The Complete Guide to C++ Strings, Part II - String Wrapper Classes

Introduction

Since C-style strings can be error-prone and difficult to manage, not to mention a target for hackers looking for buffer overrun bugs, there are lots of string wrapper classes. Unfortunately, it's not always clear which class should be used in some situations, nor how to convert from a C-style string to a wrapper class.

This article covers all the string types in the Win32 API, MFC, STL, WTL, and the Visual C++ runtime library. I will describe the usage of each class, how to construct objects, and how to convert to other classes. Nish has also contributed the section on managed strings and classes in Visual C++ 7.

In order to get the full benefit from this article, you must understand the different character types and encodings, as I covered in Part I.

Rule #1 of string classes

Casts are bad, unless they are explicitly documented.

What prompted me to write these two articles was the frequent questions about how to convert string type X to type Z, where the poster was using a cast and didn't understand why the code didn't work. The various string types, especially BSTR, are not concisely documented in any one place, so I imagine some people were throwing in casts and hoping it would work.

A cast does not do any conversion to a string, unless the source string is a wrapper class with an explicitly documented conversion operator. A cast of a string literal does nothing to the string, so writing something like:

void SomeFunc ( LPCWSTR widestr );
 
main()
{
  SomeFunc ( (LPCWSTR) "C:\\foo.txt" );  // WRONG!

}

will fail 100% of the time. It will compile, because the cast overrides the compiler's type-checking. But just because it compiles, doesn't mean the code is correct.

In the examples that follow, I will point out when casts are legal.

C-style strings and typedefs

As I covered in Part I, Windows APIs are defined and documented in terms of TCHARs, which can be MBCS or Unicode characters depending on whether you define the _MBCS or _UNICODE symbol when compiling. You should refer to Part I for a full description of TCHAR, but I will list the character typedefs here for convenience.

Type	Meaning
WCHAR	Unicode character (wchar_t)
TCHAR	MBCS or Unicode character, depending on preprocessor settings
LPSTR	string of char (char*)
LPCSTR	constant string of char (const char*)
LPWSTR	string of WCHAR (WCHAR*)
LPCWSTR	constant string of WCHAR (const WCHAR*)
LPTSTR	string of TCHAR (TCHAR*)
LPCTSTR	constant string of TCHAR (const TCHAR*)

One additional character type is the OLECHAR. This represents the character type used in Automation interfaces (such as the interfaces exposed by Word so you can manipulate documents). This type is normally defined as wchar_t, however if you define the OLE2ANSI preprocessor symbol, OLECHAR will be defined as the char type. I know of no reason these days to define OLE2ANSI (it hasn't been used by Microsoft since the days of MFC 3), so from now on I will treat an OLECHAR as a Unicode character.

Here are the OLECHAR-related typedefs you will see:

Type	Meaning
OLECHAR	Unicode character (wchar_t)
LPOLESTR	string of OLECHAR (OLECHAR*)
LPCOLESTR	constant string of OLECHAR (const OLECHAR*)

There are also two macros used around string and character literals so that the same code can be used for both MBCS and Unicode builds:

Type	Meaning
_T(x)	Prepends L to the literal in Unicode builds.
OLESTR(x)	Prepends L to the literal to make it an LPCOLESTR.

There are also variants on _T that you might encounter in documentation or sample code. There are four equivalent macros -- TEXT, _TEXT, __TEXT, and __T -- that all do the same thing.

Strings in COM - BSTR and VARIANT

Many Automation and other COM interfaces use BSTR for strings, and BSTRs have a few pitfalls, so I will give BSTR its own section here.

BSTR is a hybrid between Pascal-style strings (where the length is stored explicitly along with the data) and C-style strings (where the string length must be calculated by looking for a terminating zero character). A BSTR is a Unicode string that has its length prepended, and is also terminated by a zero character. Here is an example of "Bob" as a BSTR:

06 00 00 00	42 00	6F 00	62 00	00 00
--length--	B	o	b	EOS

Notice how the length of the string is prepended to the string data. It is a DWORD, and holds the number of bytes in the string, not counting the terminating zero. In this case, "Bob" contains 3 Unicode characters (not counting the terminating zero), for a total of 6 bytes. The length field is present so that when a BSTR is marshaled between processes or computers, the COM library knows how much data to transfer. (As a side note, a BSTR can hold any arbitrary block of data, not just characters, and can contain embedded zero characters. However, for the purposes of this article, I will not consider such cases.)

