Static cone-of-influence analysis #75

lonsing · 2020-09-01T05:42:31Z

One-time application of cone-of-influence (COI) analysis. We collect state/input variables that appear

in the term that represents the bad-state property and
in the terms that represent all the global constraints that were added to the transition system.

Then we collect all state/input variables that appear in the next-state functions of all previously collected state variables. This process is iterated until saturation.

Finally, we rebuild the transition relation of the transition system by adding all the global constraints and the next-state functions of the collected state-variables.

COI computation is done in the Prover class in the initialization function. Hence COI can basically be combined with any model checking algorithm (i.e. classes inheriting from Prover).

A new public function was added to the TransitionSystem class to rebuild the transition relation based on collected COI information (to make that possible, const qualifiers were removed e.g. from the transition system member variable in the Property class).

Command line usage: ./pono --static-coiwhere verbosity level 1 prints basic information and a statistics summary of remaining state variables, and verbosity level 3 prints in-depth information (recommended for debugging only).

COI currently is only compatible with functional transition system (when called on relational ones, an exception is thrown.)

TO BE ADDED: printing of full witnesses.

makaimann

Looks good to me. Thanks or working on this! I just left a few minor comments that could be addressed if you have a chance. In addition, I'm not sure how I feel about removing all the const specifiers, but it's fine for now. Tbh, the const stuff was causing some headaches in other areas as well. Maybe later we can discuss possible refactors that could clean that up.

I'm going to wait to merge this. I think it would be easier on me to merge the other features first and merge this on top of them.

makaimann · 2020-09-01T14:49:21Z

engines/prover.cpp

+      collect_coi_term(inputvars_in_coi_, sv);
+    }
+
+    new_coi_state_vars.clear();


Do new_coi_state_vars and new_coi_input_vars need to be parameters? It looks to me like they are only used in a local way. At least in this method because they are cleared.

Good catch! This part was left over from a previous version. I will update it.

makaimann · 2020-09-01T15:34:40Z

engines/prover.cpp

+          assert(!ts_.is_next_var(cur));
+          logger.log(3, "collect COI inputvar {}", cur);
+          collect_coi_term(new_coi_input_vars, cur);
+        }


What about the next state variable case? Constraints can have next state variables because if it's an invariant, the constraint is added both over current vars and next vars. Is it safe to ignore those cases because there's always a current state version of the same invariant?

Is it safe to ignore those cases because there's always a current state version of the same invariant?

Yes, that is exactly what I was thinking. That should work since we collect all state variables in all constraints unconditionally, i.e., without checking whether they are syntactically related to the bad-state term.

makaimann · 2020-09-01T15:35:52Z

engines/prover.cpp

+
+    /* Use global set 'coi_visited_terms' here to avoid visiting terms
+       multiple times when we call this function on different terms. */
+    if (coi_visited_terms_.find(cur) == coi_visited_terms_.end()) {


If it would be easier, you can use this method to get all the free symbolic (i.e. uninterpreted) constants: https://github.com/makaimann/smt-switch/blob/8725a64ba57ce1cddb822d9a527a5ace223271b4/include/utils.h#L58

Thanks for the pointer to that function, I was not aware of it!

As far as I see, get_free_symbolic_consts(...) uses a local cache to avoid visiting terms multiple times, whereas coi_visited_terms_ is a global cache that is maintained during the entire COI computation. That would make it a bit tricky to achieve the same functionality using get_free_symbolic_consts(...), I think.

ahmed-irfan · 2020-09-01T23:09:08Z

core/prop.h

  const smt::Term prop() const { return prop_; }

-  const TransitionSystem & transition_system() const { return ts_; }
+  TransitionSystem & transition_system() { return ts_; }


do we need this change?

