Created attachment 362827 [details] [review]
Suggested fix

GPollFileMonitor is consuming memory that is never freed. When schedule_poll_timeout() is called it creates new timeout source (with ref_count = 1 initially) and this newly created source is attached to main context (ref_count is incremented to 2):

static void
schedule_poll_timeout (GPollFileMonitor* poll_monitor)
{
   poll_monitor->timeout = g_timeout_source_new_seconds (POLL_TIME_SECS);
   g_source_set_callback (poll_monitor->timeout, poll_file_timeout, poll_monitor, NULL);
   g_source_attach (poll_monitor->timeout, g_main_context_get_thread_default ());
}

But once the timeout is called the initial ref_count is not decremented so when the main context is done with dispatching the ref_count is decremented to 1 and is never freed. This leads in constant memory consumption increase and also high CPU usage as every time the main_context_check() or main_context_prepare() is called it iterates through sources list which gets bigger and bigger:

Also poll_monitor->timeout is set to FALSE in poll_file_timeout() though timeout is of type GSource*.

Also there is a read-after-free error in file monitor cancel routine, timeout is destroyed and then unref'ed. I believe unreferencing it should be enough (and should be also safer if someone is still referencing the source).

static gboolean
g_poll_file_monitor_cancel (GFileMonitor* monitor)
{ 
   GPollFileMonitor *poll_monitor = G_POLL_FILE_MONITOR (monitor);
  
   if (poll_monitor->timeout)
     { 
       g_source_destroy (poll_monitor->timeout);
       g_source_unref (poll_monitor->timeout);
       poll_monitor->timeout = NULL;
     }
   
   return TRUE;
}

Suggested fix attached.