A BSTR variable in C++ is actually a pointer to the first character of the string. In fact, the type BSTR is defined this way:

  typedef OLECHAR* BSTR;

This is very unfortunate, because in reality a BSTR is not the same as a Unicode string. That typedef defeats type-checking and allows you to freely mix LPOLESTRs and BSTRs. Passing a BSTR to a function expecting a LPCOLESTR (or LPCWSTR) is safe, however the reverse is not. Therefore, it's important to be aware of the exact type of string that a function expects, and pass the correct type of string.

To see why it is not safe to pass a LPCWSTR to a function expecting a BSTR, remember that the four bytes immediately before the string must store its length. There is no such length with a LPCWSTR. If the BSTR needs to be marshaled to another process (for example, an instance of Word that you are controlling), the COM library will look for that length and find garbage, or some other variable on your stack, or other random data. This will either cause the method to fail, or even crash if the perceived length is too long.

There are several APIs that operate on BSTRs, however the two most important ones are the functions that create and destroy a BSTR. They are SysAllocString() and SysFreeString(). SysAllocString() copies a Unicode string into a BSTR, while SysFreeString() frees the memory used by a BSTR.

BSTR bstr = NULL;
 
  bstr = SysAllocString ( L"Hi Bob!" );
 
  if ( NULL == bstr )
    // out of memory error

 
  // Use bstr here...

 
  SysFreeString ( bstr );

Naturally, the various BSTR wrapper classes take care of the memory management for you.

The other type used in Automation interfaces is VARIANT. This is used to send data between typeless languages like JScript and VBScript, as well as Visual Basic in some cases. A VARIANT can contain data of many different types, such as long and IDispatch*. When a VARIANT contains a string, it is stored as a BSTR. I will have more to say about VARIANTs when I cover the VARIANT wrapper classes later.

String wrapper classes

Now that I've covered the various types of strings, I'll demonstrate the wrapper classes. For each one, I'll show how to construct an object and how to convert it to a C-style string pointer. The C-style pointer is often necessary for an API call, or to construct an object of a different string class. I will not cover other operators the classes provide, such as sorting or comparison.

Once again, do not blindly cast objects unless you understand exactly what the resulting code will do.

Classes provided by the CRT

_bstr_t

_bstr_t is a complete wrapper around a BSTR, and in fact it hides the underlying BSTR. It provides various constructors, as well as operators to access the underlying C-style string. However, there is no operator to access the BSTR itself, so a _bstr_t cannot be passed as an [out] parameter to COM methods. If you need a BSTR* to use as a parameter, it is easier to the ATL class CComBSTR.

A _bstr_t can be passed to a function that takes a BSTR, but only because of three coincidences. First, _bstr_t has a conversion function to wchar_t*; second, wchar_t* and BSTR appear the same to the compiler because of the definition of BSTR; and third, the wchar_t* that a _bstr_t keeps internally points to a block of memory that follows the BSTR format. So even though there is no documented conversion to BSTR, it happens to work.

// Constructing

_bstr_t bs1 = "char string";       // construct from a LPCSTR

_bstr_t bs2 = L"wide char string"; // construct from a LPCWSTR

_bstr_t bs3 = bs1;                 // copy from another _bstr_t

_variant_t v = "Bob";
_bstr_t bs4 = v;                   // construct from a _variant_t that has a string

 
// Extracting data

LPCSTR psz1 = bs1;              // automatically converts to MBCS string

LPCSTR psz2 = (LPCSTR) bs1;     // cast OK, same as previous line

LPCWSTR pwsz1 = bs1;            // returns the internal Unicode string

LPCWSTR pwsz2 = (LPCWSTR) bs1;  // cast OK, same as previous line

BSTR    bstr = bs1.copy();      // copies bs1, returns it as a BSTR

 
  // ...

