Re: EventViewer and time.

[Sami Wagiaalla <swagiaal at redhat dot com>]

Well the model I usually use to resolve issues with the interval based
implementation is printf debugging. "How does printf solve this issue ?"

So one thing that we could do is when a trace arrives at one trace
all the traces are step one interval. All are blank but only the concerned
trace gets the event. This resolves some of the issues you stated (wherever
time can be replaced by this index at least).

As for parallel events (multi cpu, multi host, etc), printf eventually 
forces
an oder that is probably arbitrary. We could do the same. The truth is any
assignment of time to an event is very likely to be inaccurate. The same 
goes
for simultaneous events. If we present them as simultaneous, do we 
really know
that for sure ?

One thing that Cagney has suggested in the past is to have buckets in 
the timeline
corresponding to iterations of the core event loop. ie every time the 
event loop makes
a loop it provides a tick. This makes sense because that is the lowest 
resolution that
we can actually guarantee.