Re: PATCH: Fix "invisible wall" problem with relative mouse mode
- From: Anthony Liguori <anthony codemonkey ws>
- To: "Daniel P. Berrange" <berrange redhat com>
- Cc: gtk-vnc-list gnome org
- Subject: Re: PATCH: Fix "invisible wall" problem with relative mouse mode
- Date: Tue, 24 Feb 2009 08:25:33 -0600
Daniel P. Berrange wrote:
A number of bugs have conspired together to make relative mouse mode not
work very well. The core problem is that when using the relative mode the
mouse pointer will appear to hit an "invisible wall" before the true edge
of the screen.
The problems start occur when we do the mouse grab and fall out from
that...
- We passed our window as the "confine-to" arg for the gdk_pointer_grab.
This meant that the client desktop pointer would stop moving / generating
events when we hit the edge of our window. In relative mode we need to
get continuous events no matter where the pointer is, since we're after
relative changes, not absolute coords.
What is supposed to fix this is auto warping. If you detect that you're
close to the edge of the screen, it should warp you toward the center of
the window.
When I first implemented this, I observed that confining the pointer to
the desktop instead of the window merely made the problem harder to
reproduce. You still stop getting pointer events if the mouse hits the
edge of the screen so under the right circumstances, you'll still hit a
wall.
The warping was supposed to be the proper fix although apparently it's
not working well enough.
Regards,
Anthony Liguori
- The 'owner-events' arg for gdk_pointer_grab was TRUE instead of FALSE.
With the first bug fixed, this meant that when the pointer was outside
the VNC window, events would be sent to another window in the app. We
always want events ourselves, no matter what window contains the grabbed
pointer
- In non-scaled mode, we discarded any mouse events whose absolute coords
was outside the range of the server desktop size. This meant that the
mouse was again effectively contained to our window, and not free to
roam, and not allowing relative mouse events to be handled correctly.
It took quite a while for me to understand exactly what was going on here
so in this patch I've added alot of comments explaining what each chunk
of code is trying to do
I've tested this patch in all 4 scenarios
- Mouse in relative mode + centered rendering of desktop
- Mouse in relative mode + scaled rendering of desktop
- Mouse in absolute mode + centered rendering of desktop
- Mouse in absolute mode + scaled rendering of desktop
Daniel
diff --git a/src/vncdisplay.c b/src/vncdisplay.c
index 5a25b91..f1082da 100644
--- a/src/vncdisplay.c
+++ b/src/vncdisplay.c
@@ -439,14 +439,22 @@ static void do_pointer_grab(VncDisplay *obj, gboolean quiet)
if (!priv->grab_keyboard)
do_keyboard_grab(obj, quiet);
+ /*
+ * For relative mouse to work correctly when grabbed we need to
+ * allow the pointer to move anywhere on the local desktop, so
+ * use NULL for the 'confine_to' argument. Furthermore we need
+ * the coords to be reported to our VNC window, regardless of
+ * what window the pointer is actally over, so use 'FALSE' for
+ * 'owner_events' parameter
+ */
gdk_pointer_grab(GTK_WIDGET(obj)->window,
- TRUE,
+ FALSE, /* All events to come to our window directly */
GDK_POINTER_MOTION_MASK |
GDK_BUTTON_PRESS_MASK |
GDK_BUTTON_RELEASE_MASK |
GDK_BUTTON_MOTION_MASK |
GDK_SCROLL_MASK,
- GTK_WIDGET(obj)->window,
+ NULL, /* Allow cursor to move over entire desktop */
priv->remote_cursor ? priv->remote_cursor : priv->null_cursor,
GDK_CURRENT_TIME);
priv->in_pointer_grab = TRUE;
@@ -551,15 +559,33 @@ static gboolean scroll_event(GtkWidget *widget, GdkEventScroll *scroll)
return TRUE;
}
+
+/*
+ * There are several scenarios to considier when handling client
+ * mouse motion events:
+ *
+ * - Mouse in relative mode + centered rendering of desktop
+ * - Mouse in relative mode + scaled rendering of desktop
+ * - Mouse in absolute mode + centered rendering of desktop
+ * - Mouse in absolute mode + scaled rendering of desktop
+ *
+ * Once scaled / offset, absolute mode is easy.
