void goto_checkt::mod_by_zero_check(
const mod_exprt &expr,
const guardt &guard)
{
if(!enable_div_by_zero_check)
return;
// add divison by zero subgoal
exprt zero=gen_zero(expr.op1().type());
if(zero.is_nil())
throw "no zero of argument type of operator "+expr.id_string();
exprt inequality(ID_notequal, bool_typet());
inequality.copy_to_operands(expr.op1(), zero);
add_guarded_claim(
inequality,
"division by zero",
"division-by-zero",
expr.find_source_location(),
expr,
guard);
}
void c_typecastt::implicit_typecast_arithmetic(
exprt &expr,
c_typet c_type)
{
typet new_type;
const typet &expr_type=ns.follow(expr.type());
switch(c_type)
{
case PTR:
if(expr_type.id()==ID_array)
{
new_type.id(ID_pointer);
new_type.subtype()=expr_type.subtype();
break;
}
return;
case BOOL: new_type=bool_typet(); break;
case CHAR: assert(false); // should always be promoted to int
case UCHAR: assert(false); // should always be promoted to int
case SHORT: assert(false); // should always be promoted to int
case USHORT: assert(false); // should always be promoted to int
case INT: new_type=int_type(); break;
case UINT: new_type=uint_type(); break;
case LONG: new_type=long_int_type(); break;
case ULONG: new_type=long_uint_type(); break;
case LONGLONG: new_type=long_long_int_type(); break;
case ULONGLONG: new_type=long_long_uint_type(); break;
case SINGLE: new_type=float_type(); break;
case DOUBLE: new_type=double_type(); break;
case LONGDOUBLE: new_type=long_double_type(); break;
case RATIONAL: new_type=rational_typet(); break;
case REAL: new_type=real_typet(); break;
case INTEGER: new_type=integer_typet(); break;
case COMPLEX: return; // do nothing
default: return;
}
if(new_type!=expr_type)
{
if(new_type.id()==ID_pointer &&
expr_type.id()==ID_array)
{
exprt index_expr(ID_index, expr_type.subtype());
index_expr.reserve_operands(2);
index_expr.move_to_operands(expr);
index_expr.copy_to_operands(gen_zero(index_type()));
expr=exprt(ID_address_of, new_type);
expr.move_to_operands(index_expr);
}
else
do_typecast(expr, new_type);
}
}
void c_typecastt::do_typecast(exprt &expr, const typet &type)
{
// special case: array -> pointer is actually
// something like address_of
const typet &expr_type=ns.follow(expr.type());
if(expr_type.id()==ID_array)
{
index_exprt index;
index.array()=expr;
index.index()=gen_zero(index_type());
index.type()=expr_type.subtype();
expr=address_of_exprt(index);
if(ns.follow(expr.type())!=ns.follow(type))
expr.make_typecast(type);
return;
}
if(expr_type!=type)
{
// C booleans are special: we compile to ?0:1
if(type.get(ID_C_c_type)==ID_bool)
{
if(expr_type.id()==ID_bool) // bool -> _Bool
{
exprt result=if_exprt(expr, gen_one(type), gen_zero(type));
expr.swap(result);
}
else // * -> _Bool
{
equal_exprt equal_zero(expr, gen_zero(expr_type));
exprt result=if_exprt(equal_zero, gen_zero(type), gen_one(type));
expr.swap(result);
}
}
else
{
expr.make_typecast(type);
}
}
}
void goto_symext::process_array_expr_rec(
exprt &expr,
const typet &type) const
{
if(expr.id()==ID_if)
{
if_exprt &if_expr=to_if_expr(expr);
process_array_expr_rec(if_expr.true_case(), type);
process_array_expr_rec(if_expr.false_case(), type);
}
else if(expr.id()==ID_index)
{
// strip index
index_exprt &index_expr=to_index_expr(expr);
exprt tmp=index_expr.array();
expr.swap(tmp);
}
else if(expr.id()==ID_typecast)
{
// strip
exprt tmp=to_typecast_expr(expr).op0();
expr.swap(tmp);
process_array_expr_rec(expr, type);
}
else if(expr.id()==ID_address_of)
{
// strip
exprt tmp=to_address_of_expr(expr).op0();
expr.swap(tmp);
process_array_expr_rec(expr, type);
}
else if(expr.id()==ID_symbol &&
expr.get_bool(ID_C_SSA_symbol) &&
to_ssa_expr(expr).get_original_expr().id()==ID_index)
{
const ssa_exprt &ssa=to_ssa_expr(expr);
const index_exprt &index_expr=to_index_expr(ssa.get_original_expr());
exprt tmp=index_expr.array();
expr.swap(tmp);
}
else
Forall_operands(it, expr)
process_array_expr_rec(*it, it->type());
if(!base_type_eq(expr.type(), type, ns))
{
byte_extract_exprt be(byte_extract_id());
be.type()=type;
be.op()=expr;
be.offset()=gen_zero(index_type());
expr.swap(be);
}
}
exprt dereferencet::dereference_rec(
const exprt &address,
const exprt &offset,
const typet &type)
{
if(address.id()==ID_address_of)
{
const address_of_exprt &address_of_expr=to_address_of_expr(address);
const exprt &object=address_of_expr.object();
return read_object(object, offset, type);
}
else if(address.id()==ID_typecast)
{
const typecast_exprt &typecast_expr=to_typecast_expr(address);
return dereference_typecast(typecast_expr, offset, type);
}
else if(address.id()==ID_plus)
{
// pointer arithmetic
if(address.operands().size()<2)
throw "plus with less than two operands";
return dereference_plus(address, offset, type);
}
else if(address.id()==ID_if)
{
const if_exprt &if_expr=to_if_expr(address);
return dereference_if(if_expr, offset, type);
}
else if(address.id()==ID_constant)
{
const typet result_type=ns.follow(address.type()).subtype();
// pointer-typed constant
if(to_constant_expr(address).get_value()==ID_NULL) // NULL
{
// we turn this into (type *)0
exprt zero=gen_zero(index_type());
return dereference_rec(
typecast_exprt(zero, address.type()), offset, type);
}
else
throw "dereferencet: unexpected pointer constant "+address.pretty();
}
else
{
throw "failed to dereference `"+address.id_string()+"'";
}
}
void goto_checkt::undefined_shift_check(
const shift_exprt &expr,
const guardt &guard)
{
if(!enable_undefined_shift_check)
return;
// Undefined for all types and shifts if distance exceeds width,
// and also undefined for negative distances.
const typet &distance_type=
ns.follow(expr.distance().type());
if(distance_type.id()==ID_signedbv)
{
binary_relation_exprt inequality(
expr.distance(), ID_ge, gen_zero(distance_type));
add_guarded_claim(
inequality,
"shift distance is negative",
"undefined-shift",
expr.find_source_location(),
expr,
guard);
}
const typet &op_type=
ns.follow(expr.op().type());
if(op_type.id()==ID_unsignedbv || op_type.id()==ID_signedbv)
{
exprt width_expr=
from_integer(to_bitvector_type(op_type).get_width(), distance_type);
if(width_expr.is_nil())
throw "no number for width for operator "+expr.id_string();
binary_relation_exprt inequality(
expr.distance(), ID_lt, width_expr);
add_guarded_claim(
inequality,
"shift distance too large",
"undefined-shift",
expr.find_source_location(),
expr,
guard);
}
}
exprt gen_one(const typet &type)
{
const irep_idt type_id=type.id();
exprt result=constant_exprt(type);
if(type_id==ID_bool ||
type_id==ID_rational ||
type_id==ID_real ||
type_id==ID_integer ||
type_id==ID_natural)
{
result.set(ID_value, ID_1);
}
else if(type_id==ID_unsignedbv ||
type_id==ID_signedbv ||
type_id==ID_c_enum)
{
std::string value;
unsigned width=to_bitvector_type(type).get_width();
for(unsigned i=0; i<width-1; i++)
value+='0';
value+='1';
result.set(ID_value, value);
}
else if(type_id==ID_fixedbv)
{
fixedbvt fixedbv;
fixedbv.spec=to_fixedbv_type(type);
fixedbv.from_integer(1);
result=fixedbv.to_expr();
}
else if(type_id==ID_floatbv)
{
ieee_floatt ieee_float;
ieee_float.spec=to_floatbv_type(type);
ieee_float.from_integer(1);
result=ieee_float.to_expr();
}
else if(type_id==ID_complex)
{
result=exprt(ID_complex, type);
result.operands().resize(2);
result.op0()=gen_one(type.subtype());
result.op1()=gen_zero(type.subtype());
}
else
result.make_nil();
return result;
}
exprt gen_zero(const typet &type)
{
exprt result;
const irep_idt type_id=type.id();
result=constant_exprt(type);
if(type_id==ID_rational ||
type_id==ID_real ||
type_id==ID_integer ||
type_id==ID_natural ||
type_id==ID_complex ||
type_id==ID_c_enum)
{
result.set(ID_value, ID_0);
}
else if(type_id==ID_unsignedbv ||
type_id==ID_signedbv ||
type_id==ID_verilogbv ||
type_id==ID_floatbv ||
type_id==ID_fixedbv)
{
std::string value;
unsigned width=to_bitvector_type(type).get_width();
for(unsigned i=0; i<width; i++)
value+='0';
result.set(ID_value, value);
}
else if(type_id==ID_complex)
{
result=exprt(ID_complex, type);
exprt sub_zero=gen_zero(type.subtype());
result.operands().resize(2, sub_zero);
}
else if(type_id==ID_bool)
{
result.make_false();
}
else if(type_id==ID_pointer)
{
result.set(ID_value, ID_NULL);
}
else
result.make_nil();
return result;
}
void java_bytecode_convertt::convert(
symbolt &class_symbol,
const fieldt &f)
{
class_typet &class_type=to_class_type(class_symbol.type);
typet member_type=java_type_from_string(f.signature);
class_type.components().push_back(class_typet::componentt());
class_typet::componentt &component=class_type.components().back();
component.set_name(f.name);
component.set_base_name(f.name);
component.set_pretty_name(f.name);
component.type()=member_type;
if(f.is_private)
component.set_access(ID_private);
else if(f.is_protected)
component.set_access(ID_protected);
else if(f.is_public)
component.set_access(ID_public);
// is this a static field?
if(f.is_static)
{
// create the symbol
symbolt new_symbol;
new_symbol.is_static_lifetime=true;
new_symbol.is_lvalue=true;
new_symbol.is_state_var=true;
new_symbol.name=id2string(class_symbol.name)+"."+id2string(f.name);
new_symbol.base_name=f.name;
new_symbol.type=member_type;
new_symbol.pretty_name=id2string(class_symbol.pretty_name)+"."+id2string(f.name);
new_symbol.mode=ID_java;
new_symbol.is_type=false;
new_symbol.value=gen_zero(member_type);
if(symbol_table.add(new_symbol))
{
error() << "failed to add static field symbol" << eom;
throw 0;
}
}
}
exprt dereferencet::operator()(const exprt &pointer)
{
if(pointer.type().id()!=ID_pointer)
throw "dereference expected pointer type, but got "+
pointer.type().pretty();
// type of the object
const typet &type=pointer.type().subtype();
#ifdef DEBUG
std::cout << "DEREF: " << from_expr(ns, "", pointer) << std::endl;
#endif
return dereference_rec(
pointer,
gen_zero(index_type()), // offset
type);
}
void c_typecastt::do_typecast(exprt &expr, const typet &dest_type)
{
// special case: array -> pointer is actually
// something like address_of
const typet &src_type=ns.follow(expr.type());
if(src_type.id()==ID_array)
{
index_exprt index;
index.array()=expr;
index.index()=gen_zero(index_type());
index.type()=src_type.subtype();
expr=address_of_exprt(index);
if(ns.follow(expr.type())!=ns.follow(dest_type))
expr.make_typecast(dest_type);
return;
}
if(src_type!=dest_type)
{
// C booleans are special; we produce the
// explicit comparision with zero.
