summaryrefslogtreecommitdiff
path: root/src/parser/smt2/Smt2.g
diff options
context:
space:
mode:
Diffstat (limited to 'src/parser/smt2/Smt2.g')
-rw-r--r--src/parser/smt2/Smt2.g112
1 files changed, 101 insertions, 11 deletions
diff --git a/src/parser/smt2/Smt2.g b/src/parser/smt2/Smt2.g
index 07ace295c..4d884d894 100644
--- a/src/parser/smt2/Smt2.g
+++ b/src/parser/smt2/Smt2.g
@@ -1107,6 +1107,10 @@ metaInfoInternal[CVC4::PtrCloser<CVC4::Command>* cmd]
sexpr.getRationalValue() == Rational(5, 2)) ||
sexpr.getValue() == "2.5" ) {
PARSER_STATE->setLanguage(language::input::LANG_SMTLIB_V2_5);
+ } else if( (sexpr.isRational() &&
+ sexpr.getRationalValue() == Rational(13, 5)) ||
+ sexpr.getValue() == "2.6" ) {
+ PARSER_STATE->setLanguage(language::input::LANG_SMTLIB_V2_6);
}
}
PARSER_STATE->setInfo(name.c_str() + 1, sexpr);
@@ -1353,7 +1357,9 @@ extendedCommand[CVC4::PtrCloser<CVC4::Command>* cmd]
}
/* Extended SMT-LIB set of commands syntax, not permitted in
* --smtlib2 compliance mode. */
- : DECLARE_DATATYPES_TOK datatypesDefCommand[false, cmd]
+ : DECLARE_DATATYPES_2_5_TOK datatypes_2_5_DefCommand[false, cmd]
+ | DECLARE_CODATATYPES_2_5_TOK datatypes_2_5_DefCommand[true, cmd]
+ | DECLARE_DATATYPES_TOK datatypesDefCommand[false, cmd]
| DECLARE_CODATATYPES_TOK datatypesDefCommand[true, cmd]
| rewriterulesCommand[cmd]
@@ -1500,7 +1506,7 @@ extendedCommand[CVC4::PtrCloser<CVC4::Command>* cmd]
;
-datatypesDefCommand[bool isCo, CVC4::PtrCloser<CVC4::Command>* cmd]
+datatypes_2_5_DefCommand[bool isCo, CVC4::PtrCloser<CVC4::Command>* cmd]
@declarations {
std::vector<CVC4::Datatype> dts;
std::string name;
@@ -1516,8 +1522,71 @@ datatypesDefCommand[bool isCo, CVC4::PtrCloser<CVC4::Command>* cmd]
RPAREN_TOK
LPAREN_TOK ( LPAREN_TOK datatypeDef[isCo, dts, sorts] RPAREN_TOK )+ RPAREN_TOK
{ PARSER_STATE->popScope();
- cmd->reset(new DatatypeDeclarationCommand(
- PARSER_STATE->mkMutualDatatypeTypes(dts)));
+ cmd->reset(new DatatypeDeclarationCommand(PARSER_STATE->mkMutualDatatypeTypes(dts)));
+ }
+ ;
+
+datatypesDefCommand[bool isCo, CVC4::PtrCloser<CVC4::Command>* cmd]
+@declarations {
+ std::vector<CVC4::Datatype> dts;
+ std::vector<Type> types;
+ std::string name;
+ std::vector<std::string> dnames;
+ std::vector<unsigned> arities;
+ std::vector<Type> params;
+}
+ : { PARSER_STATE->checkThatLogicIsSet(); PARSER_STATE->pushScope(true); }
+ LPAREN_TOK /* sorts */
+ ( LPAREN_TOK symbol[name,CHECK_UNDECLARED,SYM_SORT] n=INTEGER_LITERAL RPAREN_TOK
+ { unsigned arity = AntlrInput::tokenToUnsigned(n);
+ //Type type;
+ //if(arity == 0) {
+ // type = PARSER_STATE->mkSort(name);
+ //} else {
+ // type = PARSER_STATE->mkSortConstructor(name, arity);
+ //}
+ Debug("parser-dt") << "Datatype : " << name << ", arity = " << arity << std::endl;
+ //types.push_back(type);
+ dnames.push_back(name);
+ arities.push_back( arity );
+ }
+ )*
+ RPAREN_TOK
+ LPAREN_TOK
+ ( LPAREN_TOK {
+ params.clear();
+ Debug("parser-dt") << "Processing datatype #" << dts.size() << std::endl;
+ if( dts.size()>=dnames.size() ){
+ PARSER_STATE->parseError("Too many datatypes defined in this block.");
+ }
+ }
+ ( PAR_TOK { PARSER_STATE->pushScope(true); } LPAREN_TOK
+ ( symbol[name,CHECK_UNDECLARED,SYM_SORT]
+ { params.push_back( PARSER_STATE->mkSort(name) ); }
+ )*
+ RPAREN_TOK {
+ if( params.size()!=arities[dts.size()] ){
+ PARSER_STATE->parseError("Wrong number of parameters for datatype.");
+ }
+ Debug("parser-dt") << params.size() << " parameters for " << dnames[dts.size()] << std::endl;
