Bug 566351

I found a more consistent way to reproduce this issue.  The bug is
in Xorg::Xext/sync.c, which is exposed by the behavior of g-p-m.

Steps to Reproduce:
    1. change idle timeout to 1 minute, so it's easier to reproduce the issue. 
       gnome-screensaver-preferences 
            "Regard the computer as idle after"      1 minute 
       gnome-power-preferences 
            "Put display to sleep when inactive for" 1 minute 
 
    2. kill gnome-power-manager, and start  a new one by 
        p='idle for|idle dim|idle dim|active for' 
        b=/usr/bin/gnome-power-manager 
        $b --verbose | egrep "$p" 

    3. In F11, hit a key, idle for 35 sec, hit a key, idle for
       65 sec, hit a key, idle for 65 sec, hit a key.  Wait 10
       minutes; g-ss will not kick in, thus issue reproduced.

       The delay is actually (in F11)
           for i in 30 60 60; do
               d=$i+5
               idle $d sec
               hit a key

       In F12, use
           for i in 10 20 40 60; do
               d=$i+5
               idle $d sec
               hit a key

       g-p-m should output something like (in F12)
             - resetting idle dim time to 10s
             - increasing idle dim time to 20s
             - increasing idle dim time to 40s
             - increasing idle dim time to 80s


       Here is some bash-ish mouse script (use mouse copy and paste to
       your terminal window) to help you get the right timing:
            t() {
                n=$1
                while [ $n -gt 0 ]; do
                    printf "$(date) $n   \r"
                    let n=$n-1
                    sleep 1
                done
                printf "\n"
            }
            t1() {
                for d in $*; do
                    let d1=$d+5
                    t $d1
                done
            }
            # F11
            t1 30 60 60
            # F12
            t1 10 20 40 60

The reason for those funny numbers (10 20 40 60 in F12):
    - g-p-m schedules two alarms with Xext/sync.c:IDLETIME counter:
      1. for DIM    event (idle for 
                           30, 60, 120, 240, ... sec in F11; 
                           10, 20,  40,  80, ... sec in F12)
      2. for NORMAL event (when user is active, kbd or mouse).

    - initially g-p-m uses 10 sec (in F12) for DIM alarm; if the user
      undims screen immediately (within 10 sec) after dimming, the
      DIM time is doubled.  When DIM time (120 sec in F11, 80 sec in 
      F12) is more than 60 sec (g-ss idle time), bug566350 in
      Xorg::Xext/sync.c kicks in, thus no more idle events are sent
      to g-ss or g-p-m.

    - in F11 we have 30 60 60; the first two (30 60) are for doubling
      DIM time; the last one (60) is for g-ss timeout.

    - in F12 we have 10 20 40 60; the first three (10 20 40) are for 
      doubling DIM time; the last one (60) is for g-ss timeout.


Source code for doubling dim timeout in g-p-m.
Note: the design and implementation of this code segment is sound; I
put it here only to show the doubling behavior.

    gpm_backlight_notify_system_idle_changed (gpm-backlight.c) 
        egg_debug ("we have just been idle for %lfs", elapsed);  
        /* The user immediatly undimmed the screen!  
         * We should double the timeout to avoid this happening again */  
        if (elapsed < 10) {  
                /* double the event time */  
                backlight->priv->idle_dim_timeout *= 2.0;  
                egg_debug ("increasing idle dim time to %is",
                            backlight->priv->idle_dim_timeout);  
                gpm_idle_set_timeout_dim (backlight->priv->idle,
                            backlight->priv->idle_dim_timeout);  
        }  
        /* We reset the dimming after 2 minutes of idle,  
         * as the user will have changed tasks */  
        if (elapsed > 2*60) {  
                /* reset back to our default dimming */  
                backlight->priv->idle_dim_timeout =  
                        gconf_client_get_int (backlight->priv->conf,  
                                   GPM_CONF_BACKLIGHT_IDLE_DIM_TIME, NULL);  
                egg_debug ("resetting idle dim time to %is", 
                                backlight->priv->idle_dim_timeout);  
                gpm_idle_set_timeout_dim (backlight->priv->idle, 
                                backlight->priv->idle_dim_timeout);  
        }

I'll close this bug due to lack of response from Adam (maintainer of
Xorg; speaking of role reversal, usually it's the developer who is
eager to close bugs; in this case, the originator sees no reason to
keep bug open for no good reason).  From what I experimented, I think
this bug has been fixed in F13.  Please reopen if you disagree.