[Jason House <jhouse@xxxxxxxxx>]

Pavel Mores wrote:
> 
> > ...the effect of the change will probably be to replace a hang with a
> > not-entirely-sensible graph.
> 
> Well, if the change doesn't break handling of correct captures (which it
> shouldn't, according to your description of the change) then I guess it
> should be OK.  Getting a not-entirely-sensible graph for a
> not-entirely-sensible capture seems reasonable to me. ;-)

  I think that surprising a person with a not-entirely-sensible graph is
not quite the right way to do it.  The user should have some idea that
things might not be entirely correct.
  What about adding something like [delta t < 0] to COL_INFO?  That
would give some indication to the user that something
not-entirely-sensible is going on to begin with.  That would help excuse
other tools (such as io-stat) for not working 100% in such a scenario. 
Of course, the tools should strive to work the best way they know how
(and at the very least not hang)... but some kind of advanced warning is
always good.