  SysFreeString ( bstr );

Note that _bstr_t also has conversion operators for char* and wchar_t*. This is a questionable design, because even though those are non-constant string pointers, you must not use those pointers to modify the buffer, because that could break the internal BSTR structure.

_variant_t

_variant_t is a complete wrapper around a VARIANT, and provides many constructors and conversion functions to operate on the multitude of types that a VARIANT can contain. I will only cover the string-related operations here.

// Constructing

_variant_t v1 = "char string";       // construct from a LPCSTR

_variant_t v2 = L"wide char string"; // construct from a LPCWSTR

_bstr_t bs1 = "Bob";
_variant_t v3 = bs1;                 // copy from a _bstr_t object

 
// Extracting data

_bstr_t bs2 = v1;           // extract BSTR from the VARIANT

_bstr_t bs3 = (_bstr_t) v1; // cast OK, same as previous line

Note that the _variant_t methods can throw exceptions if the type conversion cannot be made, so be prepared to catch _com_error exceptions.

Also note that there is no direct conversion from _variant_t to an MBCS string. You will need to make an interim _bstr_t variable, use another string class that provides the Unicode to MBCS conversion, or use an ATL conversion macro.

Unlike _bstr_t, a _variant_t can be passed directly as a parameter to a COM method. _variant_t derives from the VARIANT type, so passing a _variant_t in place of a VARIANT is allowed by C++ language rules.

STL classes

STL just has one string class, basic_string. A basic_string manages a zero-terminated array of characters. The character type is given in the basic_string template parameter. In general, a basic_string should be treated as an opaque object. You can get a read-only pointer to the internal buffer, but any write operations must use basic_string operators and methods.

There are two predefined specializations for basic_string: string, which contains chars, and wstring, which contains wchar_ts. There is no built-in TCHAR specialization, but you can use the one listed below.

// Specializations

typedef basic_string<TCHAR> tstring; // string of TCHARs

 
// Constructing

string str = "char string";         // construct from a LPCSTR

wstring wstr = L"wide char string"; // construct from a LPCWSTR

tstring tstr = _T("TCHAR string");  // construct from a LPCTSTR

 
// Extracting data

LPCSTR psz = str.c_str();    // read-only pointer to str's buffer

LPCWSTR pwsz = wstr.c_str(); // read-only pointer to wstr's buffer

LPCTSTR ptsz = tstr.c_str(); // read-only pointer to tstr's buffer

Unlike _bstr_t, a basic_string cannot directly convert between character sets. However, you can pass the pointer returned by c_str() to another class's constructor if the constructor accepts the character type, for example:

// Example, construct _bstr_t from basic_string

_bstr_t bs1 = str.c_str();  // construct a _bstr_t from a LPCSTR

_bstr_t bs2 = wstr.c_str(); // construct a _bstr_t from a LPCWSTR

ATL classes

CComBSTR

CComBSTR is ATL's BSTR wrapper, and is more useful in some situations than _bstr_t. Most notably, CComBSTR allows access to the underlying BSTR, which means you can pass a CComBSTR object to COM methods, and the CComBSTR object will automatically manage the BSTR memory for you. For example, say you wanted to call methods of this interface:

// Sample interface:

struct IStuff : public IUnknown
{
  // Boilerplate COM stuff omitted...

  STDMETHOD(SetText)(BSTR bsText);
  STDMETHOD(GetText)(BSTR* pbsText);
};

CComBSTR has an operator BSTR method, so it can be passed directly to SetText(). There is also an operator & that returns a BSTR*, so you can use the & operator on a CComBSTR object to pass it to a function that takes a BSTR*.

CComBSTR bs1;
CComBSTR bs2 = "new text";
 
  pStuff->GetText ( &bs1 );       // ok, takes address of internal BSTR

  pStuff->SetText ( bs2 );        // ok, calls BSTR converter

  pStuff->SetText ( (BSTR) bs2 ); // cast ok, same as previous line

CComBSTR has similar constructors to _bstr_t, however there is no built-in converter to an MBCS string. For that, you can use an ATL conversion macro.

