Difference between revisions of "Separation Logic"

From CVC4
Jump to: navigation, search
Line 3: Line 3:
 
=Signature and semantics=
 
=Signature and semantics=
  
Given a base theory T, CVC4 supports reasoning about <math>SL(T)_{Loc,Data}</math> formulas, where "Loc" and "Data" are any sort belonging to T.
+
Given a (decidable) base theory T, CVC4 supports a [http://homepage.divms.uiowa.edu/~ajreynol/atva16.pdf decision procedure] for <math>SL(T)_{Loc,Data}</math> formulas, where "Loc" and "Data" are any sort belonging to T.
 
A <math>SL(T)_{Loc,Data}</math> formula is one from the following grammar:
 
A <math>SL(T)_{Loc,Data}</math> formula is one from the following grammar:
 
   F : L | (emp t u) | (pto t u) | (sep F1 ... Fn) | (wand F1 F2) | ~F1 | F1 op ... op Fn
 
   F : L | (emp t u) | (pto t u) | (sep F1 ... Fn) | (wand F1 F2) | ~F1 | F1 op ... op Fn

Revision as of 12:35, 28 November 2016

CVC4 supports a syntax for separation logic as an extension of the *.smt2 language.

Signature and semantics

Given a (decidable) base theory T, CVC4 supports a decision procedure for SL(T)_{Loc,Data} formulas, where "Loc" and "Data" are any sort belonging to T. A SL(T)_{Loc,Data} formula is one from the following grammar:

 F : L | (emp t u) | (pto t u) | (sep F1 ... Fn) | (wand F1 F2) | ~F1 | F1 op ... op Fn

where "op" is any classical Boolean connective, t and u are terms built from symbols in the signature of T of sort Loc and Data respectively, and L is a T-literal. The operator "emp" denotes the empty heap constraint, the operator "pto" denotes the points-to predicate, the operator "sep" denotes separation start and is variadic, and the operator "wand" denote magic wand.

Semantics

A satisfiability relation I,h \models_{SL} \varphi is defined for SL(T)_{Loc,Data} formulas, where I is an interpretation, and h is a heap. The semantics of separation logic operators are as follows:

I,h \models_{SL} L Iff I \models L, if L is a T-literal
I,h \models_{SL} (emp t \ u) Iff h = \emptyset
I,h \models_{SL} (pto t \ u) Iff h = \{(t^I,u^I)\} \text{ and } t^I\not=nil^I
I,h \models_{SL} (sep \phi_1 \ldots \phi_n) Iff there exist heaps h_1,\ldots,h_n s.t. h=h_1\uplus \ldots \uplus h_n and I,h_i \models_{SL} \phi_i, i = 1,\ldots,n
I,h \models_{SL} (wand \phi_1 \ \phi_2) Iff for all heaps h' if h'\#h and I,h' \models_{SL} \phi_1 then I,h'\uplus h \models_{SL} \phi_2

where h_1 \uplus \ldots \uplus h_n denotes the disjoint union of heaps h_1, \ldots, h_n and h'\#h denotes that heaps h' and h are disjoint. Notice that the arguments of "emp" are used to denote the type of the heap and have no meaning otherwise.

Syntax

Separation logic in CVC4 requires the "all supported" logic:

 (set-logic QF_ALL_SUPPORTED)

The syntax for the operators of separation logic is summarized in the following table:

CVC language SMTLIB language C++ API
Empty heap N/A (emp X Y) em.mkExpr(kind::SEP_EMP, X, Y);
Points-to N/A (pto X Y) em.mkExpr(kind::SEP_PTO, X, Y);
Separation star N/A (sep C1 ... Cn) em.mkExpr(kind::SEP_STAR, C1, ..., Cn);
Magic wand N/A (wand C1 C2) em.mkExpr(kind::SEP_WAND, C1, C2);
Nil element N/A (as nil T) em.mkUniqueVar(T,kind::SEP_NIL);

Examples

The following input on heaps Int -> Int is unsatisfiable:

 (set-logic QF_ALL_SUPPORTED)
 (set-info :status unsat)
 (declare-const x Int)
 (declare-const a Int)
 (declare-const b Int)
 (assert (and (pto x a) (pto x b)))
 (assert (not (= a b)))
 (check-sat)

The following input on heaps Int -> Node is satisfiable, where Node denotes a user-defined inductive datatype:

 (set-logic QF_ALL_SUPPORTED)
 (set-info :status sat)
 (declare-const x Int)
 (declare-const y Int)
 (declare-const z Int)
 (declare-datatypes () ((Node (node (data Int) (left Int) (right Int)))))
 (assert (pto x (node 0 y z)))
 (check-sat)

Given a separation logic input, the sorts Loc and Data are inferred by CVC4, and must be consistent across all predicates occurring in an input. CVC4 does not accept an input such as:

 (set-logic QF_ALL_SUPPORTED)
 (declare-sort U 0)
 (declare-const x U)
 (assert (and (pto x 0) (pto 1 2)))

since the sorts of the first arguments of the points-to predicates do not agree.