Errata for Advanced CORBA Programming with C++ (2nd Printing)

Last updated: 28 January 2004

Acknowledgments

A number of people have submitted comments, constructive criticism, and feedback since the first printing. Our thanks go to everyone who helped us weed out bugs and make this a better book. All the people who contributed are listed in the acknowledgments to our readers.

Example Code

Use of `assert` Macro

The code examples in the book in many places use the assert macro as an extremely crude form of error handling. For example, on page 396:

assert(ICP_get(m_anum, "model", buf, sizeof(buf)) == 0);

We did this mainly to keep code examples short and to avoid distracting from the main message with lots of error-handling code. However, using assert in this way means that if you compile the code with NDEBUG defined, it will no longer work correctly because the assertion has a side effect that no longer occurs when the assertion is disabled.

Quite a few of the code examples in the book rely on assertions not being disabled. A number of people have pointed out to us that using assertions in this way is bad style, and we agree. For your own code, you should ensure that assertions do not have side effects.

Use of `delete[]` with `ostrstream`

A number of examples show code along the following lines:

ostrstream ostr;
ostr << "whatever..." << ends;
char * text_p = ostr.str();
// ...
delete[] text_p;

The str member function passes responsibility for deallocation of the string to the caller. However, the use of delete[] to deallocate that string is not portable, nor is use of delete (although either or both work fine in many implementations). Instead, explicit deallocation of the string requires a call to ::operator delete(text_p).

However, another option is to use freeze(0) to return responsibility for deallocation to the ostrstream, so memory for the string returned by str is deallocated when the ostrstream goes out of scope. Therefore, the above code fragment should read:

ostrstream ostr;
ostr << "whatever..." << ends;
char * text_p = ostr.str();
// ...
ostr.rdbuf()->freeze(0);

Chapter 3

Pages 44 and 45
The code to check the number of arguments should follow the call to ORB_init instead of preceding it. This is necessary if additional ORB-specific options are passed on the commmand line.

Chapter 4

Page 74

The members of type Adata and Bdata of the ABunion are swapped. The union definition should read:

union ABunion switch (StructType) {
case A_TYPE:
    struct Acontents {
        Adata                   data;
        sequence<ABunion, 1>    nested; // Contained Bdata
    } A_member;
case B_TYPE:
    struct Bcontents {
        Bdata                   data;
        sequence<ABunion, 1>    nested; // Contained Adata
    } B_member;
};

Pages 117 and 118
The second repository identifier in the list near the top of the pages is shown as IDL:CCS/Tempttype:1.0. It should be IDL:CCS/TempType:1.0.

Chapter 6

Page 170

The prefix increment and decrement operators of class Fixed have the wrong signature. The correct signature is:

Fixed & operator++();
Fixed & operator--();

The return type of comparison operators was changed in the C++ mapping after the publication date of this printing (from int to Boolean). The new signatures are:

Boolean operator<(const Fixed &, const Fixed &);
Boolean operator>(const Fixed &, const Fixed &);
Boolean operator<=(const Fixed &, const Fixed &);
Boolean operator>=(const Fixed &, const Fixed &);
Boolean operator==(const Fixed &, const Fixed &);
Boolean operator!=(const Fixed &, const Fixed &);