// Constructing

CComBSTR bs1 = "char string";       // construct from a LPCSTR

CComBSTR bs2 = L"wide char string"; // construct from a LPCWSTR

CComBSTR bs3 = bs1;                 // copy from another CComBSTR

CComBSTR bs4;
 
  bs4.LoadString ( IDS_SOME_STR );  // load string from string table

 
// Extracting data

BSTR bstr1 = bs1;        // returns internal BSTR, but don't modify it!

BSTR bstr2 = (BSTR) bs1; // cast ok, same as previous line

BSTR bstr3 = bs1.Copy(); // copies bs1, returns it as a BSTR

BSTR bstr4;
 
  bstr4 = bs1.Detach();  // bs1 no longer manages its BSTR

 
  // ...

  SysFreeString ( bstr3 );
  SysFreeString ( bstr4 );

Note that in the last example, the Detach() method is used. After calling that method, the CComBSTR object no longer manages its BSTR or the associated memory. That's why the SysFreeString() call is necessary on bstr4.

As a footnote, the operator & override means you can't use CComBSTR directly in some STL collections, such as list. The collections require that the & operator return a pointer to the contained class, but applying & to a CComBSTR returns a BSTR*, not a CComBSTR*. However, there is an ATL class to overcome this, CAdapt. For example, to make a list of CComBSTR, declare it like this:

  std::list< CAdapt<CComBSTR> > bstr_list;

CAdapt provides the operators required by the collection, but it is invisible to your code; you can use bstr_list just as if it were a list of CComBSTR.

CComVariant

CComVariant is a wrapper around a VARIANT. However, unlike _variant_t, the VARIANT is not hidden, and in fact you need to access the members of the VARIANT directly. CComVariant provides many constructors to operate on the multitude of types that a VARIANT can contain. I will only cover the string-related operations here.

// Constructing

CComVariant v1 = "char string";       // construct from a LPCSTR

CComVariant v2 = L"wide char string"; // construct from a LPCWSTR

CComBSTR bs1 = "BSTR bob";
CComVariant v3 = (BSTR) bs1;          // copy from a BSTR

 
// Extracting data

CComBSTR bs2 = v1.bstrVal;            // extract BSTR from the VARIANT

Unlike _variant_t, there are no conversion operators to the various VARIANT types. As shown above, you must access the VARIANT members directly and ensure that the VARIANT holds data of the type you expect. You can call the ChangeType() method if you need to convert a CComVariant's data to a BSTR.

CComVariant v4 = ... // Init v4 from somewhere

CComBSTR bs3;
 
  if ( SUCCEEDED( v4.ChangeType ( VT_BSTR ) ))
    bs3 = v4.bstrVal;

As with _variant_t, there is no direct conversion to an MBCS string. You will need to make an interim _bstr_t variable, use another string class that provides the Unicode to MBCS conversion, or use an ATL conversion macro.

ATL conversion macros

ATL's string conversion macros are a very convenient way to convert between character encodings, and are especially useful in function calls. They are named according to the scheme [source type]2[new type] or [source type]2C[new type]. Macros named with the second form convert to a constant pointer (thus the "C" in the name). The type abbreviations are:

A: MBCS string, char* (A for ANSI)
W: Unicode string, wchar_t* (W for wide)
T: TCHAR string, TCHAR*
OLE: OLECHAR string, OLECHAR* (in practice, equivalent to W)
BSTR: BSTR (used as the destination type only)

So, for example, W2A() converts a Unicode string to an MBCS string, and T2CW() converts a TCHAR string to a constant Unicode string.

To use the macros, first include the atlconv.h header file. You can do this even in non-ATL projects, since that header file has no dependencies on other parts of ATL, and doesn't require a _Module global variable. Then, when you use a conversion macro in a function, put the USES_CONVERSION macro at the beginning of the function. This defines some local variables used by the macros.

When the destination type is anything other than BSTR, the converted string is stored on the stack, so if you want to keep the string around for longer than the current function, you'll need to copy the string into another string class. When the destination type is BSTR, the memory is not automatically freed, so you must assign the return value to a BSTR variable or a BSTR wrapper class to avoid memory leaks.