// Note that this requires ieee_float_notequal
// in case of floating-point numbers.
if(dest_type.get(ID_C_c_type)==ID_bool)
{
expr=is_not_zero(expr, ns);
expr.make_typecast(dest_type);
}
else if(dest_type.id()==ID_bool)
{
expr=is_not_zero(expr, ns);
}
else
{
expr.make_typecast(dest_type);
}
}
}
void goto_checkt::div_by_zero_check(const exprt &expr, const guardt &guard)
{
if (options.get_bool_option("no-div-by-zero-check"))
return;
if (expr.operands().size() != 2)
throw expr.id_string() + " takes two arguments";
// add divison by zero subgoal
exprt zero = gen_zero(expr.op1().type());
if (zero.is_nil())
throw "no zero of argument type of operator " + expr.id_string();
exprt inequality("notequal", bool_typet());
inequality.copy_to_operands(expr.op1(), zero);
add_guarded_claim(inequality, "division by zero", "division-by-zero",
expr.find_location(), guard);
}
bool simplify_exprt::simplify_address_of(exprt &expr)
{
if(expr.operands().size()!=1) return true;
if(ns.follow(expr.type()).id()!=ID_pointer) return true;
exprt &object=expr.op0();
bool result=simplify_address_of_arg(object);
if(object.id()==ID_index)
{
index_exprt &index_expr=to_index_expr(object);
if(!index_expr.index().is_zero())
{
// we normalize &a[i] to (&a[0])+i
exprt offset;
offset.swap(index_expr.op1());
index_expr.op1()=gen_zero(offset.type());
exprt addition(ID_plus, expr.type());
addition.move_to_operands(expr, offset);
expr.swap(addition);
return false;
}
}
else if(object.id()==ID_dereference)
{
// simplify &*p to p
assert(object.operands().size()==1);
exprt tmp=object.op0();
expr=tmp;
return false;
}
return result;
}
exprt convert_float_literal(const std::string &src)
{
mp_integer significand;
mp_integer exponent;
bool is_float, is_long, is_imaginary;
bool is_decimal, is_float80, is_float128; // GCC extensions
unsigned base;
parse_float(src, significand, exponent, base,
is_float, is_long, is_imaginary,
is_decimal, is_float80, is_float128);
exprt result=exprt(ID_constant);
result.set(ID_C_cformat, src);
// In ANSI-C, float literals are double by default,
// unless marked with 'f'.
// All of these can be complex as well.
// This can be overriden with
// config.ansi_c.single_precision_constant.
if(is_float)
result.type()=float_type();
else if(is_long)
result.type()=long_double_type();
else if(is_float80)
{
result.type()=ieee_float_spect(64, 15).to_type();
result.type().set(ID_C_c_type, ID_long_double);
}
else if(is_float128)
{
result.type()=ieee_float_spect::quadruple_precision().to_type();
result.type().set(ID_C_c_type, ID_gcc_float128);
}
else
{
// default
if(config.ansi_c.single_precision_constant)
result.type()=float_type(); // default
else
result.type()=double_type(); // default
}
if(is_decimal)
{
// TODO - should set ID_gcc_decimal32/ID_gcc_decimal64/ID_gcc_decimal128,
// but these aren't handled anywhere
}
if(config.ansi_c.use_fixed_for_float)
{
unsigned width=result.type().get_int(ID_width);
unsigned fraction_bits;
const irep_idt integer_bits=result.type().get(ID_integer_bits);
assert(width!=0);
if(integer_bits==irep_idt())
fraction_bits=width/2; // default
else
fraction_bits=width-safe_string2int(id2string(integer_bits));
mp_integer factor=mp_integer(1)<<fraction_bits;
mp_integer value=significand*factor;
if(value!=0)
{
if(exponent<0)
value/=power(base, -exponent);
else
{
value*=power(base, exponent);
if(value>=power(2, width-1))
{
// saturate: use "biggest value"
value=power(2, width-1)-1;
}
else if(value<=-power(2, width-1)-1)
{
// saturate: use "smallest value"
value=-power(2, width-1);
}
}
}
result.set(ID_value, integer2binary(value, width));
}
else
{
ieee_floatt a;
a.spec=to_floatbv_type(result.type());
if(base==10)
a.from_base10(significand, exponent);
else if(base==2) // hex
a.build(significand, exponent);
else
assert(false);
result.set(ID_value,
integer2binary(a.pack(), a.spec.width()));
}
if(is_imaginary)
{
complex_typet complex_type;
complex_type.subtype()=result.type();
exprt complex_expr(ID_complex, complex_type);
complex_expr.operands().resize(2);
complex_expr.op0()=gen_zero(result.type());
complex_expr.op1()=result;
return complex_expr;
}
return result;
}
bool c_typecheck_baset::zero_initializer(exprt &value, const typet &type) const
{
const std::string &type_id = type.id_string();
if(type_id == "bool")
{
value.make_false();
return false;
}
if(
type_id == "unsignedbv" || type_id == "signedbv" || type_id == "floatbv" ||
type_id == "fixedbv" || type_id == "pointer")
{
value = gen_zero(type);
return false;
}
else if(type_id == "code")
return false;
else if(type_id == "c_enum" || type_id == "incomplete_c_enum")
{
value = exprt("constant", type);
value.value(i2string(0));
return false;
}
else if(type_id == "array")
{
const array_typet &array_type = to_array_type(type);
exprt tmpval;
if(zero_initializer(tmpval, array_type.subtype()))
return true;
const exprt &size_expr = array_type.size();
if(size_expr.id() == "infinity")
{
}
else
{
mp_integer size;
if(to_integer(size_expr, size))
return true;
// Permit GCC zero sized arrays; disallow negative sized arrays.
// Cringe slightly when doing it though.
if(size < 0)
return true;
}
value = exprt("array_of", type);
value.move_to_operands(tmpval);
return false;
}
else if(type_id == "struct")
{
const irept::subt &components = type.components().get_sub();
value = exprt("struct", type);
forall_irep(it, components)
{
exprt tmp;
if(zero_initializer(tmp, (const typet &)it->type()))
return true;
value.move_to_operands(tmp);
}
return false;
}
exprt convert_integer_literal(const std::string &src)
{
bool is_unsigned=false, is_imaginary=false;
unsigned long_cnt=0;
unsigned width_suffix=0;
unsigned base=10;
for(unsigned i=0; i<src.size(); i++)
{
register char ch=src[i];
if(ch=='u' || ch=='U')
is_unsigned=true;
else if(ch=='l' || ch=='L')
long_cnt++;
else if(ch=='i' || ch=='I')
{
// This can be "1i128" in MS mode,
// and "10i" (imaginary) for GCC.
// If it's followed by a number, we do MS mode.
if((i+1)<src.size() && isdigit(src[i+1]))
width_suffix=atoi(src.c_str()+i+1);
else
is_imaginary=true;
}
else if(ch=='j' || ch=='J')
is_imaginary=true;
}
mp_integer value;
if(src.size()>=2 && src[0]=='0' && tolower(src[1])=='x')
{
// hex; strip "0x"
base=16;
std::string without_prefix(src, 2, std::string::npos);
value=string2integer(without_prefix, 16);
}
else if(src.size()>=2 && src[0]=='0' && tolower(src[1])=='b')
{
// binary; strip "0x"
// see http://gcc.gnu.org/onlinedocs/gcc/Binary-constants.html
base=2;
std::string without_prefix(src, 2, std::string::npos);
value=string2integer(without_prefix, 2);
}
else if(src.size()>=2 && src[0]=='0')
{
// octal
base=8;
value=string2integer(src, 8);
}
else
{
// The default is base 10.
value=string2integer(src, 10);
}
if(width_suffix!=0)
{
// this is a Microsoft extension
irep_idt c_type;
if(width_suffix<=config.ansi_c.int_width)
c_type=is_unsigned?ID_unsigned_int:ID_signed_int;
else if(width_suffix<=config.ansi_c.long_int_width)
c_type=is_unsigned?ID_unsigned_long_int:ID_signed_long_int;
else
c_type=is_unsigned?ID_unsigned_long_long_int:ID_signed_long_long_int;
typet type=typet(is_unsigned?ID_unsignedbv:ID_signedbv);
type.set(ID_width, width_suffix);
type.set(ID_C_c_type, c_type);
exprt result=from_integer(value, type);
result.set(ID_C_cformat, src);
return result;
}
mp_integer value_abs=value;
if(value<0)
value_abs.negate();
bool is_hex_or_oct_or_bin=(base==8) || (base==16) || (base==2);
#define FITS(width, signed) \
((signed?!is_unsigned:(is_unsigned || is_hex_or_oct_or_bin)) && \
(power(2, signed?width-1:width)>value_abs))
unsigned width;
bool is_signed=false;
irep_idt c_type;
if(FITS(config.ansi_c.int_width, true) && long_cnt==0) // int
{
width=config.ansi_c.int_width;
is_signed=true;
c_type=ID_signed_int;
}
else if(FITS(config.ansi_c.int_width, false) && long_cnt==0) // unsigned int
{
width=config.ansi_c.int_width;
is_signed=false;
c_type=ID_unsigned_int;
}
else if(FITS(config.ansi_c.long_int_width, true) && long_cnt!=2) // long int
{
width=config.ansi_c.long_int_width;
is_signed=true;
c_type=ID_signed_long_int;
}
else if(FITS(config.ansi_c.long_int_width, false) && long_cnt!=2) // unsigned long int
{
width=config.ansi_c.long_int_width;
is_signed=false;
c_type=ID_unsigned_long_int;
}
else if(FITS(config.ansi_c.long_long_int_width, true)) // long long int
{
width=config.ansi_c.long_long_int_width;
is_signed=true;
c_type=ID_signed_long_long_int;
}
else if(FITS(config.ansi_c.long_long_int_width, false)) // unsigned long long int
{
width=config.ansi_c.long_long_int_width;
is_signed=false;
c_type=ID_unsigned_long_long_int;
}
else
{
// Way too large. Should consider issuing a warning.
