (!c->lost && c->wl) {
+        if (wl_display_flush(c->wl) < 0 && errno != EAGAIN)
+            display_lose(c);
+    }
     display_leave(c);
 }
 
@@ -163,6 +230,10 @@ static void display_free(struct yasl_display *c)
     if (c->compositor) wl_compositor_destroy(c->compositor);
     if (c->registry) wl_registry_destroy(c->registry);
     if (c->wl) wl_display_disconnect(c->wl);
+    if (c->owns_loop && c->loop) {
+        yasl_loop_destroy(c->loop);
+        c->loop = NULL;
+    }
     free(c->app_id);
     free(c);
 }
@@ -171,7 +242,13 @@ struct yasl_display *yasl_display_open(struct yasl_loop *loop, const struct yasl
 {
     struct yasl_display *c = calloc(1, sizeof(*c));
     if (!c) return NULL;
-    c->loop = loop;
+    if (!loop) {
+        c->loop = yasl_loop_create();
+        c->owns_loop = true;
+    } else {
+        c->loop = loop;
+    }
+    if (!c->loop) goto err;
     if (spec) {
         c->cbs = spec->cbs;
         c->ud = spec->ud;
@@ -189,7 +266,7 @@ struct yasl_display *yasl_display_open(struct yasl_loop *loop, const struct yasl
 
     surface_init_cursor(c);
     if (c->loop) {
-        c->watch = yasl_loop_add_fd(loop, wl_display_get_fd(c->wl), POLLIN, display_ready, c);
+        c->watch = yasl_loop_add_fd(c->loop, wl_display_get_fd(c->wl), POLLIN, display_ready, c);
         if (c->watch <= 0) { c->watch = 0; goto err; }
     }
     if (display_input_init(c) < 0)
@@ -371,7 +448,10 @@ struct yasl_display *yasl_surface_display(const struct yasl_surface *w)
 
 void yasl_surface_schedule_redraw(struct yasl_surface *w)
 {
-    if (w) w->want_draw = true;
+    if (!w || w->closed) return;
+    w->want_draw = true;
+    if (w->conn && w->conn->loop)
+        yasl_loop_kick(w->conn->loop);
 }
 
 struct yasl_vector2i yasl_surface_size(const struct yasl_surface *w)