Here are some examples showing various conversion macros:

// Functions taking various strings:

void Foo ( LPCWSTR wstr );
void Bar ( BSTR bstr );
// Functions returning strings:

void Baz ( BSTR* pbstr );
 
#include <atlconv.h>

 
main()
{
using std::string;
USES_CONVERSION;    // declare locals used by the ATL macros

 
// Example 1: Send an MBCS string to Foo()

LPCSTR psz1 = "Bob";
string str1 = "Bob";
 
  Foo ( A2CW(psz1) );
  Foo ( A2CW(str1.c_str()) );
 
// Example 2: Send a MBCS and Unicode string to Bar()

LPCSTR psz2 = "Bob";
LPCWSTR wsz = L"Bob";
BSTR bs1;
CComBSTR bs2;
 
  bs1 = A2BSTR(psz2);         // create a BSTR

  bs2.Attach ( W2BSTR(wsz) ); // ditto, assign to a CComBSTR

 
  Bar ( bs1 );
  Bar ( bs2 );
 
  SysFreeString ( bs1 );      // free bs1 memory

  // No need to free bs2 since CComBSTR will do it for us.

 
// Example 3: Convert the BSTR returned by Baz()

BSTR bs3 = NULL;
string str2;
 
  Baz ( &bs3 );          // Baz() fills in bs3

 
  str2 = W2CA(bs3);      // convert to an MBCS string

  SysFreeString ( bs3 ); // free bs3 memory

}

As you can see, the macros are very handy when passing parameters to a function if you have a string in one format and the function takes a different format.

MFC classes

CString

An MFC CString holds TCHARs, so the exact character type depends on the preprocessor symbols you have defined. In general, a CString is like an STL string, in that you should treat it as an opaque object and modify it only with CString methods. One nice advantage CString has over the STL string is that it has constructors that accept both MBCS and Unicode strings, and it has a converter to LPCTSTR, so you can pass a CString object directly to a function that accepts an LPCTSTR; there is no c_str() method you have to call.

// Constructing

CString s1 = "char string";  // construct from a LPCSTR

CString s2 = L"wide char string";  // construct from a LPCWSTR

CString s3 ( ' ', 100 );  // pre-allocate a 100-byte buffer, fill with spaces

CString s4 = "New window text";
 
  // You can pass a CString in place of an LPCTSTR:

  SetWindowText ( hwndSomeWindow, s4 );
 
  // Or, equivalently, explicitly cast the CString:

  SetWindowText ( hwndSomeWindow, (LPCTSTR) s4 );

You can also load a string from your string table. There is a CString constructor that will do it, along with LoadString(). The Format() method can optionally read a format string from the string table as well.

// Constructing/loading from string table

CString s5 ( (LPCTSTR) IDS_SOME_STR );  // load from string table

CString s6, s7;
 
  // Load from string table.

  s6.LoadString ( IDS_SOME_STR );
 
  // Load printf-style format string from the string table:

  s7.Format ( IDS_SOME_FORMAT, "bob", nSomeStuff, ... );

That first constructor looks odd, but that is actually the documented that way to load a string.

Note that the only legal cast you can apply to a CString is a cast to LPCTSTR. Casting to an LPTSTR (that is, a non-const pointer) is wrong. Getting in the habit of casting a CString to an LPTSTR will only hurt yourself, as when the code does break later on, you might not see why, because you used the same code elsewhere and it happened to work. The correct way to get a non-const pointer to the buffer is the GetBuffer() method.

As an example of the correct usage, consider the case of setting the text of an item in a list control:

CString str = _T("new text");
LVITEM item = {0};
 
  item.mask = LVIF_TEXT;
  item.iItem = 1;
  item.pszText = (LPTSTR)(LPCTSTR) str; // WRONG!

  item.pszText = str.GetBuffer(0);      // correct

 
  ListView_SetItem ( &item );
  str.ReleaseBuffer();  // return control of the buffer to str

The pszText member is an LPTSTR, a non-const pointer, therefore you call GetBuffer() on str. The parameter to GetBuffer() is the minimum length you want CString to allocate for the buffer. If for some reason you wanted a modifiable buffer large enough to hold 1K TCHARs, you would call GetBuffer(1024). Passing 0 as the length just returns a pointer to the current contents of the string.