Page 173
The fifth line of the code example should read:
```
r = f1.round(1);       // 0.4
```
Page 190
The code to fill a string sequence buffer contains a memory management bug:
```
if (len == max) {
    // Double size if out of room
    char ** tmp = StrSeq::allocbuf(max *= 2);
    memcpy(tmp, strvec, len * sizeof(*strvec));
    StrSeq::freebuf(strvec);
    strvec = tmp;
}
```
This is incorrect because freebuf deallocates each string in the buffer. This results in eventual double deallocation of the sequence elements.
To avoid the problem, the code must copy each string, deallocating the empty strings that are placed into the buffer by allocbuf:
```
if (len == max) {
    // Double size if out of room
    char ** tmp = StrSeq::allocbuf(max *= 2);
    for (CORBA::ULong i = 0; i < len; i++) {
        CORBA::string_free(tmp[i]);
        tmp[i] = CORBA::string_dup(strvec[i]);
    }
    StrSeq::freebuf(strvec);
    strvec = tmp;
}
```
The memory management rules for allocbuf with string sequences were changed with the final revision of the C++ mapping for CORBA 2.3. This makes it highly likely that your ORB will deviate from the behavior currently prescribed by the standard. In addition, there are still ongoing discussions as to the correct behavior of the sequence allocation and deallocation functions for sequences of strings and sequences of object references.
To keep your code portable, we recommend that you do not use the buffer manipulation functions for sequences of strings and object references until this matter is resolved in the OMG.
Page 194
The forward declaration for DoubleSeq contains a typo (calss instead of class).
Page 200
The call to StrArray_copy on the third-last line of the code example passes the address of the array instead of the array itself. The call should read:
```
StrArray_copy(x, sp1);  // Copy contents of sp1 into x
```

Page 207

The class definition for ShippingInfo disagrees with its IDL definition. The correct class definition is:

class ShippingInfo {
public:
                    // Other member functions as before...

    const Details & packaging_info() const;            // Accessor
    void            packaging_info(const Details &);   // Modifier
    Details &       packaging_info();                  // Referent

    const TextSeq & other_info() const;                // Accessor
    void            other_info(const TextSeq &);       // Modifier
    TextSeq &       other_info();                      // Referent
};

Pages 212 and 214
The return type of operator[] is incorrect. The operators return a reference to the element type of the sequence (not the sequence type itself). Assuming an element type TE, the correct declarations are:
```
      TE & operator[](CORBA::ULong);       // For sequences
const TE & operator[](CORBA::ULong) const; // For sequences
```
Page 217
The first sentence of the section Additional T_var Member Functions for Fixed-Length Types should read:
For a T_var for a fixed-length structure or union of type T, the IDL compiler generates the following:
Page 218
The return type T & of the T_var assignment operator is incorrect. The correct signature is:
```
T_var & operator=(const T &)
```

Chapter 7

Pages 234 and 235

The definitions for _var_type and _ptr_type appear at the wrong scope. They are incorrectly shown as part of the CORBA and CCS namespaces, but instead should appear as nested definitions in each proxy class.

In addition, the signature for the _narrow member function is incorrect in both the Thermometer and Thermostat classes: the parameter type of these should be CORBA::Object_ptr.

The correct class definitions for pages 234 and 235 are:

namespace CORBA {
	class Object;
	class Object_var;
	typedef Object * Object_ptr;
	class Object {
	public:
		static Object_ptr _duplicate(Object_ptr p);
		static Object_ptr _nil();
		static Object_ptr _narrow(Object_ptr p);
		// Other member functions here...
		typedef Object_var _var_type;
		typedef Object_ptr _ptr_type;
	};
	Boolean is_nil(Object_ptr p);
	// ...
}

namespace CCS {
	// ...
	class Thermometer;
	class Thermometer_var;
	typedef Thermometer * Thermometer_ptr;
	class Thermometer : public virtual CORBA::Object {
	public:
		static Thermometer_ptr _duplicate(Thermometer_ptr p);
		static Thermometer_ptr _nil();
		static Thermometer_ptr _narrow(CORBA::Object_ptr p);
		// Member functions for attributes of Thermometer here...
		typedef Thermometer_var _var_type;
		typedef Thermometer_ptr _ptr_type;
	};

	class Thermostat;
	class Thermostat_var;
	typedef Thermostat * Thermostat_ptr;
	class Thermostat : public virtual Thermometer {
	public:
		static Thermostat_ptr _duplicate(Thermostat_ptr p);
		static Thermostat_ptr _nil();
		static Thermostat_ptr _narrow(CORBA::Object_ptr p);
		// Member functions for operations of Thermostat here...
		typedef Thermostat_var _var_type;
		typedef Thermostat_ptr _ptr_type;
	};
	// ...
}