+ dts.push_back(Datatype(dnames[dts.size()],params,isCo));
+ }
+ LPAREN_TOK
+ ( LPAREN_TOK constructorDef[dts.back()] RPAREN_TOK )+
+ RPAREN_TOK { PARSER_STATE->popScope(); }
+ | { if( 0!=arities[dts.size()] ){
+ PARSER_STATE->parseError("No parameters given for datatype.");
+ }
+ Debug("parser-dt") << params.size() << " parameters for " << dnames[dts.size()] << std::endl;
+ dts.push_back(Datatype(dnames[dts.size()],params,isCo));
+ }
+ ( LPAREN_TOK constructorDef[dts.back()] RPAREN_TOK )+
+ )
+ RPAREN_TOK
+ )+
+ RPAREN_TOK
+ { PARSER_STATE->popScope();
+ cmd->reset(new DatatypeDeclarationCommand(PARSER_STATE->mkMutualDatatypeTypes(dts)));
}
;
@@ -1890,8 +1959,13 @@ term[CVC4::Expr& expr, CVC4::Expr& expr2]
| LPAREN_TOK
( /* An indexed function application */
- indexedFunctionName[op] termList[args,expr] RPAREN_TOK
- { expr = MK_EXPR(op, args);
+ indexedFunctionName[op, kind] termList[args,expr] RPAREN_TOK
+ {
+ if( kind!=kind::NULL_EXPR ){
+ expr = MK_EXPR( kind, op, args );
+ }else{
+ expr = MK_EXPR(op, args);
+ }
PARSER_STATE->checkOperator(expr.getKind(), args.size());
}
| /* Array constant (in Z3 syntax) */
@@ -2267,7 +2341,11 @@ attribute[CVC4::Expr& expr, CVC4::Expr& retExpr, std::string& attr]
/**
* Matches a bit-vector operator (the ones parametrized by numbers)
*/
-indexedFunctionName[CVC4::Expr& op]
+indexedFunctionName[CVC4::Expr& op, CVC4::Kind& kind]
+@init {
+ Expr expr;
+ Expr expr2;
+}
: LPAREN_TOK INDEX_TOK
( 'extract' n1=INTEGER_LITERAL n2=INTEGER_LITERAL
{ op = MK_CONST(BitVectorExtract(AntlrInput::tokenToUnsigned($n1),
@@ -2344,6 +2422,13 @@ indexedFunctionName[CVC4::Expr& op]
{ op = MK_CONST(FloatingPointToUBV(AntlrInput::tokenToUnsigned($m))); }
| FP_TO_SBV_TOK m=INTEGER_LITERAL
{ op = MK_CONST(FloatingPointToSBV(AntlrInput::tokenToUnsigned($m))); }
+ | TESTER_TOK term[expr, expr2] {
+ if( !expr.getType().isConstructor() ){
+ PARSER_STATE->parseError("Bad syntax for test (_ is X), X must be a constructor.");
+ }
+ op = Datatype::datatypeOf(expr)[Datatype::indexOf(expr)].getTester();
+ kind = CVC4::kind::APPLY_TESTER;
+ }
| badIndexedFunctionName
)
RPAREN_TOK
@@ -2795,7 +2880,7 @@ GET_ASSERTIONS_TOK : 'get-assertions';
GET_PROOF_TOK : 'get-proof';
GET_UNSAT_CORE_TOK : 'get-unsat-core';
EXIT_TOK : 'exit';
-RESET_TOK : { PARSER_STATE->v2_5() }? 'reset';
+RESET_TOK : { PARSER_STATE->v2_5(false) }? 'reset';
RESET_ASSERTIONS_TOK : 'reset-assertions';
ITE_TOK : 'ite';
LET_TOK : 'let';
@@ -2815,8 +2900,13 @@ AS_TOK : 'as';
CONST_TOK : 'const';
// extended commands
-DECLARE_DATATYPES_TOK : 'declare-datatypes';
-DECLARE_CODATATYPES_TOK : 'declare-codatatypes';
+DECLARE_DATATYPE_TOK : { PARSER_STATE->v2_6() }? 'declare-datatype';
+DECLARE_DATATYPES_2_5_TOK : { !PARSER_STATE->v2_6() }?'declare-datatypes';
+DECLARE_DATATYPES_TOK : { PARSER_STATE->v2_6() }?'declare-datatypes';
+DECLARE_CODATATYPES_2_5_TOK : { !PARSER_STATE->v2_6() }?'declare-codatatypes';
+DECLARE_CODATATYPES_TOK : { PARSER_STATE->v2_6() }?'declare-codatatypes';
+PAR_TOK : { PARSER_STATE->v2_6() }?'par';
+TESTER_TOK : { PARSER_STATE->v2_6() }?'is';
GET_MODEL_TOK : 'get-model';
ECHO_TOK : 'echo';
REWRITE_RULE_TOK : 'assert-rewrite';
@@ -3039,7 +3129,7 @@ STRING_LITERAL_2_0
* will be part of the token text. Use the str[] parser rule instead.
*/
STRING_LITERAL_2_5
- : { PARSER_STATE->v2_5() || PARSER_STATE->sygus() }?=>
+ : { PARSER_STATE->v2_5(false) || PARSER_STATE->sygus() }?=>
'"' (~('"') | '""')* '"'
;
generated by cgit on debian on lair
contact matthew@masot.net with questions or feedback