width=config.ansi_c.long_long_int_width;
if(is_unsigned)
{
is_signed=false;
c_type=ID_unsigned_long_long_int;
}
else
c_type=ID_signed_long_long_int;
}
typet type=typet(is_signed?ID_signedbv:ID_unsignedbv);
type.set(ID_width, width);
type.set(ID_C_c_type, c_type);
exprt result;
if(is_imaginary)
{
complex_typet complex_type;
complex_type.subtype()=type;
result=exprt(ID_complex, complex_type);
result.operands().resize(2);
result.op0()=gen_zero(type);
result.op1()=from_integer(value, type);
}
else
{
result=from_integer(value, type);
result.set(ID_C_cformat, src);
}
return result;
}
void cpp_typecheckt::do_virtual_table(const symbolt &symbol)
{
assert(symbol.type.id()==ID_struct);
// builds virtual-table value maps: (class x virtual_name x value)
std::map<irep_idt, std::map<irep_idt,exprt> > vt_value_maps;
const struct_typet &struct_type = to_struct_type(symbol.type);
for(unsigned i = 0; i < struct_type.components().size(); i++)
{
const struct_typet::componentt& compo = struct_type.components()[i];
if(!compo.get_bool("is_virtual"))
continue;
const code_typet& code_type = to_code_type(compo.type());
assert(code_type.arguments().size() > 0);
const pointer_typet& pointer_type =
static_cast<const pointer_typet&>(code_type.arguments()[0].type());
irep_idt class_id = pointer_type.subtype().get("identifier");
std::map<irep_idt,exprt>& value_map =
vt_value_maps[class_id];
exprt e = symbol_exprt(compo.get_name(),code_type);
if(compo.get_bool("is_pure_virtual"))
{
pointer_typet pointer_type(code_type);
e = gen_zero(pointer_type);
assert(e.is_not_nil());
value_map[compo.get("virtual_name")] = e;
}
else
{
address_of_exprt address(e);
value_map[compo.get("virtual_name")] = address;
}
}
// create virtual-table symbol variables
for(std::map<irep_idt, std::map<irep_idt,exprt> >::const_iterator cit =
vt_value_maps.begin(); cit != vt_value_maps.end(); cit++)
{
const std::map<irep_idt,exprt>& value_map = cit->second;
const symbolt& late_cast_symb = namespacet(symbol_table).lookup(cit->first);
const symbolt& vt_symb_type = namespacet(symbol_table).lookup("virtual_table::"+id2string(late_cast_symb.name));
symbolt vt_symb_var;
vt_symb_var.name= id2string(vt_symb_type.name) + "@"+ id2string(symbol.name);
vt_symb_var.base_name= id2string(vt_symb_type.base_name) + "@" + id2string(symbol.base_name);
vt_symb_var.mode=ID_cpp;
vt_symb_var.module=module;
vt_symb_var.location=vt_symb_type.location;
vt_symb_var.type = symbol_typet(vt_symb_type.name);
vt_symb_var.is_lvalue = true;
vt_symb_var.is_static_lifetime = true;
// do the values
const struct_typet &vt_type = to_struct_type(vt_symb_type.type);
exprt values(ID_struct, symbol_typet(vt_symb_type.name));
for(unsigned i=0; i < vt_type.components().size(); i++)
{
const struct_typet::componentt& compo = vt_type.components()[i];
std::map<irep_idt,exprt>::const_iterator cit2 =
value_map.find( compo.get("base_name"));
assert(cit2 != value_map.end());
const exprt& value = cit2->second;
assert(value.type() == compo.type());
values.operands().push_back(value);
}
vt_symb_var.value = values;
bool failed = symbol_table.move(vt_symb_var);
assert(!failed);
}
}
void goto_checkt::pointer_validity_check(
const dereference_exprt &expr,
const guardt &guard)
{
if(!enable_pointer_check)
return;
const exprt &pointer=expr.op0();
const typet &pointer_type=to_pointer_type(ns.follow(pointer.type()));
assert(base_type_eq(pointer_type.subtype(), expr.type(), ns));
local_bitvector_analysist::flagst flags=
local_bitvector_analysis->get(t, pointer);
const typet &dereference_type=pointer_type.subtype();
// For Java, we only need to check for null
if(mode==ID_java)
{
if(flags.is_unknown() || flags.is_null())
{
notequal_exprt not_eq_null(pointer, gen_zero(pointer.type()));
add_guarded_claim(
not_eq_null,
"reference is null",
"pointer dereference",
expr.find_source_location(),
expr,
guard);
}
}
else
{
if(flags.is_unknown() || flags.is_null())
{
add_guarded_claim(
not_exprt(null_pointer(pointer)),
"dereference failure: pointer NULL",
"pointer dereference",
expr.find_source_location(),
expr,
guard);
}
if(flags.is_unknown())
add_guarded_claim(
not_exprt(invalid_pointer(pointer)),
"dereference failure: pointer invalid",
"pointer dereference",
expr.find_source_location(),
expr,
guard);
if(flags.is_uninitialized())
add_guarded_claim(
not_exprt(invalid_pointer(pointer)),
"dereference failure: pointer uninitialized",
"pointer dereference",
expr.find_source_location(),
expr,
guard);
if(mode != ID_java)
{
if(flags.is_unknown() || flags.is_dynamic_heap())
add_guarded_claim(
not_exprt(deallocated(pointer, ns)),
"dereference failure: deallocated dynamic object",
"pointer dereference",
expr.find_source_location(),
expr,
guard);
if(flags.is_unknown() || flags.is_dynamic_local())
add_guarded_claim(
not_exprt(dead_object(pointer, ns)),
"dereference failure: dead object",
"pointer dereference",
expr.find_source_location(),
expr,
guard);
}
if(enable_bounds_check)
{
if(flags.is_unknown() || flags.is_dynamic_heap())
{
exprt dynamic_bounds=
or_exprt(dynamic_object_lower_bound(pointer),
dynamic_object_upper_bound(pointer, dereference_type, ns));
add_guarded_claim(
implies_exprt(malloc_object(pointer, ns), not_exprt(dynamic_bounds)),
"dereference failure: dynamic object bounds",
"pointer dereference",
expr.find_source_location(),
expr,
guard);
}
}
if(enable_bounds_check)
{
if(flags.is_unknown() ||
flags.is_dynamic_local() ||
flags.is_static_lifetime())
{
exprt object_bounds=
or_exprt(object_lower_bound(pointer),
object_upper_bound(pointer, dereference_type, ns));
add_guarded_claim(
or_exprt(dynamic_object(pointer), not_exprt(object_bounds)),
"dereference failure: object bounds",
"pointer dereference",
expr.find_source_location(),
expr,
guard);
}
}
}
}
static inline __m128 gen_ones(void)
{
__m128 x = gen_zero();
__m128 ones = _mm_cmpeq_ps(x, x);
return ones;
}
void goto_convertt::do_java_new_array(
const exprt &lhs,
const side_effect_exprt &rhs,
goto_programt &dest)
{
if(lhs.is_nil())
throw "do_java_new_array without lhs is yet to be implemented";
source_locationt location=rhs.source_location();
assert(rhs.operands().size()>=1); // one per dimension
if(rhs.type().id()!=ID_pointer)
throw "do_java_new_array returns pointer";
typet object_type=rhs.type().subtype();
// build size expression
exprt object_size=size_of_expr(object_type, ns);
if(object_size.is_nil())
throw "do_java_new_array got nil object_size";
// we produce a malloc side-effect, which stays
side_effect_exprt malloc_expr(ID_malloc);
malloc_expr.copy_to_operands(object_size);
malloc_expr.type()=pointer_typet(object_type);
goto_programt::targett t_n=dest.add_instruction(ASSIGN);
t_n->code=code_assignt(lhs, malloc_expr);
t_n->source_location=location;
// multi-dimensional?
assert(ns.follow(object_type).id()==ID_struct);
const struct_typet &struct_type=to_struct_type(ns.follow(object_type));
assert(struct_type.components().size()==3);
// if it's an array, we need to set the length field
dereference_exprt deref(lhs, object_type);
member_exprt length(deref, struct_type.components()[1].get_name(), struct_type.components()[1].type());
goto_programt::targett t_s=dest.add_instruction(ASSIGN);
t_s->code=code_assignt(length, rhs.op0());
t_s->source_location=location;
// we also need to allocate space for the data
member_exprt data(deref, struct_type.components()[2].get_name(), struct_type.components()[2].type());
side_effect_exprt data_cpp_new_expr(ID_cpp_new_array, data.type());
data_cpp_new_expr.set(ID_size, rhs.op0());
goto_programt::targett t_p=dest.add_instruction(ASSIGN);
t_p->code=code_assignt(data, data_cpp_new_expr);
t_p->source_location=location;
// zero-initialize the data
exprt zero_element=gen_zero(data.type().subtype());
codet array_set(ID_array_set);
array_set.copy_to_operands(data, zero_element);
goto_programt::targett t_d=dest.add_instruction(OTHER);
t_d->code=array_set;
t_d->source_location=location;
if(rhs.operands().size()>=2)
{
// produce
// for(int i=0; i<size; i++) tmp[i]=java_new(dim-1);
// This will be converted recursively.
goto_programt tmp;
symbol_exprt tmp_i=
new_tmp_symbol(index_type(), "index", tmp, location).symbol_expr();
code_fort for_loop;
side_effect_exprt sub_java_new=rhs;
sub_java_new.operands().erase(sub_java_new.operands().begin());
side_effect_exprt inc(ID_assign);
inc.operands().resize(2);
inc.op0()=tmp_i;
inc.op1()=plus_exprt(tmp_i, gen_one(tmp_i.type()));
dereference_exprt deref_expr(plus_exprt(data, tmp_i), data.type().subtype());
for_loop.init()=code_assignt(tmp_i, gen_zero(tmp_i.type()));
for_loop.cond()=binary_relation_exprt(tmp_i, ID_lt, rhs.op0());
for_loop.iter()=inc;
for_loop.body()=code_skipt();
for_loop.body()=code_assignt(deref_expr, sub_java_new);
convert(for_loop, tmp);
dest.destructive_append(tmp);
}
}
void dplib_convt::convert_dplib_expr(const exprt &expr)
{
if(expr.id()==ID_symbol)
{
convert_identifier(expr.get_string(ID_identifier));
}
else if(expr.id()==ID_nondet_symbol)
{
convert_identifier("nondet$"+expr.get_string(ID_identifier));
}
else if(expr.id()==ID_typecast)
{
assert(expr.operands().size()==1);
const exprt &op=expr.op0();
if(expr.type().id()==ID_bool)
{
if(op.type().id()==ID_signedbv ||
op.type().id()==ID_unsignedbv ||
op.type().id()==ID_pointer)
{
convert_dplib_expr(op);
dplib_prop.out << "/=";
convert_dplib_expr(gen_zero(op.type()));
}
else
{
throw "TODO typecast1 "+op.type().id_string()+" -> bool";
}
}
else if(expr.type().id()==ID_signedbv ||
expr.type().id()==ID_unsignedbv)
{
unsigned to_width=unsafe_string2unsigned(id2string(expr.type().get(ID_width)));
if(op.type().id()==ID_signedbv)
{
unsigned from_width=unsafe_string2unsigned(id2string(op.type().get(ID_width)));
if(from_width==to_width)
convert_dplib_expr(op);
else if(from_width<to_width)
{
dplib_prop.out << "SX(";
convert_dplib_expr(op);
dplib_prop.out << ", " << to_width << ")";
}
else
{
dplib_prop.out << "(";
convert_dplib_expr(op);
dplib_prop.out << ")[" << (to_width-1) << ":0]";
}
}
else if(op.type().id()==ID_unsignedbv)
{
unsigned from_width=unsafe_string2unsigned(id2string(op.type().get(ID_width)));
if(from_width==to_width)
convert_dplib_expr(op);
else if(from_width<to_width)
{
dplib_prop.out << "(0bin";
for(unsigned i=from_width; i<to_width; i++)
dplib_prop.out << "0";
dplib_prop.out << " @ ";
dplib_prop.out << "(";
convert_dplib_expr(op);
dplib_prop.out << "))";
}
else
{
dplib_prop.out << "(";
convert_dplib_expr(op);
dplib_prop.out << ")[" << (to_width-1) << ":0]";
}
}
else if(op.type().id()==ID_bool)
{
if(to_width>1)
{
dplib_prop.out << "(0bin";
for(unsigned i=1; i<to_width; i++)
dplib_prop.out << "0";
dplib_prop.out << " @ ";
dplib_prop.out << "IF ";
convert_dplib_expr(op);
dplib_prop.out << " THEN 0bin1 ELSE 0bin0 ENDIF)";
}
else
{
dplib_prop.out << "IF ";
convert_dplib_expr(op);
dplib_prop.out << " THEN 0bin1 ELSE 0bin0 ENDIF";
}
}
else
{
throw "TODO typecast2 "+op.type().id_string()+
" -> "+expr.type().id_string();
}
}
else if(expr.type().id()==ID_pointer)
{
if(op.type().id()==ID_pointer)
{
convert_dplib_expr(op);
}
else
throw "TODO typecast3 "+op.type().id_string()+" -> pointer";
}
else
throw "TODO typecast4 ? -> "+expr.type().id_string();
}
else if(expr.id()==ID_struct)
{
dplib_prop.out << "(# ";
const struct_typet &struct_type=to_struct_type(expr.type());
const struct_typet::componentst &components=
struct_type.components();
assert(components.size()==expr.operands().size());
unsigned i=0;
for(struct_typet::componentst::const_iterator
it=components.begin();
it!=components.end();
it++, i++)
{
if(i!=0) dplib_prop.out << ", ";
dplib_prop.out << it->get(ID_name);
dplib_prop.out << ":=";
convert_dplib_expr(expr.operands()[i]);
}
dplib_prop.out << " #)";
}
else if(expr.id()==ID_constant)
{
if(expr.type().id()==ID_unsignedbv ||
expr.type().id()==ID_signedbv ||
expr.type().id()==ID_bv)
{
dplib_prop.out << "0bin" << expr.get(ID_value);
}
else if(expr.type().id()==ID_pointer)
{
const irep_idt &value=expr.get(ID_value);
if(value=="NULL")
{
dplib_prop.out << "(# object:="
<< pointer_logic.get_null_object()
<< ", offset:="
<< bin_zero(config.ansi_c.pointer_width) << " #)";
}
else
throw "unknown pointer constant: "+id2string(value);
}
else if(expr.type().id()==ID_bool)
{
if(expr.is_true())
dplib_prop.out << "TRUE";
else if(expr.is_false())
dplib_prop.out << "FALSE";
else
throw "unknown boolean constant";
}
else if(expr.type().id()==ID_array)
{
dplib_prop.out << "ARRAY (i: " << array_index_type() << "):";
assert(!expr.operands().empty());
unsigned i=0;
forall_operands(it, expr)
{
if(i==0)
dplib_prop.out << "\n IF ";
else
dplib_prop.out << "\n ELSIF ";
dplib_prop.out << "i=" << array_index(i) << " THEN ";
convert_array_value(*it);
i++;
}
dplib_prop.out << "\n ELSE ";
convert_dplib_expr(expr.op0());
dplib_prop.out << "\n ENDIF";
}
else if(expr.type().id()==ID_integer ||
void goto_checkt::goto_check(goto_functiont &goto_function)
{
{
const symbolt *init_symbol;
if(!ns.lookup(CPROVER_PREFIX "initialize", init_symbol))
mode=init_symbol->mode;
}
assertions.clear();
local_bitvector_analysist local_bitvector_analysis_obj(goto_function);
local_bitvector_analysis=&local_bitvector_analysis_obj;
goto_programt &goto_program=goto_function.body;
Forall_goto_program_instructions(it, goto_program)
{
t=it;
goto_programt::instructiont &i=*it;
new_code.clear();
// we clear all recorded assertions if
// 1) we want to generate all assertions or
// 2) the instruction is a branch target
if(retain_trivial ||
i.is_target())
assertions.clear();
check(i.guard);
// magic ERROR label?