Pages 240 and 241

The code example to illustrate incorrect use of down-casts uses incorrect types. The example should read:

CCS::Thermostat_ptr tmstat = ...;   // Get reference
CORBA::Object_ptr o = tmstat;       // OK
CCS::Thermostat_ptr tmstat2;

tmstat2 = dynamic_cast<CCS::Thermostat_ptr>(o);     // Bad!
tmstat2 = static_cast<CCS::Thermostat_ptr>(o);      // Bad!
tmstat2 = reinterpret_cast<CCS::Thermostat_ptr>(o); // Bad!
tmstat2 = (CCS::Thermostat_ptr)o;                   // Bad!

tmstat2 = CCS::Thermostat::_narrow(o);              // OK

Page 262

The second code example has the variable p in the wrong scope. The example should read:

CCS::Thermometer_ptr p = ...;   // Get reference
{
    CCS::Thermometer_var v;
    v = p;                      // v takes ownership
    // Use v...
} // No leak here - v's destructor calls
  // release and p now dangles.

Page 269
The last sentence of Section 7.12.4 should read:
On the other hand, the _ptr reference tp now dangles, because ownership passed to seq[1] during the assignment.
Page 276
The operation name in the IDL example at the bottom of the page should be long_op instead of op_long.
Page 277
The final statement of the second code example uses an incorrect variable (ret_val instead of out_val). The final statement should read:
```
cout << "out_val: " << out_val << endl;
```

Page 281

The second code example passes the wrong variable to get_darr4 (d3 instead of in_val). The code example should read:

Foo_var fv = ...;   // Get reference

Darr3 in_val = { 1, 2, 3 };
Darr3_var ret_val;

ret_val = fv->get_darr4(in_val);    // Double disaster!!!

Page 290
The second bullet point mentions the wrong structure member (s instead of s_mem). The third sentence of the second bullet point should read:
If the callee wants to modify the string member s_mem of the structure, it can do so simply by assignment.
Page 292
The type definition for Varr_slice should read:
```
typedef Vls Varr_slice;
```

Page 294

The IDL at the top of the page should read:

interface Foo {
    Foo ref_op(
            in Foo    ref_in,
            inout Foo ref_inout,
            out Foo   ref_out
    );
};

Pages 297-299
The example for using _var types to hide differences in parameter rules is wrong. (It uses inout parameters instead of out parameters.) We have provided PDF with change bars for the affected pages.
The C++ mapping for _var types is under-specified in CORBA 2.3 because it does not show how _var types for fixed-length underlying types should work. On 10 January 2000, the OMG C++ Mapping RTF voted to clarify the situation by specifying the mapping for _var types for fixed-length underlying types. (Note that the implementation shown is not binding in the sense that its internals may differ in each ORB; however, the semantics must be the same.) The upshot of this mapping is that, once initialized (which is always via a deep copy), a _var variable for a fixed-length underlying type behaves like a value of the underlying type.
Pages 308 and 309
The BadTemp exception is used with the wrong scoped name. The correct scope is CCS::Thermostat::BadTemp.

Page 312

The example code tries to down-cast a system exception to a BadTemp exception, which cannot possibly work. The correct code is:

try {
    tmstat_ref->set_nominal(500);
} catch (const CORBA::Exception & e) {
    // Check what sort of exception it is...
    const CORBA::SystemException * se;
    const CCS::Thermostat::BadTemp * bt;
    if (se = CORBA::OBJECT_NOT_EXIST::_downcast(&e)) {
        // It is an OBJECT_NOT_EXIST exception
    } else if (bt = CCS::Thermostat::BadTemp::_downcast(&e)) {
        // It is a BadTemp exception
    } else {
        // It is some other exception
    }
    // Do not deallocate se here -- the exception
    // still owns the memory pointed to by se or bt.
}

