1 / 25

Proving termination conditions

Proving termination conditions. Mentor : Dr. Ben Livshits & Dr. Stephan Tobies. The Project. The aim: Investigate state of the art approaches for termination proof; Prove termination of sample algorithms. We focused on the following cases: Nested loops ; Recursion ;

Download Presentation

Proving termination conditions

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Proving termination conditions Mentor: Dr. Ben Livshits & Dr. StephanTobies

  2. The Project • The aim: • Investigate state of the art approaches for termination proof; • Prove termination of sample algorithms. • We focused on the following cases: • Nested loops; • Recursion; • Linked list data structures.

  3. Termination • A program is terminating if all its executions of all its executions are finite. • A program is non-terminating it there exits at least one infinite execution.

  4. Motivation Functional correctness + termination proof = Total correctness • The halting problem: termination is undecidable (Alan Turing). • Does not mean we can’t not prove termination in every case. We can prove termination via introducing termination metrics (ranking functions).

  5. Termination proof in Dafny • Dafny is a programming language and verifier that enables to prove terminations of algorithms. • Dafny provides annotations to specify termination metrics • Many verifiers do not support termination proofs • http://research.microsoft.com/en-us/projects/dafny/

  6. DafnyApproach-Formalism • Let U be a non-empty set of disjoint union of algebraic datatypes, tuples and variables. • Algebraicdatatypes: sets, sequences, lists. • Let S be set of states of program. • Let (Y,>=) be a well-ordered set. • Forinstance, ℕ is a well-ordered set.

  7. FormalismCont. • Define metric φ:UxSYsuchthat • For∀ transition (s,s’), φ(u,s)> φ(u,s’) • For∀ state s, ∀ u in U, φ(u,s)>=0 • ∃δ>0 suchthatfor∀ transition (s,s’) and ∀ u in U,φ(u,s)> φ(u,s’)+δ • Usuallythismapping is calledprogressmeasureorrankingfunction

  8. Node #1 ReachableNodes: {1, 2, 3, 4, 5} Node #2 ReachableNodes: {2, 4, 5} Node #3 ReachableNodes: {3} Node #4 ReachableNodes: {4} Node #5 ReachableNodes: {5} method Find(x: int): returns (found:bool) decreases ReachableNodes; { if (x == data) {found := true;} else if (left != null && x < data) {found := left.Find(x);} else if (right != null && x > data) {found := right.Find(x);} else {found := false;} }

  9. OtherApproaches • Compose termination arguments since constructing a ranking function can be difficult • Decompose the termination check into easier ones • Terminator http://research.microsoft.com/en-us/um/cambridge/projects/terminator/development of automatic methods for proving program termination and general liveness properties

  10. Our Contribution • Provedtermination of several Dafny programs: • QuickSort • MergeSort • Insertion sort • Insertion and search for binary search tree • We have investigated state of art for termination proof. • Idea: reusing termination metric to prove computational complexity of algorithms.

  11. What we have learned What remains to be done • More in-depth understanding of formal verification. • Got information about proving termination properties. • Got additional insights how complexity analysis can be implemented by reusing termination metric.

  12. Thank you!Questions?

  13. Possible tools • partial correctness + termination = total correctness • partial correctness + time > total correctness VCC provides only partial correctness Other tools: • Code contracts • Dafny • f* • spec#

  14. Why dafny It is imperative, sequential, supports generic classes and dynamic allocation, and builds in specification constructs. Also • Updatable ghost variables • Recursive functions • Proof of total correctness • Types like algebraic datatypes (sets, sequences)

  15. Principles of termination by Byron Cook • A program is terminating if all its executions are finite. A program is non-terminating it there exits at least one infinite execution. • When trying to prove termination, one tries to prove program’s transition relation is well-founded. • Turing’s suggestion for proving well-foundedness: Find a map form (R,S) to a well-ordered set and then prove this map is a homomorphism. These maps are typically ranking functions. • Constructing a ranking function can be difficult: Compose termination arguments. • Another idea: Decompose the termination check into easier ones.

  16. Notation {P}C{Q} : partial correctness [P]C[Q] : total correctness Problems with: • While loop ( for is alias in C-like languages) • Recursion • Function invocation

  17. Total Correctness Specification A total correctness specification [P]C[Q] true if • Whenever C is executed in a state satisfying P, then the execution of C terminates • After C terminates Q holds With the exception of the WHILE, FOR loops, recursion and function invocation, all the axioms and rules described so far are sound for total correctness as well as partial correctness

  18. Rules for Non-Looping Command • Replace { and } by [ and ], respectively, in: - Assignment axiom - Consequence rules - Conditional rules - Sequencing rule - Block rule

  19. Total Correctness Assignment Axiom - Assignment axiom for total correctness: Ⱶ [P [E/V ]] V := E [P ] • Note that the assignment axiom for total correctness states that assignment commands always terminate • So all function applications in expressions must terminate • This might not be the case if functions could be defined recursively EXAMPLE: X := fact(-1), where fact(n) = if n = 0 then 1 else n * fact(n - 1)

  20. WHILE Rule for Total Correctnes • WHILE commands are the only commands that can cause non-termination • The idea behind the WHILE rule for total correctness is: 1) to prove WHILE S DO C terminates 2) show that some non-negative quantity decreases on each iteration of C 3) this decreasing quantity is called a variant

  21. WHILE Rule for Total Correctness

  22. Derived rules Multiple step rules for total correctness can be derived in the same way as for partial correctness • The rules are the same up to the brackets used • Same derivations with total correctness rules replacing partial correctness ones

  23. Example 3: ghost data class ListNode { var data: int; var next: ListNode; ghost varreachableNodes: set<ListNode>; // ... function sum(): int reads *; decreases reachableNodes; { if next == null then data else data + next.Sum() } }

  24. Example 1: no termination proof is required method processArray(numbers: array<int>) requires numbers != null; { vari := 0; var sum := 0; while (i < numbers.Length) { sum := sum + numbers[i]; i := i + 1; } }

  25. Dafny programs that we have proved Termination of: • QuickSort • MergeSort • Insertion and search for binary search tree • TODO: outline the most interesting aspects of these programs

More Related