for(optionst::value_listt::const_iterator
l_it=error_labels.begin();
l_it!=error_labels.end();
l_it++)
{
if(std::find(i.labels.begin(), i.labels.end(), *l_it)!=i.labels.end())
{
goto_program_instruction_typet type=
enable_assert_to_assume?ASSUME:ASSERT;
goto_programt::targett t=new_code.add_instruction(type);
t->guard=false_exprt();
t->source_location=i.source_location;
t->source_location.set_property_class("error label");
t->source_location.set_comment("error label "+*l_it);
t->source_location.set("user-provided", true);
}
}
if(i.is_other())
{
const irep_idt &statement=i.code.get(ID_statement);
if(statement==ID_expression)
{
check(i.code);
}
else if(statement==ID_printf)
{
forall_operands(it, i.code)
check(*it);
}
}
else if(i.is_assign())
{
const code_assignt &code_assign=to_code_assign(i.code);
check(code_assign.lhs());
check(code_assign.rhs());
// the LHS might invalidate any assertion
invalidate(code_assign.lhs());
}
else if(i.is_function_call())
{
const code_function_callt &code_function_call=
to_code_function_call(i.code);
// for Java, need to check whether 'this' is null
// on non-static method invocations
if(mode==ID_java &&
enable_pointer_check &&
!code_function_call.arguments().empty() &&
code_function_call.function().type().id()==ID_code &&
to_code_type(code_function_call.function().type()).has_this())
{
exprt pointer=code_function_call.arguments()[0];
local_bitvector_analysist::flagst flags=
local_bitvector_analysis->get(t, pointer);
if(flags.is_unknown() || flags.is_null())
{
notequal_exprt not_eq_null(pointer, gen_zero(pointer.type()));
add_guarded_claim(
not_eq_null,
"this is null on method invokation",
"pointer dereference",
i.source_location,
pointer,
guardt());
}
}
forall_operands(it, code_function_call)
check(*it);
// the call might invalidate any assertion
assertions.clear();
}
else if(i.is_return())
{
if(i.code.operands().size()==1)
{
check(i.code.op0());
// the return value invalidate any assertion
invalidate(i.code.op0());
}
}
else if(i.is_throw())
{
if(i.code.get_statement()==ID_expression &&
i.code.operands().size()==1 &&
i.code.op0().operands().size()==1)
{
// must not throw NULL
exprt pointer=i.code.op0().op0();
if(pointer.type().subtype().get(ID_identifier)!="java::java.lang.AssertionError")
{
notequal_exprt not_eq_null(pointer, gen_zero(pointer.type()));
add_guarded_claim(
not_eq_null,
"throwing null",
"pointer dereference",
i.source_location,
pointer,
guardt());
}
}
// this has no successor
assertions.clear();
}
else if(i.is_assert())
{
if(i.source_location.get_bool("user-provided") &&
i.source_location.get_property_class()!="error label" &&
!enable_assertions)
i.type=SKIP;
}
else if(i.is_assume())
{
if(!enable_assumptions)
i.type=SKIP;
}
else if(i.is_dead())
{
if(enable_pointer_check)
{
assert(i.code.operands().size()==1);
const symbol_exprt &variable=to_symbol_expr(i.code.op0());
// is it dirty?
if(local_bitvector_analysis->dirty(variable))
{
// need to mark the dead variable as dead
goto_programt::targett t=new_code.add_instruction(ASSIGN);
exprt address_of_expr=address_of_exprt(variable);
exprt lhs=ns.lookup(CPROVER_PREFIX "dead_object").symbol_expr();
if(!base_type_eq(lhs.type(), address_of_expr.type(), ns))
address_of_expr.make_typecast(lhs.type());
exprt rhs=if_exprt(
side_effect_expr_nondett(bool_typet()), address_of_expr, lhs, lhs.type());
t->source_location=i.source_location;
t->code=code_assignt(lhs, rhs);
t->code.add_source_location()=i.source_location;
}
}
}
else if(i.is_end_function())
{
if(i.function==goto_functionst::entry_point() &&
enable_memory_leak_check)
{
const symbolt &leak=ns.lookup(CPROVER_PREFIX "memory_leak");
const symbol_exprt leak_expr=leak.symbol_expr();
// add self-assignment to get helpful counterexample output
goto_programt::targett t=new_code.add_instruction();
t->make_assignment();
t->code=code_assignt(leak_expr, leak_expr);
source_locationt source_location;
source_location.set_function(i.function);
equal_exprt eq(leak_expr, gen_zero(ns.follow(leak.type)));
add_guarded_claim(
eq,
"dynamically allocated memory never freed",
"memory-leak",
source_location,
eq,
guardt());
}
}
for(goto_programt::instructionst::iterator
i_it=new_code.instructions.begin();
i_it!=new_code.instructions.end();
i_it++)
{
if(i_it->source_location.is_nil())
{
i_it->source_location.id(irep_idt());
if(it->source_location.get_file()!=irep_idt())
i_it->source_location.set_file(it->source_location.get_file());
if(it->source_location.get_line()!=irep_idt())
i_it->source_location.set_line(it->source_location.get_line());
if(it->source_location.get_function()!=irep_idt())
i_it->source_location.set_function(it->source_location.get_function());
if(it->source_location.get_column()!=irep_idt())
i_it->source_location.set_column(it->source_location.get_column());
}
if(i_it->function==irep_idt()) i_it->function=it->function;
}
// insert new instructions -- make sure targets are not moved
while(!new_code.instructions.empty())
{
goto_program.insert_before_swap(it, new_code.instructions.front());
new_code.instructions.pop_front();
it++;
}
}
void string_instrumentationt::invalidate_buffer(
goto_programt &dest,
goto_programt::const_targett target,
const exprt &buffer,
const typet &buf_type,
const mp_integer &limit)
{
irep_idt cntr_id="string_instrumentation::$counter";
if(context.symbols.find(cntr_id)==context.symbols.end())
{
symbolt new_symbol;
new_symbol.base_name="$counter";
new_symbol.pretty_name=new_symbol.base_name;
new_symbol.name=cntr_id;
new_symbol.mode="C";
new_symbol.type=uint_type();
new_symbol.is_statevar=true;
new_symbol.lvalue=true;
new_symbol.static_lifetime=true;
context.move(new_symbol);
}
const symbolt &cntr_sym=ns.lookup(cntr_id);
// create a loop that runs over the buffer
// and invalidates every element
goto_programt::targett init=dest.add_instruction(ASSIGN);
init->location=target->location;
init->code=code_assignt(symbol_expr(cntr_sym), gen_zero(cntr_sym.type));
goto_programt::targett check=dest.add_instruction();
check->location=target->location;
goto_programt::targett invalidate=dest.add_instruction(ASSIGN);
invalidate->location=target->location;
goto_programt::targett increment=dest.add_instruction(ASSIGN);
increment->location=target->location;
exprt plus("+", uint_type());
plus.copy_to_operands(symbol_expr(cntr_sym));
plus.copy_to_operands(gen_one(uint_type()));
increment->code=code_assignt(symbol_expr(cntr_sym), plus);
goto_programt::targett back=dest.add_instruction();
back->location=target->location;
back->make_goto(check);
back->guard=true_exprt();
goto_programt::targett exit=dest.add_instruction();
exit->location=target->location;
exit->make_skip();
exprt cnt_bs, bufp;
if(buf_type.id()=="pointer")
bufp = buffer;
else
{
index_exprt index;
index.array()=buffer;
index.index()=gen_zero(uint_type());
index.type()=buf_type.subtype();
bufp = address_of_exprt(index);
}
exprt deref("dereference", buf_type.subtype());
exprt b_plus_i("+", bufp.type());
b_plus_i.copy_to_operands(bufp);
b_plus_i.copy_to_operands(symbol_expr(cntr_sym));
deref.copy_to_operands(b_plus_i);
check->make_goto(exit);
if(limit==0)
check->guard=
binary_relation_exprt(symbol_expr(cntr_sym), ">=",
buffer_size(bufp));
else
check->guard=
binary_relation_exprt(symbol_expr(cntr_sym), ">",
from_integer(limit, uint_type()));
exprt nondet=side_effect_expr_nondett(buf_type.subtype());
invalidate->code=code_assignt(deref, nondet);
}
void goto_checkt::bounds_check(
const index_exprt &expr,
const guardt &guard)
{
if(!enable_bounds_check)
return;
if(expr.find("bounds_check").is_not_nil() &&
!expr.get_bool("bounds_check"))
return;
typet array_type=ns.follow(expr.array().type());
if(array_type.id()==ID_pointer)
return; // done by the pointer code
else if(array_type.id()==ID_incomplete_array)
throw "index got incomplete array";
else if(array_type.id()!=ID_array && array_type.id()!=ID_vector)
throw "bounds check expected array or vector type, got "
+array_type.id_string();
std::string name=array_name(expr.array());
const exprt &index=expr.index();
object_descriptor_exprt ode;
ode.build(expr, ns);
if(index.type().id()!=ID_unsignedbv)
{
// we undo typecasts to signedbv
if(index.id()==ID_typecast &&
index.operands().size()==1 &&
index.op0().type().id()==ID_unsignedbv)
{
// ok
}
else
{
mp_integer i;
if(!to_integer(index, i) && i>=0)
{
// ok
}
else
{
exprt effective_offset=ode.offset();
if(ode.root_object().id()==ID_dereference)
{
exprt p_offset=pointer_offset(
to_dereference_expr(ode.root_object()).pointer());
assert(p_offset.type()==effective_offset.type());
effective_offset=plus_exprt(p_offset, effective_offset);
}
exprt zero=gen_zero(ode.offset().type());
assert(zero.is_not_nil());
// the final offset must not be negative
binary_relation_exprt inequality(effective_offset, ID_ge, zero);
add_guarded_claim(
inequality,
name+" lower bound",
"array bounds",
expr.find_source_location(),
expr,
guard);
}
}
}
if(ode.root_object().id()==ID_dereference)
{
const exprt &pointer=
to_dereference_expr(ode.root_object()).pointer();
if_exprt size(
dynamic_object(pointer),
typecast_exprt(dynamic_size(ns), object_size(pointer).type()),
object_size(pointer));
plus_exprt effective_offset(ode.offset(), pointer_offset(pointer));
assert(effective_offset.op0().type()==effective_offset.op1().type());
assert(effective_offset.type()==size.type());
binary_relation_exprt inequality(effective_offset, ID_lt, size);
or_exprt precond(
and_exprt(
dynamic_object(pointer),
not_exprt(malloc_object(pointer, ns))),
inequality);
add_guarded_claim(
precond,
name+" upper bound",
"array bounds",
expr.find_source_location(),
expr,
guard);
return;
}
const exprt &size=array_type.id()==ID_array ?