Page 313

The example code tries to down-cast a system exception to a BadTemp exception, which cannot possibly work. The correct code is:

try {
    tmstat_ref->set_nominal(500);
} catch (const CORBA::Exception & e) {
    // Check what sort of exception it is...
    const CORBA::SystemException * se;
    const CCS::Thermostat::BadTemp * bt;
    if (se = dynamic_cast<const CORBA::OBJECT_NOT_EXIST *>(&e)) {
        // It is an OBJECT_NOT_EXIST exception
    } else if (bt =
            dynamic_cast<const CCS::Thermostat::BadTemp *>(&e)) {
        // It is a BadTemp exception
    } else {
        // It is some other exception
    }
}

Chapter 9

Page 362
The first definition in the code example should read:
```
const int array_length = sizeof(Darr)/sizeof(*Darr);
```
Page 371
The first definition in the second code example should read:
```
const int array_length = sizeof(Varr)/sizeof(*Varr);
```

Page 376

The declaration of the set_nominal function uses the wrong scope for the BadTemp exception. The correct declaration is:

virtual CCS::TempType set_nominal(CCS::TempType new_tmpe)
    throw(CORBA::SystemException, CCS::Thermostat::BadTemp) = 0;

Page 377
The BadTemp exception is used with the wrong scoped name in three places. The correct scope is CCS::Thermostat::BadTemp.
Page 380/381
The code example deletes the previously allocated vls_out pointer and then throws an exception without changing the pointer value. This is a grey area in the C++ mapping: it is not clear whether the skeleton will attempt to delete the value of an out parameter when an exception is thrown. (Currently, we are aware of skeleton implementions that call delete on the parameter when an exception is thrown and others that do not.)
To make the code portable for either kind of skeleton implementation, it is necessary to set the vls_out argument to zero before throwing the exception:
```
catch (const CORBA::Exception &) {
    delete vls_out.ptr();
    vls_out = 0;		// Required for portability
    delete result;
    throw;
}
```
The same situation arises for inout parameters, which also need to be set to zero before throwing an exception.
Note that there is no need to set the return value to zero for either kind of skeleton implementation.
Instead of writing code such as the example on page 380/381, we recommend that you use a _var type to hold an allocated out or inout parameter and call its _retn member function once no more exceptions can be thrown, as shown on page 381/382. This is both simpler and correct for either kind of skeleton implementation.
Page 385
The type name on the second last line is incorrect. It should be MyLegacyClass instead of MyLegacyType.

Chapter 10

Page 393
The code example at the top of page 393 uses the wrong identifier in the printf statement (nominal_temp instead of temp). The statement should read:
```
printf("nominal_temp: %d\n", temp);
```
Page 395
The definition of the m_anum protected member misses a const qualifier. It should read:
```
const CCS:AssetType m_anum; // My asset number
```

Page 407

The StrFinder function contains a memory leak. The two calls to strcmp should read:

return strcmp(CORBA::String_var(p.second->location()), m_str) == 0;

and

return strcmp(CORBA::String_var(p.second->model()), m_str) == 0;

The same error appears on page 417.

Page 412

The last statement of the code example passes an incorrect parameter to the find algorithm:

where = find_if(
            ++where, m_assets.end(),
            StrFinder(sc, slist[i].key.loc())
        );

The statement should read:

where = find_if(
            ++where, m_assets.end(),
            StrFinder(sc, search_str)
        );

Chapter 11

Page 447
The IDL definitions for the AdapterAlreadyExists and InvalidPolicy exceptions should be in the scope of the POA interface.
Page 463
The IDL definitions for the ServantAlreadyActive, ObjectAlreadyActive, and WrongPolicy exceptions should be in the scope of the POA interface.
Page 467
The call to activate_object incorrectly passes the servant instead of the address of the servant. The call should read:
poa->activate_object(&ctrl_servant);
Page 496
The last bullet point on the page finishes with the parenthetical remark
(the servant manager must be a ServantLocator)

The parenthetical remark is incorrect and should be deleted.
Page 505
The transition labelled F should be deleted from the diagram and the transition table.
Page 506
The description in the second paragraph talks about POAs, when it should be talking about POAManagers instead. Destruction of a POAManager is automatic and occurs when the last POA using the POAManager is destroyed.
Page 519
The code example makes a number of calls to find_POA and then activates the POA manager. However, the POA manager is created in the holding state, so the calls to find_POA will not be dispatched and the code deadlocks.
To fix the problem, the statements
```
// Activate our POAManager.
PortableServer::POAManager_var mgr = root_poa->the_POAManager();
mgr->activate();
```
must be moved to precede the calls to find_POA.

