Keep state variables occurring in INIT in COI analysis #179
Conversation
|
FYI, since this is a simple change, I wasn't planning to run on the cluster. However, I'll be running lots of other jobs this week so if this introduces performance issues, I'll notice after it's merged. |
|
@makaimann, I agree with your comments on potential difficulties in analyzing |
That's a good question. I think a conjunction of equalities is the most common case but is too restrictive in general. One possible solution is to just have a flag that gets set if a more general constraint is added and then we can check to flag to see how much simplification we can do. What do you think about that? I was thinking we could address that later and just merge this PR for now though if that works for you? |
|
Yes of course, sounds good, let's merge that PR. |
As discussed elsewhere, there was an issue with state variables being removed from the system. Since COI ignores the initial state constraints, some removed variables may appear in
init. When those are removed, the transition system becomes invalid (i.e. it contains non-variables in init). This prevented us from use COI with some techniques. This PR simply doesn't remove those state variables. This should not impact performance too much because those variables only occur in init.If it ends up affecting performance negatively, we can look into analyzing init to remove the state variables. This is not so simple though, because even for functional transition systems
initcan be arbitrary formulas over current state variables.