to_array_type(array_type).size() :
to_vector_type(array_type).size();
if(size.is_nil())
{
// Linking didn't complete, we don't have a size.
// Not clear what to do.
}
else if(size.id()==ID_infinity)
{
}
else if(size.is_zero() &&
expr.array().id()==ID_member)
{
// a variable sized struct member
}
else
{
binary_relation_exprt inequality(index, ID_lt, size);
// typecast size
if(inequality.op1().type()!=inequality.op0().type())
inequality.op1().make_typecast(inequality.op0().type());
// typecast size
if(inequality.op1().type()!=inequality.op0().type())
inequality.op1().make_typecast(inequality.op0().type());
add_guarded_claim(
inequality,
name+" upper bound",
"array bounds",
expr.find_source_location(),
expr,
guard);
}
}
void string_instrumentationt::do_format_string_read(
goto_programt &dest,
goto_programt::const_targett target,
const code_function_callt::argumentst &arguments,
unsigned format_string_inx,
unsigned argument_start_inx,
const std::string &function_name)
{
const exprt &format_arg = arguments[format_string_inx];
if(format_arg.id()=="address_of" &&
format_arg.op0().id()=="index" &&
format_arg.op0().op0().id()==ID_string_constant)
{
format_token_listt token_list;
parse_format_string(format_arg.op0().op0(), token_list);
unsigned args=0;
for(format_token_listt::const_iterator it=token_list.begin();
it!=token_list.end();
it++)
{
if(it->type==format_tokent::STRING)
{
const exprt &arg = arguments[argument_start_inx+args];
const typet &arg_type = ns.follow(arg.type());
if(arg.id()!=ID_string_constant) // we don't need to check constants
{
goto_programt::targett assertion=dest.add_instruction();
assertion->location=target->location;
assertion->location.set("property", "string");
std::string comment("zero-termination of string argument of ");
comment += function_name;
assertion->location.set("comment", comment);
exprt temp(arg);
if(arg_type.id()!="pointer")
{
index_exprt index;
index.array()=temp;
index.index()=gen_zero(uint_type());
index.type()=arg_type.subtype();
temp=address_of_exprt(index);
}
assertion->make_assertion(is_zero_string(temp));
}
}
if(it->type!=format_tokent::TEXT &&
it->type!=format_tokent::UNKNOWN) args++;
if(find(it->flags.begin(), it->flags.end(), format_tokent::ASTERISK)!=
it->flags.end())
args++; // just eat the additional argument
}
}
else // non-const format string
{
goto_programt::targett format_ass=dest.add_instruction();
format_ass->make_assertion(is_zero_string(arguments[1]));
format_ass->location=target->location;
format_ass->location.set("property", "string");
std::string comment("zero-termination of format string of ");
comment += function_name;
format_ass->location.set("comment", comment);
for(unsigned i=2; i<arguments.size(); i++)
{
const exprt &arg = arguments[i];
const typet &arg_type=ns.follow(arguments[i].type());
if(arguments[i].id()!=ID_string_constant &&
is_string_type(arg_type))
{
goto_programt::targett assertion=dest.add_instruction();
assertion->location=target->location;
assertion->location.set("property", "string");
std::string comment("zero-termination of string argument of ");
comment += function_name;
assertion->location.set("comment", comment);
exprt temp(arg);
if(arg_type.id()!="pointer")
{
index_exprt index;
index.array()=temp;
index.index()=gen_zero(uint_type());
index.type()=arg_type.subtype();
temp=address_of_exprt(index);
}
assertion->make_assertion(is_zero_string(temp));
}
}
}
}
void string_instrumentationt::do_format_string_write(
goto_programt &dest,
goto_programt::const_targett target,
const code_function_callt::argumentst &arguments,
unsigned format_string_inx,
unsigned argument_start_inx,
const std::string &function_name)
{
const exprt &format_arg = arguments[format_string_inx];
if(format_arg.id()=="address_of" &&
format_arg.op0().id()=="index" &&
format_arg.op0().op0().id()==ID_string_constant) // constant format
{
format_token_listt token_list;
parse_format_string(format_arg.op0().op0(), token_list);
unsigned args=0;
for(format_token_listt::const_iterator it=token_list.begin();
it!=token_list.end();
it++)
{
if(find(it->flags.begin(), it->flags.end(), format_tokent::ASTERISK)!=
it->flags.end())
continue; // asterisk means `ignore this'
switch(it->type)
{
case format_tokent::STRING:
{
const exprt &argument=arguments[argument_start_inx+args];
const typet &arg_type=ns.follow(argument.type());
goto_programt::targett assertion=dest.add_instruction();
assertion->location=target->location;
assertion->location.set("property", "string");
std::string comment("format string buffer overflow in ");
comment += function_name;
assertion->location.set("comment", comment);
if(it->field_width!=0)
{
exprt fwidth = from_integer(it->field_width, uint_type());
exprt fw_1("+", uint_type());
exprt one = gen_one(uint_type());
fw_1.move_to_operands(fwidth);
fw_1.move_to_operands(one); // +1 for 0-char
exprt fw_lt_bs;
if(arg_type.id()=="pointer")
fw_lt_bs=binary_relation_exprt(fw_1, "<=", buffer_size(argument));
else
{
index_exprt index;
index.array()=argument;
index.index()=gen_zero(uint_type());
address_of_exprt aof(index);
fw_lt_bs=binary_relation_exprt(fw_1, "<=", buffer_size(aof));
}
assertion->make_assertion(fw_lt_bs);
}
else
{
// this is a possible overflow.
assertion->make_assertion(false_exprt());
}
// now kill the contents
invalidate_buffer(dest, target, argument, arg_type, it->field_width);
args++;
break;
}
case format_tokent::TEXT:
case format_tokent::UNKNOWN:
{
// nothing
break;
}
default: // everything else
{
const exprt &argument=arguments[argument_start_inx+args];
const typet &arg_type=ns.follow(argument.type());
goto_programt::targett assignment=dest.add_instruction(ASSIGN);
assignment->location=target->location;
exprt lhs("dereference", arg_type.subtype());
lhs.copy_to_operands(argument);
exprt rhs=side_effect_expr_nondett(lhs.type());
rhs.location()=target->location;
assignment->code=code_assignt(lhs, rhs);
args++;
break;
}
}
}
}
else // non-const format string
{
for(unsigned i=argument_start_inx; i<arguments.size(); i++)
{
const typet &arg_type=ns.follow(arguments[i].type());
// Note: is_string_type() is a `good guess' here. Actually
// any of the pointers could point into an array. But it
// would suck if we had to invalidate all variables.
// Luckily this case isn't needed too often.
if(is_string_type(arg_type))
{
goto_programt::targett assertion=dest.add_instruction();
assertion->location=target->location;
assertion->location.set("property", "string");
std::string comment("format string buffer overflow in ");
comment += function_name;
assertion->location.set("comment", comment);
// as we don't know any field width for the %s that
// should be here during runtime, we just report a
// possibly false positive
assertion->make_assertion(false_exprt());
invalidate_buffer(dest, target, arguments[i], arg_type, 0);
}
else
{
goto_programt::targett assignment = dest.add_instruction(ASSIGN);
assignment->location=target->location;
exprt lhs("dereference", arg_type.subtype());
lhs.copy_to_operands(arguments[i]);
exprt rhs=side_effect_expr_nondett(lhs.type());
rhs.location()=target->location;
assignment->code=code_assignt(lhs, rhs);
}
}
}
}
codet java_bytecode_convertt::convert_instructions(
const instructionst &instructions,
const code_typet &method_type)
{
// Run a worklist algorithm, assuming that the bytecode has not
// been tampered with. See "Leroy, X. (2003). Java bytecode
// verification: algorithms and formalizations. Journal of Automated
// Reasoning, 30(3-4), 235-269." for a more complete treatment.
// first pass: get targets and map addresses to instructions
struct converted_instructiont
{
converted_instructiont(
const instructionst::const_iterator &it,
const codet &_code):source(it), code(_code), done(false)
{
}
instructionst::const_iterator source;
std::list<unsigned> successors;
std::set<unsigned> predecessors;
codet code;
stackt stack;
bool done;
};
typedef std::map<unsigned, converted_instructiont> address_mapt;
address_mapt address_map;
std::set<unsigned> targets;
for(instructionst::const_iterator
i_it=instructions.begin();
i_it!=instructions.end();
i_it++)
{
std::pair<address_mapt::iterator, bool> a_entry=
address_map.insert(std::make_pair(
i_it->address,
converted_instructiont(i_it, code_skipt())));
assert(a_entry.second);
// addresses are strictly increasing, hence we must have inserted
// a new maximal key
assert(a_entry.first==--address_map.end());
if(i_it->statement!="goto" &&
i_it->statement!="return" &&
!(i_it->statement==patternt("?return")) &&
i_it->statement!="athrow")
{
instructionst::const_iterator next=i_it;
if(++next!=instructions.end())
a_entry.first->second.successors.push_back(next->address);
}
if(i_it->statement=="goto" ||
i_it->statement==patternt("if_?cmp??") ||
i_it->statement==patternt("if??") ||
i_it->statement=="ifnonnull" ||
i_it->statement=="ifnull")
{
assert(!i_it->args.empty());
const unsigned target=safe_string2unsigned(
id2string(to_constant_expr(i_it->args[0]).get_value()));
targets.insert(target);
a_entry.first->second.successors.push_back(target);
}
else if(i_it->statement=="tableswitch" ||
i_it->statement=="lookupswitch")
{
bool is_label=true;
for(instructiont::argst::const_iterator
a_it=i_it->args.begin();
a_it!=i_it->args.end();
a_it++, is_label=!is_label)
{
if(is_label)
{
const unsigned target=safe_string2unsigned(
id2string(to_constant_expr(*a_it).get_value()));
targets.insert(target);
a_entry.first->second.successors.push_back(target);
}
}
}
}
for(address_mapt::iterator
it=address_map.begin();
it!=address_map.end();
++it)
{
for(unsigned s : it->second.successors)
{
address_mapt::iterator a_it=address_map.find(s);
assert(a_it!=address_map.end());
a_it->second.predecessors.insert(it->first);
}
}
std::set<unsigned> working_set;
if(!instructions.empty())
working_set.insert(instructions.front().address);
while(!working_set.empty())
{
std::set<unsigned>::iterator cur=working_set.begin();
address_mapt::iterator a_it=address_map.find(*cur);
assert(a_it!=address_map.end());
working_set.erase(cur);
if(a_it->second.done) continue;
working_set.insert(a_it->second.successors.begin(),
a_it->second.successors.end());
instructionst::const_iterator i_it=a_it->second.source;
stack.swap(a_it->second.stack);
a_it->second.stack.clear();
codet &c=a_it->second.code;
assert(stack.empty() ||
a_it->second.predecessors.size()<=1 ||
has_prefix(stack.front().get_string(ID_C_base_name),
"$stack"));
irep_idt statement=i_it->statement;
exprt arg0=i_it->args.size()>=1?i_it->args[0]:nil_exprt();
exprt arg1=i_it->args.size()>=2?i_it->args[1]:nil_exprt();
const bytecode_infot &bytecode_info=get_bytecode_info(statement);
// deal with _idx suffixes
if(statement.size()>=2 &&
statement[statement.size()-2]=='_' &&
isdigit(statement[statement.size()-1]))
{
arg0=constant_exprt(
std::string(id2string(statement), statement.size()-1, 1),
integer_typet());
statement=std::string(id2string(statement), 0, statement.size()-2);
}
exprt::operandst op=pop(bytecode_info.pop);
exprt::operandst results;
results.resize(bytecode_info.push, nil_exprt());
if(statement=="aconst_null")
{
assert(results.size()==1);
results[0]=gen_zero(java_reference_type(void_typet()));
}
else if(statement=="athrow")
{
assert(op.size()==1 && results.size()==1);
side_effect_expr_throwt throw_expr;
throw_expr.add_source_location()=i_it->source_location;
throw_expr.copy_to_operands(op[0]);
c=code_expressiont(throw_expr);
results[0]=op[0];
}
else if(statement=="checkcast")
{
// checkcast throws an exception in case a cast of object
// on stack to given type fails.
