Bugs: dmfilth

This is only detectable with many (hundreds) of runs.

The tool will fail if this method is used with exposure-corrected images or if there are negative pixels in the source or background regions.

unix% dmstat "events.fits[sky=region(source.reg)]"
EVENTS_IMAGE(x, y)
    min:        -2.0534541607         @:        ( 4620.5 4868.5 )
    max:        63.77072525           @:        ( 4588.5 4876.5 )
cntrd[log] :    ( 5.7060806841 5.973494067 )
cntrd[phys]:    ( 4586.1486455 4876.2879525 )
sigma_cntrd:    ( 6.3023091173 6.8273372496 )
   good:        70 
   null:        40 

dmfilth is re-initializing the random seed to the same value for each source/background pair in the stack. This makes the filled values highly correlated (if not identical).

The bug also affects the rand=0 case, where the system time is chosen as the seed. dmfilth only gets the system time once, then resets the seed to the same time value for each source.