Document how log origin and FSM type impact invariants and model

[GoogleCodeExporter]

Logs could be created in two different ways: an event could be logged as it's 
starting (let's call it log-on-start) or as it's finishing (log-on-finish). For 
an event "read-file," the former might mean a "read-file" is logged as soon as 
some reading function is entered, whereas a log-on-finish system would create 
the log line as when the file reading finished. This has an impact on how 
Perfume invariants and models are interpreted.

For log-on-start, invariant bounds are between the start of event A and the 
start of event B (so they don't include the metric of event B). In addition, 
metrics for one node/event are in its outgoing edges, and the last event in any 
trace will never have a metric. This is because a metric for an event means 
"metric value difference between when the next event starts and when I started."

For log-on-finish, invariant bounds are between the end of event A and the end 
of event B (so they don't include the metric of event A). In addition, metrics 
for one node/event are in its incoming edges, and the first event in any trace 
will never have a metric. This is because a metric for an event means "metric 
value difference between when I finished and when the previous event finished."

Document these important distinctions somewhere.

Original issue reported on code.google.com by [email protected] on 24 Apr 2014 at 11:00