// The stack isn't modified.
assert(op.size()==1 && results.size()==1);
results[0]=op[0];
}
else if(statement=="invokedynamic")
{
// not used in Java
code_typet &code_type=to_code_type(arg0.type());
const code_typet::parameterst ¶meters(code_type.parameters());
pop(parameters.size());
const typet &return_type=code_type.return_type();
if(return_type.id()!=ID_empty)
{
results.resize(1);
results[0]=nil_exprt();
}
}
else if(statement=="invokeinterface" ||
statement=="invokespecial" ||
statement=="invokevirtual" ||
statement=="invokestatic")
{
const bool use_this(statement != "invokestatic");
const bool is_virtual(
statement == "invokevirtual" || statement == "invokeinterface");
code_typet &code_type=to_code_type(arg0.type());
code_typet::parameterst ¶meters(code_type.parameters());
if(use_this)
{
if(parameters.empty() || !parameters[0].get_this())
{
const empty_typet empty;
pointer_typet object_ref_type(empty);
code_typet::parametert this_p(object_ref_type);
this_p.set_this();
this_p.set_base_name("this");
parameters.insert(parameters.begin(), this_p);
}
}
code_function_callt call;
call.add_source_location()=i_it->source_location;
call.arguments() = pop(parameters.size());
// double-check a bit
if(use_this)
{
const exprt &this_arg=call.arguments().front();
assert(this_arg.type().id()==ID_pointer);
}
// do some type adjustment for the arguments,
// as Java promotes arguments
for(unsigned i=0; i<parameters.size(); i++)
{
const typet &type=parameters[i].type();
if(type==java_boolean_type() ||
type==java_char_type() ||
type==java_byte_type() ||
type==java_short_type())
{
assert(i<call.arguments().size());
call.arguments()[i].make_typecast(type);
}
}
// do some type adjustment for return values
const typet &return_type=code_type.return_type();
if(return_type.id()!=ID_empty)
{
// return types are promoted in Java
call.lhs()=tmp_variable("return", return_type);
exprt promoted=java_bytecode_promotion(call.lhs());
results.resize(1);
results[0]=promoted;
}
assert(arg0.id()==ID_virtual_function);
// does the function symbol exist?
irep_idt id=arg0.get(ID_identifier);
if(symbol_table.symbols.find(id)==symbol_table.symbols.end())
{
// no, create stub
symbolt symbol;
symbol.name=id;
symbol.base_name=arg0.get(ID_C_base_name);
symbol.type=arg0.type();
symbol.value.make_nil();
symbol.mode=ID_java;
symbol_table.add(symbol);
}
if(is_virtual)
{
// dynamic binding
assert(use_this);
assert(!call.arguments().empty());
call.function()=arg0;
}
else
{
// static binding
/*if(id == "java::java.lang.String.charAt:(I)C")
call.function()=symbol_exprt("java::__CPROVER_uninterpreted_char_at", arg0.type());
else*/
call.function()=symbol_exprt(arg0.get(ID_identifier), arg0.type());
}
call.function().add_source_location()=i_it->source_location;
c = call;
}
else if(statement=="return")
{
assert(op.empty() && results.empty());
c=code_returnt();
}
else if(statement==patternt("?return"))
{
// Return types are promoted in java, so this might need
// conversion.
assert(op.size()==1 && results.empty());
exprt r=op[0];
if(r.type()!=method_return_type) r=typecast_exprt(r, method_return_type);
c=code_returnt(r);
}
else if(statement==patternt("?astore"))
{
assert(op.size()==3 && results.empty());
char type_char=statement[0];
exprt pointer=
typecast_exprt(op[0], java_array_type(type_char));
const dereference_exprt deref(pointer, pointer.type().subtype());
const member_exprt data_ptr(
deref, "data", pointer_typet(java_type_from_char(type_char)));
plus_exprt data_plus_offset(data_ptr, op[1], data_ptr.type());
typet element_type=data_ptr.type().subtype();
const dereference_exprt element(data_plus_offset, element_type);
c=code_assignt(element, op[2]);
}
else if(statement==patternt("?store"))
{
// store value into some local variable
assert(op.size()==1 && results.empty());
exprt var=variable(arg0, statement[0]);
const bool is_array('a' == statement[0]);
if(is_array)
var.type()=op[0].type();
c=code_assignt(var, op[0]);
}
else if(statement==patternt("?aload"))
{
assert(op.size() == 2 && results.size() == 1);
char type_char=statement[0];
exprt pointer=
typecast_exprt(op[0], java_array_type(type_char));
const dereference_exprt deref(pointer, pointer.type().subtype());
const member_exprt data_ptr(
deref, "data", pointer_typet(java_type_from_char(type_char)));
plus_exprt data_plus_offset(data_ptr, op[1], data_ptr.type());
typet element_type=data_ptr.type().subtype();
dereference_exprt element(data_plus_offset, element_type);
results[0]=java_bytecode_promotion(element);
}
else if(statement==patternt("?load"))
{
// load a value from a local variable
results[0]=variable(arg0, statement[0]);
}
else if(statement=="ldc" || statement=="ldc_w" ||
statement=="ldc2" || statement=="ldc2_w")
{
assert(op.empty() && results.size()==1);
// 1) Pushing a String causes a reference to a java.lang.String object
// to be constructed and pushed onto the operand stack.
// 2) Pushing an int or a float causes a primitive value to be pushed
// onto the stack.
// 3) Pushing a Class constant causes a reference to a java.lang.Class
// to be pushed onto the operand stack
if(arg0.id()==ID_java_string_literal)
{
// these need to be references to java.lang.String
results[0]=arg0;
symbol_typet string_type("java::java.lang.String");
results[0].type()=pointer_typet(string_type);
}
else if(arg0.id()==ID_type)
{
irep_idt class_id=arg0.type().get(ID_identifier);
symbol_typet java_lang_Class("java::java.lang.Class");
symbol_exprt symbol_expr(id2string(class_id)+"@class_model", java_lang_Class);
address_of_exprt address_of_expr(symbol_expr);
results[0]=address_of_expr;
}
else if(arg0.id()==ID_constant)
{
results[0]=arg0;
}
else
{
error() << "unexpected ldc argument" << eom;
throw 0;
}
}
else if(statement=="goto" || statement=="goto_w")
{
assert(op.empty() && results.empty());
irep_idt number=to_constant_expr(arg0).get_value();
code_gotot code_goto(label(number));
c=code_goto;
}
else if(statement=="iconst_m1")
{
assert(results.size()==1);
results[0]=from_integer(-1, java_int_type());
}
else if(statement==patternt("?const"))
{
assert(results.size() == 1);
const char type_char=statement[0];
const bool is_double('d' == type_char);
const bool is_float('f' == type_char);
if(is_double || is_float)
{
const ieee_float_spect spec(
is_float ?
ieee_float_spect::single_precision() :
ieee_float_spect::double_precision());
ieee_floatt value(spec);
const typet &arg_type(arg0.type());
if(ID_integer == arg_type.id())
value.from_integer(arg0.get_int(ID_value));
else
value.from_expr(to_constant_expr(arg0));
results[0] = value.to_expr();
}
else
{
const unsigned int value(arg0.get_unsigned_int(ID_value));
const typet type=java_type_from_char(statement[0]);
results[0] = as_number(value, type);
}
}
else if(statement==patternt("?ipush"))
{
assert(results.size()==1);
results[0]=typecast_exprt(arg0, java_int_type());
}
else if(statement==patternt("if_?cmp??"))
{
irep_idt number=to_constant_expr(arg0).get_value();
assert(op.size()==2 && results.empty());
code_ifthenelset code_branch;
const irep_idt cmp_op=get_if_cmp_operator(statement);
binary_relation_exprt condition(op[0], cmp_op, op[1]);
cast_if_necessary(condition);
code_branch.cond()=condition;
code_branch.then_case()=code_gotot(label(number));
code_branch.then_case().add_source_location()=i_it->source_location;
code_branch.add_source_location()=i_it->source_location;
c=code_branch;
}
else if(statement==patternt("if??"))
