Difference between revisions of "Datatypes"
(→Syntax/API) |
|||
(2 intermediate revisions by the same user not shown) | |||
Line 106: | Line 106: | ||
| <code> (declare-fun t () '''(Tuple Int Int)''') </code> | | <code> (declare-fun t () '''(Tuple Int Int)''') </code> | ||
| <code>std::vector<Type> types;</code><br> | | <code>std::vector<Type> types;</code><br> | ||
− | <code>types.push_back(em.mkIntegerType());</code>< | + | <code>types.push_back(em.mkIntegerType());</code><br> |
− | <code>types.push_back(em.mkIntegerType());</code>< | + | <code>types.push_back(em.mkIntegerType());</code><br> |
<code>em.'''mkTupleType'''( types );</code> | <code>em.'''mkTupleType'''( types );</code> | ||
|- | |- | ||
Line 133: | Line 133: | ||
| | | | ||
| <code>t: '''[#''' fst:'''INT''', snd:'''INT]''';</code> | | <code>t: '''[#''' fst:'''INT''', snd:'''INT]''';</code> | ||
− | | | + | | n/a |
− | | <code>std::vector< std::pair<std::string, Type> > vec;</code> | + | | <code>std::vector< std::pair<std::string, Type> > vec;</code><br> |
<code>vec.push_back(std::pair<std::string, Type>("fst",em.integerType());</code><br> | <code>vec.push_back(std::pair<std::string, Type>("fst",em.integerType());</code><br> | ||
<code>vec.push_back(std::pair<std::string, Type>("snd",em.integerType());</code><br> | <code>vec.push_back(std::pair<std::string, Type>("snd",em.integerType());</code><br> | ||
Line 156: | Line 156: | ||
<code>const Datatype& dt = tt.getDatatype();</code><br> | <code>const Datatype& dt = tt.getDatatype();</code><br> | ||
<code>Expr s = dt[0][i].getSelector();</code><br> | <code>Expr s = dt[0][i].getSelector();</code><br> | ||
− | <code>em.mkExpr('''kind:: | + | <code>em.mkExpr('''kind::APPLY_SELECTOR''', s, t);</code> |
|-} | |-} |
Latest revision as of 10:51, 8 May 2018
Contents
Logic
To enable CVC4's decision procedure for datatypes, include "DT" in the logic:
(set-logic QF_UFDT)
Alternatively, use the "ALL_SUPPORTED" logic:
(set-logic ALL_SUPPORTED)
Syntax
CVC4 supports the following syntax for declaring mutually recursive blocks of datatypes in *.smt2 input files in the smt lib 2.6 format:
(declare-datatypes ((D1 n1)...(Dk nk)) (((C1 (S1 T1)....(Si Ti))...(Cj .... )) ... ((....) ... (....)))
where D1...Dk are datatype types, C1...Cj are the constructors for datatype D1, S1....Si are the selectors (or "destructors") of constructor C1, and each T1...Ti is a previously declared type or one of D1...k. The symbols U1...Un are type parameters (fresh symbols). The numbers n1...nk denote the number of type parameters for the datatype, where 0 is used for non-parametric datatypes.
In addition to declaring symbols for constructors and selectors, the above command also adds tester (or "discriminator") indexed symbols of the form (_ is C) for each constructor C, which are unary predicates which evaluate to true iff their argument has top-symbol C.
CVC4 also supports datatypes in its native format [1].
Semantics
The decision procedure for inductive datatypes can be found here.
Example Declarations
An enumeration:
(declare-datatypes ((Color 0)) (((Red) (Black))))
A List of Int with "cons" and "nil" as constructors:
(declare-datatypes ((list 0)) (((cons (head Int) (tail list)) (nil))))
A parametric List of T's:
(declare-datatypes ((list 1)) ((par (T) ((cons (head T) (tail (list T))) (nil)))))
Mutual recursion:
(declare-datatypes ((list 0) (tree 0)) (((cons (head tree) (tail list)) (nil)) ((node (data Int) (children list)))))
A (non-recursive) record type:
(declare-datatypes ((record 0)) (((rec (fname String) (lname String) (id Int)))))
Examples
(declare-datatypes ((list 0)) (((cons (head Int) (tail list)) (nil)))) (declare-fun a () list) (declare-fun b () list) (assert (and (= (tail a) b) (not ((_ is nil) b)) (> (head b) 0))) (check-sat)
(declare-datatypes ((record 0)) (((rec (fname String) (lname String) (id Int))))) (declare-const x record) (assert (and (= (fname x) "John") (= (lname x) "Smith"))) (check-sat)
Parametric Datatypes
Instances of parametric datatypes must have their arguments instantiated with concrete types. For instance, in the example:
(declare-datatypes ((list 1)) ((par (T) (cons (head T) (tail (list T))) (nil))))
To declare a list of Int, use the command:
(declare-fun f () (list Int))
Use of constructors that are ambiguously typed must be cast to a concrete type, for instance all occurrences of "nil" for the above datatype must be cast with the syntax:
(as nil (list Int))
Tuples
Tuples are a particular instance of an inductive datatype. CVC4 supports special syntax for tuples as an extension of the SMT-LIB version 2 format. For example:
(declare-fun t () (Tuple Int Int)) (assert (= ((_ tupSel 0) t) 3)) (assert (not (= t (mkTuple 3 4))))
Codatatypes
CVC4 also supports co-inductive datatypes, as described here. The syntax for declaring mutually recursive coinductive datatype blocks is identical to inductive datatypes, except that "declare-datatypes" is replaced by "declare-codatatypes". For example, the following declares the type denote streams of Int:
(declare-codatatypes ((stream 0)) (((cons (head Int) (tail stream)))))
Syntax/API
CVC language | SMTLIB language | C++ API | |
---|---|---|---|
Logic string | Not needed | (set-logic QF_DT)
|
smt.setLogic("QF_DT");
|
Tuple Sort | [<Sort_1>, ..., <Sort_n>] | (Tuple <Sort_1>, ..., <Sort_n>) | CVC4::ExprManager::mkTupleType(std::vector<CVC4::Type>& types) |
t: [INT, INT];
|
(declare-fun t () (Tuple Int Int))
|
std::vector<Type> types;
| |
Tuple constructor | (t1, ..., tn)
|
(mkTuple t1, ..., tn)
|
DatatypeType tt = em.mkTupleType(types);
|
Tuple selector | t.i
|
((_ tupSel i) t)
|
DatatypeType tt = em.mkTupleType(types);
|
Record Sort | [# f1:<Sort_1>, ..., fn:<Sort_n> #] | n/a | CVC4::ExprManager::mkRecordType(const Record& r) |
t: [# fst:INT, snd:INT];
|
n/a | std::vector< std::pair<std::string, Type> > vec;
| |
Record constructor | (# f1 := t1, ..., fn := tn #)
|
n/a | Record r(vec);
|
Record selector | f1(t)
|
n/a | Record r(vec);
|