We need it for the current version but we will refactor the design and revert that change.

core/ts.cpp

ahmed-irfan · 2020-09-01T23:37:02Z

engines/prover.cpp

 ProverResult Prover::prove() { return check_until(INT_MAX); }

+/* For debugging only. */
+void Prover::print_term_dfs(const Term & term)


this could be a static non-member function

I agree. I suggest to leave it for now as it will change during refactoring.

ahmed-irfan · 2020-09-01T23:38:09Z

engines/prover.cpp

+}
+
+/* Add 'term' to 'set' if it does not already appear there. */
+void Prover::collect_coi_term(UnorderedTermSet & set, const Term & term)


this could be a static non-member function

Like above, I agree. I suggest to leave it for now as it will change during refactoring.

engines/prover.cpp

ahmed-irfan · 2020-09-01T23:55:20Z

engines/prover.cpp

+{
+  assert(term != NULL);
+  TermVec open_terms;
+  open_terms.push_back (term);


can a term contain a next state variable?

Yes, term could contain a next-state variable depending on the context in which this function is called.

if term is the bad-state term, then it cannot contain a next-state variable.

if term is a next-state function, then it cannot contain a next-state variable because the transition system is functional (not yet supporting relational ones).

if term is a constraint, then it could contain a next-state variable, which we ignore because there exists always the corresponding constraint over current-state variables, which we will collect. That is related to the comment by @makaimann above.

lonsing · 2020-09-02T01:58:02Z

Thanks for your feedback @makaimann and @ahmed-irfan!

I just pushed an update to address your comments.

Florian Lonsing added 29 commits July 7, 2020 15:55

added test file

7a240c8

added test file, minor comments

0ebbe11

Merge branch 'master' into coi-analysis

b91656e

Merge branch 'master' into coi-analysis

5d7a27e

cleanup

bfc3f6e

added static COI option

ce8bd92

added static COI option

d4a6e4b

added coi printing function (for testing)

a43b2b6

modified example to slow down model checking

b2b11ca

progress

7f9013e

added notes

47e0a2f

small tweaks in visiting terms

e9e6699

added exceptions to safeguard functionality

6ebebde

coi analysis of next-state functions

37b8a86

added recompute-function to TS

e09bd72

removing const qualifiers to allow recomputation of transition function

144b20b

first running version but buggy

cf3e08f

added test: state-var connected to bad_ via constraint

43cd4aa

added note

7e71685

progress but ts-constraints empty in prover but not after parsing

19a38af

fix in copy-constructor of transition system

524aed5

Merge remote-tracking branch 'upstream/master' into coi-analysis

f798940

statistics output

ff3c8d7

fixed merge conflict

6a7e1ec

consider all global constraints

4e8544d

fixed output

4d0a00b

minor things and comments

caceb5c

cleanup

a4b0076

clang format

98285b5

lonsing added the enhancement New feature or request label Sep 1, 2020

lonsing requested review from ahmed-irfan and makaimann September 1, 2020 05:42

makaimann approved these changes Sep 1, 2020

View reviewed changes

ahmed-irfan reviewed Sep 1, 2020

View reviewed changes

core/ts.cpp Show resolved Hide resolved

ahmed-irfan reviewed Sep 1, 2020

View reviewed changes

engines/prover.cpp Outdated Show resolved Hide resolved

remove term sets from parameter list

8569aa2

ahmed-irfan reviewed Sep 1, 2020

View reviewed changes

Florian Lonsing added 2 commits September 1, 2020 17:56

use a reference to state_updates

73eb47e

change state-updates based on coi

a682d07

ahmed-irfan linked an issue Sep 6, 2020 that may be closed by this pull request

Implement Cone of Influence reduction #15

Closed

ahmed-irfan mentioned this pull request Sep 15, 2020

Bug fixes for HWMCC2020 #82

Merged

makaimann changed the base branch from hwmcc2020 to master September 16, 2020 00:31

makaimann merged commit 6d72613 into stanford-centaur:master Sep 16, 2020

Static cone-of-influence analysis #75

Static cone-of-influence analysis #75

Uh oh!

Conversation

lonsing commented Sep 1, 2020

Uh oh!

makaimann left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

lonsing Sep 2, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

ahmed-irfan Sep 1, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

lonsing commented Sep 2, 2020

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

lonsing Sep 2, 2020 •

edited

Loading

ahmed-irfan Sep 1, 2020 •

edited

Loading