{
const irep_idt id=
statement=="ifeq"?ID_equal:
statement=="ifne"?ID_notequal:
statement=="iflt"?ID_lt:
statement=="ifge"?ID_ge:
statement=="ifgt"?ID_gt:
statement=="ifle"?ID_le:
(assert(false), "");
irep_idt number=to_constant_expr(arg0).get_value();
assert(op.size()==1 && results.empty());
code_ifthenelset code_branch;
code_branch.cond()=binary_relation_exprt(op[0], id, gen_zero(op[0].type()));
code_branch.cond().add_source_location()=i_it->source_location;
code_branch.then_case()=code_gotot(label(number));
code_branch.then_case().add_source_location()=i_it->source_location;
code_branch.add_source_location()=i_it->source_location;
c=code_branch;
}
else if(statement==patternt("ifnonnull"))
{
irep_idt number=to_constant_expr(arg0).get_value();
assert(op.size()==1 && results.empty());
code_ifthenelset code_branch;
const typecast_exprt lhs(op[0], pointer_typet());
const exprt rhs(gen_zero(lhs.type()));
code_branch.cond()=binary_relation_exprt(lhs, ID_notequal, rhs);
code_branch.then_case()=code_gotot(label(number));
code_branch.then_case().add_source_location()=i_it->source_location;
code_branch.add_source_location()=i_it->source_location;
c=code_branch;
}
else if(statement==patternt("ifnull"))
{
assert(op.size()==1 && results.empty());
irep_idt number=to_constant_expr(arg0).get_value();
code_ifthenelset code_branch;
const typecast_exprt lhs(op[0], pointer_typet(empty_typet()));
const exprt rhs(gen_zero(lhs.type()));
code_branch.cond()=binary_relation_exprt(lhs, ID_equal, rhs);
code_branch.then_case()=code_gotot(label(number));
code_branch.then_case().add_source_location()=i_it->source_location;
code_branch.add_source_location()=i_it->source_location;
c=code_branch;
}
else if(statement=="iinc")
{
code_assignt code_assign;
code_assign.lhs()=variable(arg0, 'i');
code_assign.rhs()=plus_exprt(
variable(arg0, 'i'),
typecast_exprt(arg1, java_int_type()));
c=code_assign;
}
else if(statement==patternt("?xor"))
{
assert(op.size()==2 && results.size()==1);
results[0]=bitxor_exprt(op[0], op[1]);
}
else if(statement==patternt("?or"))
{
assert(op.size()==2 && results.size()==1);
results[0]=bitor_exprt(op[0], op[1]);
}
else if(statement==patternt("?and"))
{
assert(op.size()==2 && results.size()==1);
results[0]=bitand_exprt(op[0], op[1]);
}
else if(statement==patternt("?shl"))
{
assert(op.size()==2 && results.size()==1);
results[0]=shl_exprt(op[0], op[1]);
}
else if(statement==patternt("?shr"))
{
assert(op.size()==2 && results.size()==1);
results[0]=ashr_exprt(op[0], op[1]);
}
else if(statement==patternt("?ushr"))
{
assert(op.size()==2 && results.size()==1);
const typet type(java_type_from_char(statement[0]));
const unsigned int width(type.get_unsigned_int(ID_width));
typet target=unsigned_long_int_type();
target.set(ID_width, width);
const typecast_exprt lhs(op[0], target);
const typecast_exprt rhs(op[1], target);
results[0]=lshr_exprt(lhs, rhs);
}
else if(statement==patternt("?add"))
{
assert(op.size()==2 && results.size()==1);
results[0]=plus_exprt(op[0], op[1]);
}
else if(statement==patternt("?sub"))
{
assert(op.size()==2 && results.size()==1);
results[0]=minus_exprt(op[0], op[1]);
}
else if(statement==patternt("?div"))
{
assert(op.size()==2 && results.size()==1);
results[0]=div_exprt(op[0], op[1]);
}
else if(statement==patternt("?mul"))
{
assert(op.size()==2 && results.size()==1);
results[0]=mult_exprt(op[0], op[1]);
}
else if(statement==patternt("?neg"))
{
assert(op.size()==1 && results.size()==1);
results[0]=unary_minus_exprt(op[0], op[0].type());
}
else if(statement==patternt("?rem"))
{
assert(op.size()==2 && results.size()==1);
if(statement=="frem" || statement=="drem")
results[0]=rem_exprt(op[0], op[1]);
else
results[0]=mod_exprt(op[0], op[1]);
}
else if(statement==patternt("?cmp"))
{
assert(op.size() == 2 && results.size() == 1);
// The integer result on the stack is:
// 0 if op[0] equals op[1]
// -1 if op[0] is less than op[1]
// 1 if op[0] is greater than op[1]
const typet t=java_int_type();
results[0]=
if_exprt(binary_relation_exprt(op[0], ID_equal, op[1]), gen_zero(t),
if_exprt(binary_relation_exprt(op[0], ID_gt, op[1]), from_integer(1, t),
from_integer(-1, t)));
}
else if(statement==patternt("?cmp?"))
{
assert(op.size()==2 && results.size()==1);
const floatbv_typet type(to_floatbv_type(java_type_from_char(statement[0])));
const ieee_float_spect spec(type);
const ieee_floatt nan(ieee_floatt::NaN(spec));
const constant_exprt nan_expr(nan.to_expr());
const int nan_value(statement[4] == 'l' ? -1 : 1);
const typet result_type(java_int_type());
const exprt nan_result(from_integer(nan_value, result_type));
// (value1 == NaN || value2 == NaN) ? nan_value : value1 < value2 ? -1 : value2 < value1 1 ? 1 : 0;
// (value1 == NaN || value2 == NaN) ? nan_value : value1 == value2 ? 0 : value1 < value2 -1 ? 1 : 0;
results[0]=
if_exprt(or_exprt(ieee_float_equal_exprt(nan_expr, op[0]), ieee_float_equal_exprt(nan_expr, op[1])), nan_result,
if_exprt(ieee_float_equal_exprt(op[0], op[1]), gen_zero(result_type),
if_exprt(binary_relation_exprt(op[0], ID_lt, op[1]), from_integer(-1, result_type), from_integer(1, result_type))));
}
else if(statement==patternt("?cmpl"))
{
assert(op.size()==2 && results.size()==1);
results[0]=binary_relation_exprt(op[0], ID_lt, op[1]);
}
else if(statement=="dup")
{
assert(op.size()==1 && results.size()==2);
results[0]=results[1]=op[0];
}
else if(statement=="dup_x1")
{
assert(op.size()==2 && results.size()==3);
results[0]=op[1];
results[1]=op[0];
results[2]=op[1];
}
else if(statement=="dup_x2")
{
assert(op.size()==3 && results.size()==4);
results[0]=op[2];
results[1]=op[0];
results[2]=op[1];
results[3]=op[2];
}
// dup2* behaviour depends on the size of the operands on the
// stack
else if(statement=="dup2")
{
assert(!stack.empty() && results.empty());
if(stack.back().type().get_unsigned_int(ID_width)==32)
op=pop(2);
else
op=pop(1);
results.insert(results.end(), op.begin(), op.end());
results.insert(results.end(), op.begin(), op.end());
}
else if(statement=="dup2_x1")
{
assert(!stack.empty() && results.empty());
if(stack.back().type().get_unsigned_int(ID_width)==32)
op=pop(3);
else
op=pop(2);
results.insert(results.end(), op.begin()+1, op.end());
results.insert(results.end(), op.begin(), op.end());
}
else if(statement=="dup2_x2")
{
assert(!stack.empty() && results.empty());
if(stack.back().type().get_unsigned_int(ID_width)==32)
op=pop(2);
else
op=pop(1);
assert(!stack.empty());
exprt::operandst op2;
if(stack.back().type().get_unsigned_int(ID_width)==32)
op2=pop(2);
else
op2=pop(1);
results.insert(results.end(), op.begin(), op.end());
results.insert(results.end(), op2.begin(), op2.end());
results.insert(results.end(), op.begin(), op.end());
}
else if(statement=="dconst")
{
assert(op.empty() && results.size()==1);
}
else if(statement=="fconst")
{
assert(op.empty() && results.size()==1);
}
else if(statement=="getfield")
{
assert(op.size()==1 && results.size()==1);
results[0]=to_member(op[0], arg0);
}
else if(statement=="getstatic")
{
assert(op.empty() && results.size()==1);
symbol_exprt symbol_expr(arg0.type());
symbol_expr.set_identifier(arg0.get_string(ID_class)+"."+arg0.get_string(ID_component_name));
results[0]=symbol_expr;
}
else if(statement=="putfield")
{
assert(op.size()==2 && results.size()==0);
c = code_assignt(to_member(op[0], arg0), op[1]);
}
else if(statement=="putstatic")
{
assert(op.size()==1 && results.empty());
symbol_exprt symbol_expr(arg0.type());
symbol_expr.set_identifier(arg0.get_string(ID_class)+"."+arg0.get_string(ID_component_name));
c=code_assignt(symbol_expr, op[0]);
}
else if(statement==patternt("?2?")) // i2c etc.
{
assert(op.size()==1 && results.size()==1);
results[0]=typecast_exprt(op[0], java_type_from_char(statement[2]));
}
else if(statement=="new")
{
// use temporary since the stack symbol might get duplicated
assert(op.empty() && results.size()==1);
const pointer_typet ref_type(arg0.type());
exprt java_new_expr=side_effect_exprt(ID_java_new, ref_type);
if(!i_it->source_location.get_line().empty())
java_new_expr.add_source_location()=i_it->source_location;
const exprt tmp=tmp_variable("new", ref_type);
c=code_assignt(tmp, java_new_expr);
results[0]=tmp;
}
else if(statement=="newarray" ||
statement=="anewarray")
{
// the op is the array size
assert(op.size()==1 && results.size()==1);
char element_type;
if(statement=="newarray")
{
irep_idt id=arg0.type().id();
if(id==ID_bool)
element_type='z';
else if(id==ID_char)
element_type='c';
else if(id==ID_float)
element_type='f';
else if(id==ID_double)
element_type='d';
else if(id==ID_byte)
element_type='b';
else if(id==ID_short)
element_type='s';
else if(id==ID_int)
element_type='i';
else if(id==ID_long)
element_type='j';
else
element_type='?';
}
else
element_type='a';
const pointer_typet ref_type=java_array_type(element_type);
side_effect_exprt java_new_array(ID_java_new_array, ref_type);
java_new_array.copy_to_operands(op[0]);
if(!i_it->source_location.get_line().empty())
java_new_array.add_source_location()=i_it->source_location;
const exprt tmp=tmp_variable("newarray", ref_type);
c=code_assignt(tmp, java_new_array);
results[0]=tmp;
}
else if(statement=="multianewarray")
{
// The first argument is the type, the second argument is the dimension.
// The size of each dimension is on the stack.
irep_idt number=to_constant_expr(arg1).get_value();
unsigned dimension=safe_c_str2unsigned(number.c_str());
op=pop(dimension);
assert(results.size()==1);
// arg0.type()
const pointer_typet ref_type=java_array_type('a');
side_effect_exprt java_new_array(ID_java_new_array, ref_type);
java_new_array.operands()=op;
if(!i_it->source_location.get_line().empty())
java_new_array.add_source_location()=i_it->source_location;
const exprt tmp=tmp_variable("newarray", ref_type);
c=code_assignt(tmp, java_new_array);
results[0]=tmp;
}
else if(statement=="arraylength")
{
assert(op.size()==1 && results.size()==1);
exprt pointer=
typecast_exprt(op[0], java_array_type(statement[0]));
const dereference_exprt array(pointer, pointer.type().subtype());
assert(pointer.type().subtype().id()==ID_symbol);
const member_exprt length(array, "length", java_int_type());
results[0]=length;
}
else if(statement=="tableswitch" ||
statement=="lookupswitch")
{
assert(op.size()==1 && results.size()==0);
// we turn into switch-case
code_switcht code_switch;
code_switch.add_source_location()=i_it->source_location;
code_switch.value()=op[0];
code_blockt code_block;
code_block.add_source_location()=i_it->source_location;
bool is_label=true;
for(instructiont::argst::const_iterator
a_it=i_it->args.begin();
a_it!=i_it->args.end();
a_it++, is_label=!is_label)
{
if(is_label)
{
code_switch_caset code_case;
code_case.add_source_location()=i_it->source_location;
irep_idt number=to_constant_expr(*a_it).get_value();
code_case.code()=code_gotot(label(number));
code_case.code().add_source_location()=i_it->source_location;
if(a_it==i_it->args.begin())
code_case.set_default();
else
{
instructiont::argst::const_iterator prev=a_it;
prev--;
code_case.case_op()=typecast_exprt(*prev, op[0].type());
code_case.case_op().add_source_location()=i_it->source_location;
}
code_block.add(code_case);
}
}
code_switch.body()=code_block;
c=code_switch;
}
else if(statement=="pop" || statement=="pop2")
{
// these are skips
c=code_skipt();
// pop2 removes two single-word items from the stack (e.g. two
// integers, or an integer and an object reference) or one
// two-word item (i.e. a double or a long).
