Persisting a Scene and analysis results

[Manoj Plakal <plakal@cs.wisc.edu>]

Right now, doing whole-program analysis
in Soot is an expensive affair and 
experimenting with whole-program passes
can be slow due to the startup cost
of building the scene and call graph
and points-to graph from scratch.

Has anyone ever tried serializing an
entire Scene after it has been
constructed, including the points-to
analysis, call graph and side-effect
analysis? And then deserializing it
on subsequent executions?

Seems like it would be a big win
to do this once for each benchmark,
and then simply read the stored state 
on subsequent runs where we experiment
with our own passes.

What are the issues in
persisting and later reconstruction
of an entire Soot Scene?

Serialized scenes of standard benchmarks,
for standard Java library versions,
including expensively-accurate analysis graphs
of all flavors (call, points-to, CFG, PDG etc),
might be valuable downloads for the 
entire community (much like program
execution traces are useful to the 
simulation community).

Manoj