Chapter 12

Pages 582 and 583
The call to push_front(oid) at the end of the incarnate function should precede the instantiation of the new servant.

Chapter 13

Page 611
The third paragraph following the IDL example states:
Note that the size of the structure CD in Figure 13.2 is 13 bytes...
This should read:
Note that the size of the structure CD in Figure 13.2 is 14 bytes...
Page 625
The last line states that the components field is used only by IIOP 1.1. This is incorrect; the components field is used by both IIOP 1.1 and IIOP 1.2.
Page 628
The first sentence states that each profile contains information for one protocol. This is only partially correct because a single profile, if it is a multicomponent profile, can contain information for more than one protocol.
Using a multicomponent profile to hold information for more than one protocol is useful if the object key is the same for both protocols. In that case, using a single multicomponent profile avoids having to write two copies of the same object key into an IOR.

Chapter 16

Page 703
The member_name accessor is missing a parameter declaration. It should read:
```
const char * member_name(ULong index) const;
```
Page 713
The code dealing with wide characters incorrectly extracts the wide character from an Any value (because it omits to use the to_wchar helper). The code should read:
```
case CORBA::tk_wchar:
    CORBA::WChar wc;
    *ap >>= CORBA::Any::to_wchar(wc);
    cout << "'" << wc << "'";
    break;
```

Page 720

The final three statements of the code example incorrectly use >> instead of >>= for extraction. They should read:

foo_any >>= foo_p;      // Succeeds
aof_any >>= foo_p;      // Succeeds in 2.3, undefined in 2.2
bar_any >>= foo_p;      // Fails in 2.3, undefined in 2.2

Page 722
The list of type code constants misses an entry for _tc_long:
```
const CORBA::TypeCode_ptr _tc_long = ...;
```

Page 728

The signatures for create_exception_tc, create_interface_tc, and create_string_tc have an extraneous underscore. The correct signatures are:

TypeCode create_exception_tc(
                in RepositoryId         id,
                in Identifier           name,
                in StructMemberSeq      members
         );

TypeCode create_interface_tc(
                in RepositoryId         id,
                in Identifier           name
         );

TypeCode create_string_tc(
                in unsigned long        bound
         );

Page 730
The type code for the Node structure is correct, but TypeCode::equal does not return true when comparing this type code to the _tc_Node type code that is generated by the IDL compiler. (Only TypeCode::equivalent returns true for the comparison.)
To get TypeCode::equal to return true when comparing the struct_tc in the example with the generated _tc_Node type code, the second-last statement of the example should read:
```
members[1].type = orb->create_sequence_tc(0, rec_tc);
```

Chapter 17

The DynAny specification was changed substantially after the second printing was finalized, making it necessary to rewrite the chapter. You can download a replacement chapter in PDF format.

Chapter 18

Page 785
The first two sentences of the second paragraph should read:
resolve_initial_references cannot return a nil reference, unless someone has misconfigured the ORB.

Chapter 19

Page 838
The final paragraph says that
The Address property has the type code Manufacturing::_tc_AddressType.
This should read
The Manufacturer property has the type code Manufacturing::_tc_AddressType.