// http://cs.au.dk/~mis/dOvs/jvmspec/ref-pop2.html
if(statement=="pop2" &&
op[0].type().get_unsigned_int(ID_width)==32)
pop(1);
}
else if(statement=="instanceof")
{
assert(op.size()==1 && results.size()==1);
results[0]=
binary_predicate_exprt(op[0], "java_instanceof", arg0);
}
else
{
c=codet(statement);
c.operands()=op;
}
if(!i_it->source_location.get_line().empty())
c.add_source_location()=i_it->source_location;
push(results);
a_it->second.done=true;
for(std::list<unsigned>::iterator
it=a_it->second.successors.begin();
it!=a_it->second.successors.end();
++it)
{
address_mapt::iterator a_it2=address_map.find(*it);
assert(a_it2!=address_map.end());
if(!stack.empty() && a_it2->second.predecessors.size()>1)
{
// copy into temporaries
code_blockt more_code;
// introduce temporaries when successor is seen for the first
// time
if(a_it2->second.stack.empty())
{
for(stackt::iterator s_it=stack.begin();
s_it!=stack.end();
++s_it)
{
symbol_exprt lhs=tmp_variable("$stack", s_it->type());
code_assignt a(lhs, *s_it);
more_code.copy_to_operands(a);
s_it->swap(lhs);
}
}
else
{
assert(a_it2->second.stack.size()==stack.size());
stackt::const_iterator os_it=a_it2->second.stack.begin();
for(stackt::iterator s_it=stack.begin();
s_it!=stack.end();
++s_it)
{
assert(has_prefix(os_it->get_string(ID_C_base_name),
"$stack"));
symbol_exprt lhs=to_symbol_expr(*os_it);
code_assignt a(lhs, *s_it);
more_code.copy_to_operands(a);
s_it->swap(lhs);
++os_it;
}
}
if(results.empty())
{
more_code.copy_to_operands(c);
c.swap(more_code);
}
else
{
c.make_block();
forall_operands(o_it, more_code)
c.copy_to_operands(*o_it);
}
}
a_it2->second.stack=stack;
}
}
// TODO: add exception handlers from exception table
// review successor computation of athrow!
code_blockt code;
// temporaries
for(const auto & var : tmp_vars)
{
code.add(code_declt(var));
}
for(const auto & it : address_map)
{
const unsigned address=it.first;
assert(it.first==it.second.source->address);
const codet &c=it.second.code;
if(targets.find(address)!=targets.end())
code.add(code_labelt(label(i2string(address)), c));
else if(c.get_statement()!=ID_skip)
code.add(c);
}
return code;
}
void goto_checkt::nan_check(
const exprt &expr,
const guardt &guard)
{
if(!enable_nan_check)
return;
// first, check type
if(expr.type().id()!=ID_floatbv)
return;
if(expr.id()!=ID_plus &&
expr.id()!=ID_mult &&
expr.id()!=ID_div &&
expr.id()!=ID_minus)
return;
// add NaN subgoal
exprt isnan;
if(expr.id()==ID_div)
{
assert(expr.operands().size()==2);
// there a two ways to get a new NaN on division:
// 0/0 = NaN and x/inf = NaN
// (note that x/0 = +-inf for x!=0 and x!=inf)
exprt zero_div_zero=and_exprt(
ieee_float_equal_exprt(expr.op0(), gen_zero(expr.op0().type())),
ieee_float_equal_exprt(expr.op1(), gen_zero(expr.op1().type())));
exprt div_inf=unary_exprt(ID_isinf, expr.op1(), bool_typet());
isnan=or_exprt(zero_div_zero, div_inf);
}
else if(expr.id()==ID_mult)
{
if(expr.operands().size()>=3)
return nan_check(make_binary(expr), guard);
assert(expr.operands().size()==2);
// Inf * 0 is NaN
exprt inf_times_zero=and_exprt(
unary_exprt(ID_isinf, expr.op0(), bool_typet()),
ieee_float_equal_exprt(expr.op1(), gen_zero(expr.op1().type())));
exprt zero_times_inf=and_exprt(
ieee_float_equal_exprt(expr.op1(), gen_zero(expr.op1().type())),
unary_exprt(ID_isinf, expr.op0(), bool_typet()));
isnan=or_exprt(inf_times_zero, zero_times_inf);
}
else if(expr.id()==ID_plus)
{
if(expr.operands().size()>=3)
return nan_check(make_binary(expr), guard);
assert(expr.operands().size()==2);
// -inf + +inf = NaN and +inf + -inf = NaN,
// i.e., signs differ
ieee_float_spect spec=ieee_float_spect(to_floatbv_type(expr.type()));
exprt plus_inf=ieee_floatt::plus_infinity(spec).to_expr();
exprt minus_inf=ieee_floatt::minus_infinity(spec).to_expr();
isnan=or_exprt(
and_exprt(equal_exprt(expr.op0(), minus_inf), equal_exprt(expr.op1(), plus_inf)),
and_exprt(equal_exprt(expr.op0(), plus_inf), equal_exprt(expr.op1(), minus_inf)));
}
else if(expr.id()==ID_minus)
{
assert(expr.operands().size()==2);
// +inf - +inf = NaN and -inf - -inf = NaN,
// i.e., signs match
ieee_float_spect spec=ieee_float_spect(to_floatbv_type(expr.type()));
exprt plus_inf=ieee_floatt::plus_infinity(spec).to_expr();
exprt minus_inf=ieee_floatt::minus_infinity(spec).to_expr();
isnan=or_exprt(
and_exprt(equal_exprt(expr.op0(), plus_inf), equal_exprt(expr.op1(), plus_inf)),
and_exprt(equal_exprt(expr.op0(), minus_inf), equal_exprt(expr.op1(), minus_inf)));
}
else
assert(false);
isnan.make_not();
add_guarded_claim(
isnan,
"NaN on "+expr.id_string(),
"NaN",
expr.find_source_location(),
expr,
guard);
}
bool ssa_may_alias(
const exprt &e1,
const exprt &e2,
const namespacet &ns)
{
#ifdef DEBUG
std::cout << "MAY ALIAS1 " << from_expr(ns, "", e1) << " "
<< from_expr(ns, "", e2) << "\n";
#endif
// The same?
if(e1==e2)
return true;
// Both symbol?
if(e1.id()==ID_symbol &&
e2.id()==ID_symbol)
{
return to_symbol_expr(e1).get_identifier()==
to_symbol_expr(e2).get_identifier();
}
// __CPROVER symbols
if(e1.id()==ID_symbol &&
has_prefix(
id2string(to_symbol_expr(e1).get_identifier()), CPROVER_PREFIX))
return false;
if(e2.id()==ID_symbol &&
has_prefix(
id2string(to_symbol_expr(e2).get_identifier()), CPROVER_PREFIX))
return false;
if(e1.id()==ID_symbol &&
has_suffix(
id2string(to_symbol_expr(e1).get_identifier()), "#return_value"))
return false;
if(e2.id()==ID_symbol &&
has_suffix(
id2string(to_symbol_expr(e2).get_identifier()), "#return_value"))
return false;
// Both member?
if(e1.id()==ID_member &&
e2.id()==ID_member)
{
const member_exprt &m1=to_member_expr(e1);
const member_exprt &m2=to_member_expr(e2);
// same component?
if(m1.get_component_name()!=m2.get_component_name())
return false;
return ssa_may_alias(m1.struct_op(), m2.struct_op(), ns);
}
// Both index?
if(e1.id()==ID_index &&
e2.id()==ID_index)
{
const index_exprt &i1=to_index_expr(e1);
const index_exprt &i2=to_index_expr(e2);
return ssa_may_alias(i1.array(), i2.array(), ns);
}
const typet &t1=ns.follow(e1.type());
const typet &t2=ns.follow(e2.type());
// If one is an array and the other not, consider the elements
if(t1.id()==ID_array && t2.id()!=ID_array)
if(ssa_may_alias(
index_exprt(e1, gen_zero(index_type()), t1.subtype()), e2, ns))
return true;
if(t2.id()==ID_array && t2.id()!=ID_array)
if(ssa_may_alias(
e1, index_exprt(e2, gen_zero(index_type()), t2.subtype()), ns))
return true;
// Pointers only alias with other pointers,
// which is a restriction.
if(t1.id()==ID_pointer)
return t2.id()==ID_pointer;
if(t2.id()==ID_pointer)
return t1.id()==ID_pointer;
// Is one a scalar pointer?
if(e1.id()==ID_dereference &&
(t1.id()==ID_signedbv || t1.id()==ID_unsignedbv || t1.id()==ID_floatbv))
return true;
if(e2.id()==ID_dereference &&
(t2.id()==ID_signedbv || t2.id()==ID_unsignedbv || t1.id()==ID_floatbv))
return true;
// Is one a pointer?
if(e1.id()==ID_dereference ||
e2.id()==ID_dereference)
{
// look at the types
// same type?
if(base_type_eq(t1, t2, ns))
{
return true;
}
// should consider further options, e.g., struct prefixes
return false;
}
return false; // both different objects
}
void string_instrumentationt::do_strerror(
goto_programt &dest,
goto_programt::targett it,
code_function_callt &call)
{
if(call.lhs().is_nil())
{
it->make_skip();
return;
}
irep_idt identifier_buf="c::__strerror_buffer";
irep_idt identifier_size="c::__strerror_buffer_size";
if(context.symbols.find(identifier_buf)==context.symbols.end())
{
symbolt new_symbol_size;
new_symbol_size.base_name="__strerror_buffer_size";
new_symbol_size.pretty_name=new_symbol_size.base_name;
new_symbol_size.name=identifier_size;
new_symbol_size.mode="C";
new_symbol_size.type=uint_type();
new_symbol_size.is_statevar=true;
new_symbol_size.lvalue=true;
new_symbol_size.static_lifetime=true;
array_typet type;
type.subtype()=char_type();
type.size()=symbol_expr(new_symbol_size);
symbolt new_symbol_buf;
new_symbol_buf.mode="C";
new_symbol_buf.type=type;
new_symbol_buf.is_statevar=true;
new_symbol_buf.lvalue=true;
new_symbol_buf.static_lifetime=true;
new_symbol_buf.base_name="__strerror_buffer";
new_symbol_buf.pretty_name=new_symbol_buf.base_name;
new_symbol_buf.name="c::"+id2string(new_symbol_buf.base_name);
context.move(new_symbol_buf);
context.move(new_symbol_size);
}
const symbolt &symbol_size=ns.lookup(identifier_size);
const symbolt &symbol_buf=ns.lookup(identifier_buf);
goto_programt tmp;
{
goto_programt::targett assignment1=tmp.add_instruction(ASSIGN);
exprt nondet_size=side_effect_expr_nondett(uint_type());
assignment1->code=code_assignt(symbol_expr(symbol_size), nondet_size);
assignment1->location=it->location;
goto_programt::targett assumption1=tmp.add_instruction();
assumption1->make_assumption(binary_relation_exprt(
symbol_expr(symbol_size), "notequal",
gen_zero(symbol_size.type)));
assumption1->location=it->location;
}
// return a pointer to some magic buffer
exprt index=exprt("index", char_type());
index.copy_to_operands(symbol_expr(symbol_buf), gen_zero(uint_type()));
exprt ptr=exprt("address_of", pointer_typet());
ptr.type().subtype()=char_type();
ptr.copy_to_operands(index);
// make that zero-terminated
{
goto_programt::targett assignment2=tmp.add_instruction(ASSIGN);
assignment2->code=code_assignt(is_zero_string(ptr, true), true_exprt());
assignment2->location=it->location;
}
// assign address
{
goto_programt::targett assignment3=tmp.add_instruction(ASSIGN);
exprt rhs=ptr;
make_type(rhs, call.lhs().type());
assignment3->code=code_assignt(call.lhs(), rhs);
assignment3->location=it->location;
}
it->make_skip();
dest.insert_before_swap(it, tmp);
}