+ *
+ * Relative mode has a couple of special complications
+ *
+ * - Need to turn client absolute events into a delta
+ * - Need to warp local pointer to avoid hitting a wall
+ */
static gboolean motion_event(GtkWidget *widget, GdkEventMotion *motion)
{
VncDisplayPrivate *priv = VNC_DISPLAY(widget)->priv;
- int dx, dy;
int ww, wh;
if (priv->gvnc == NULL || !gvnc_is_initialized(priv->gvnc))
return FALSE;
+ /* In relative mode, only move the server mouse pointer
+ * if the client grab is active */
if (!priv->absolute && !priv->in_pointer_grab)
return FALSE;
@@ -568,11 +594,14 @@ static gboolean motion_event(GtkWidget *widget, GdkEventMotion *motion)
gdk_drawable_get_size(widget->window, &ww, &wh);
+ /* First apply adjustments to the coords in the motion event */
if (priv->allow_scaling) {
double sx, sy;
sx = (double)priv->fb.width / (double)ww;
sy = (double)priv->fb.height / (double)wh;
+ /* Scaling the desktop, so scale the mouse coords
+ * by same ratio */
motion->x *= sx;
motion->y *= sy;
} else {
@@ -583,21 +612,28 @@ static gboolean motion_event(GtkWidget *widget, GdkEventMotion *motion)
if (wh > priv->fb.height)
mh = (wh - priv->fb.height) / 2;
+ /* Not scaling, drawing the desktop centered
+ * in the larger window, so offset the mouse
+ * coords to match centering */
motion->x -= mw;
motion->y -= mh;
-
- if (motion->x < 0 || motion->x >= priv->fb.width ||
- motion->y < 0 || motion->y >= priv->fb.height)
- return FALSE;
}
- if (!priv->absolute && priv->in_pointer_grab) {
+ /* Next adjust the real client pointer */
+ if (!priv->absolute) {
GdkDrawable *drawable = GDK_DRAWABLE(widget->window);
GdkDisplay *display = gdk_drawable_get_display(drawable);
GdkScreen *screen = gdk_drawable_get_screen(drawable);
int x = (int)motion->x_root;
int y = (int)motion->y_root;
+ /* In relative mode check to see if client pointer hit
+ * one of the screen edges, and if so move it back by
+ * 200 pixels. This is important because the pointer
+ * in the server doesn't correspond 1-for-1, and so
+ * may still be only half way across the screen. Without
+ * this warp, the server pointer would thus appear to hit
+ * an invisible wall */
if (x == 0) x += 200;
if (y == 0) y += 200;
if (x == (gdk_screen_get_width(screen) - 1)) x -= 200;
@@ -611,11 +647,20 @@ static gboolean motion_event(GtkWidget *widget, GdkEventMotion *motion)
}
}
+ /* Finally send the event to server */
if (priv->last_x != -1) {
+ int dx, dy;
if (priv->absolute) {
dx = (int)motion->x;
dy = (int)motion->y;
+
+ /* Drop out of bounds motion to avoid upsetting
+ * the server */
+ if (dx < 0 || dx >= priv->fb.width ||
+ dy < 0 || dy >= priv->fb.height)
+ return FALSE;
} else {
+ /* Just send the delta since last motion event */
dx = (int)motion->x + 0x7FFF - priv->last_x;
dy = (int)motion->y + 0x7FFF - priv->last_y;
}
[
Date Prev][
Date Next] [
Thread Prev][
Thread Next]
[
Thread Index]
[
Date Index]
[
Author Index]