Page 852

The code to create a service type for the controller uses property modes that disagree with the ones shown in Table 19.1 on page 838 for the Supports and MaxDevices properties. The code for these two properties should read:

props[3].name = CORBA::string_dup("Supports");
props[3].value_type = CORBA::TypeCode::_duplicate(
                        Airconditioning::_tc_ModelType
                      );
props[3].mode = ServiceTypeRepository::PROP_READONLY;

props[4].name = CORBA::string_dup("MaxDevices");
props[4].value_type = CORBA::TypeCode::_duplicate(
                        CORBA::_tc_ulong
                      );
props[4].mode = ServiceTypeRepository::PROP_NORMAL;

Pages 864, 866, 897, and 919
The code examples call the export method on the proxy for the Register interface to export service offers. ISO/IEC C++ has added export as a keyword to C++. This means that, for ISO/IEC C++ compilers, the code will not compile as shown. Instead of export, the code should call _cxx_export. (Note that this will work only if your IDL compiler correctly maps away from the ISO/IEC C++ keyword.)
Page 874
The IDL on page 874 shows the type definition for Preference in the CosTrading scope. This is incorrect. The definition instead is part of the Lookup interface, so the correct fully-qualified name is CosTrading::Lookup::Preference.
Page 885
The expression near the bottom of the page is missing a * operator and should read:
min(12.3 * mem_size + 4.6 * file_size)
Page 902
The final line of the first bullet point mentions the limiting_follow_policy. This should read limiting_follow_rule instead.

Chapter 20

Page 947
The first code example uses the wrong member name for the assignment to the name component for the event channel (name instead of id). The beginning of the example should read:
```
CosNaming::Name ec_name;
ec_name.length(1);
ec_name[0].id = CORBA::string_dup("event_channel");
```

Page 956

The code example near the bottom of the page omits to connect the consumer to the event channel, which results in a Disconnected exception when the consumer calls pull. The correct code is:

// Assume the "channel" variable refers to our event channel.
CosEventChannelAdmin::ConsumerAdmin_var consumer_admin =
    channel->for_consumers();

// Obtain a ProxyPullSupplier from the ConsumerAdmin and connect.
CosEventChannelAdmin::ProxyPullSupplier_var supplier =
    consumer_admin->obtain_pull_supplier();
supplier->connect_pull_consumer(
    CosEventComm::PullConsumer::_nil()
);

bool done;                  // Now enter our GUI event loop.
do {
    check_for_thermostat_event(supplier);
    done = check_for_gui_event();
} while (!done);

Chapter 21

Page 975
The second paragraph states that even for requests made locally, they must be dispatched on the POA's single thread instead of the caller's thread.
Different ORB vendors have interpreted the wording in the specification differently, so the semantics of the SINGLE_THREAD_MODEL policy with respect to collocated calls are currently under discussion. Until this matter is settled in the OMG Core Revision Task Force, you must assume that it is implementation-dependent whether the call is dispatched on the POA's single thread or the caller's thread.
Page 993
The implementation of preinvoke should include a call to servant->_add_ref() before returning, and postinvoke should include a corresponding call to servant->_remove_ref(). These calls are necessary to ensure that the servant remains in existence for the duration of the operation.

Appendix A

Page 1026

The implementation of ICP_set does not always correctly NUL-terminate the location string. The correct implementation is:

extern "C"
int
ICP_set(unsigned long id, const char * attr, const void * value)
{
    // Look for id in state map
    StateMap::iterator pos = dstate.find(id);
    if (pos == dstate.end())
        return -1;                          // No such device

    // Change either location or nominal temp, depending on attr.
    if (strcmp(attr, "location") == 0) {
        pos->second.location = (const char *)value;
    if (pos->second.location.size() >= MAXSTR)
        pos->second.location[MAXSTR - 1] = '\0';
    } else if (strcmp(attr, "nominal_temp") == 0) {
        if (pos->second.type != thermostat) 
            return -1;                      // Must be thermostat
        short temp;
        memcpy(&temp, value, sizeof(temp));
        if (temp < MIN_TEMP || temp > MAX_TEMP)
            return -1;
        pos->second.nominal_temp = temp;
    } else {
        return -1;                          // No such attribute
    }
    return 0;                               // OK
}

Page 1028
The implementation of ~ICP_Persist does not always correctly NUL-terminate the location string. A call to c_str is missing in the loop that writes the device attributes:
```
db << i->second.location.c_str() << endl;
```