The timeout value in dunst wraps around

[hbakken]

dunst internally stores timeouts in microseconds, in a signed 32 bit integer, which wraps around at 2147484, which corresponds to ~40 minutes. Having timeouts longer than that is not unreasonable, so the wrapping should be fixed.

notify-send -t 2147483 foo
will stay on the screen for 35 minutes, while
notify-send -t 2147484 foo
will use the default (for me, 10s)

Installation info

• Version: Dunst - A customizable and lightweight notification-daemon 1.3.2
• Install type: package
• Distro and version: Fedora 30

[tsipinakis]

I prompted @hbakken to created this issue after a short discussion about the issue on IRC.

What completely flew over my head in that discussion was that it was overflowing on the 32 bit boundary not the 64 bit one. After working that out it was dead simple to pinpoint this, and it got me wondering how I missed that in the review 🤦‍♂️.

@bebehei Posting the patch here since you self-assigned this, not sure if you're preparing a PR with anything else.

diff --git a/src/dbus.c b/src/dbus.c
index 741a1c5..b2fc2cb 100644
--- a/src/dbus.c
+++ b/src/dbus.c
@@ -285,8 +285,9 @@ static struct notification *dbus_message_to_notification(const gchar *sender, GV
                 }
         }
 
-        if (timeout >= 0)
-                n->timeout = timeout * 1000;
+        if (timeout >= 0) {
+                n->timeout = ((gint64)timeout) * 1000;
+        }
 
         g_variant_unref(hints);
         g_variant_type_free(required_type);

[bebehei]

Thanks for the hint. I would have searched pretty long. I'll write a DBus test for this probably.