severe reliability problems with gdk_input_add

[Vlad Harchev <hvv hippo ru>]

 Hello!

 It seems I'm encountering reliability problems with gdk_input_add - it seems
sometimes Glib (which implements source notification) gives up calling
callback functions, the only way to cure is to kill the application. All my
apps run under RedHat 7.2 on x86 with 512MB RAM with all updates installed,
glib-1.2.10 and gtk-1.2.10.. I'm using it in the 2 apps that are better called
daemons - they use dedicated dummy XVnc server with screen size 20x20 pixels
to which nobody can attach (so I mean the problem is not with XVnc server
since apps start to work just fine if I restart them).
 All input sources in my problems are monitored for 'read' and 'exception'
events. None of my apps wait for 'write' events.
 The sources are fds that correspond to sockets - either listening for
incoming connections or used for data transfer. The clients they
accept/communicate with are both on localhost and on remote computer in the
LAN with reliable cables and switches (so it can't be attributed network
problems).
 The swap partion is never used by more than 30% (it.s 1.5 gb).
 The system load never goes upper than 0.9.
 I'm using RH-compiled kernel from updates.
 
 The callbacks stop from being called after about 1 or more days of the
program's life. Netstat shows that sockets are still opened (or listening
correspondingly).
 Programs do not use threads.

 I thought that it was a problem in my app, but today I encountered it in the
trivial program - definetly bug can't be attributed to my code.
This problem really annoys me!

 Have anybody experienced such problems?
 Does anybody remember fixing bugs regarding the input source callback
support in glib/gdk since the release of gtk-1.2.10 that could fix it?

 Any help is greatly appreciated.

 Thank you in advance.

 Best regards,
  -Vlad