The crossed-out line above will compile, and it will even work, in this case. But that doesn't mean the code is correct. By using the non-const cast, you're breaking object-oriented encapsulation and assuming something about the internal implementation of CString. If you make a habit of casting like that, you will eventually run into a case where the code breaks, and you'll wonder why it isn't working, because you use the same code everywhere else and it (apparently) works.

You know how people are always complaining about how buggy software is these days? Bugs are caused by the programmers writing incorrect code. Do you really want to write code you know is wrong, and thus contribute to the perception that all software is buggy? Take the time to learn the correct way of using a CString and have your code work 100% of the time.

CString also has two functions that create a BSTR from the CString contents, converting to Unicode if necessary. They are AllocSysString() and SetSysString(). Aside from the BSTR* parameter that SetSysString() takes, they work identically.

// Converting to BSTR

CString s5 = "Bob!";
BSTR bs1 = NULL, bs2 = NULL;
 
  bs1 = s5.AllocSysString();
  s5.SetSysString ( &bs2 );
 
  // ...

  SysFreeString ( bs1 );
  SysFreeString ( bs2 );

COleVariant

COleVariant is pretty similar to CComVariant. COleVariant derives from VARIANT, so it can be passed to a function that takes a VARIANT. However, unlike CComVariant, COleVariant only has an LPCTSTR constructor. There are not separate constructors for LPCSTR and LPCWSTR. In most cases this is not a problem, since your strings will likely be LPCTSTRs anyway, but it is a point to be aware of. COleVariant also has a constructor that accepts a CString.

// Constructing

CString s1 = _T("tchar string");
COleVariant v1 = _T("Bob"); // construct from an LPCTSTR

COleVariant v2 = s1; // copy from a CString

As with CComVariant, you must access the VARIANT members directly, using the ChangeType() method if necessary to convert the VARIANT to a string. However, COleVariant::ChangeType() throws an exception if it fails, instead of returning a failure HRESULT code.

// Extracting data

COleVariant v3 = ...; // fill in v3 from somewhere

BSTR bs = NULL;
 
  try
    {
    v3.ChangeType ( VT_BSTR );
    bs = v3.bstrVal;
    }
  catch ( COleException* e )
    {
    // error, couldn't convert

    }
 
  SysFreeString ( bs );

WTL classes

CString

WTL's CString behaves exactly like MFC's CString, so refer to the description of the MFC CString above.

CLR and VC 7 classes

System::String is the .NET class for handling strings. Internally, a String object holds an immutable sequence of characters. Any String method that supposedly manipulates the String object actually returns a new String object, because the original String is immutable. A peculiarity of Strings is that if you have more than one String containing the same series, of characters all of them actually refer the same object. The Managed Extensions to C++ have a new string literal prefix S, which is used to represent a managed string literal.

// Constructing

String* ms = S"This is a nice managed string";

You can construct a String object by passing an unmanaged string, but this is slightly less efficient than when you construct a String object by passing a managed string. This is because all instances of identical S prefixed strings represent the same object, but this is not true for unmanaged strings. The following code will make this clear:

String* ms1 = S"this is nice";
String* ms2 = S"this is nice";
String* ms3 = L"this is nice";
 
  Console::WriteLine ( ms1 == ms2 ); // prints true

  Console::WriteLine ( ms1 == ms3);  // prints false

The right way to compare strings that may not have been created using S prefixed strings is to use the String::CompareTo() method as shown below:

  Console::WriteLine ( ms1->CompareTo(ms2) );
  Console::WriteLine ( ms1->CompareTo(ms3) );

Both the above lines will print 0, which means the strings are equal.

Converting between a String and the MFC 7 CString is easy. CString has a converter to LPCTSTR and String has two constructors that take a char* and wchar_t*, therefore you can pass a CString straight to a String constructor.

CString s1 ( "hello world" );
String* s2 ( s1 );  // copy from a CString

Converting the other way works similarly:

String* s1 = S"Three cats";
CString s2 ( s1 );

This might puzzle you a bit, but it works because starting with VS.NET, CString has a constructor that accepts a String object:

  CStringT ( System::String* pString );

For some speedy manipulations, you might sometimes want to access the underlying string:

String* s1 = S"Three cats";
 
  Console::WriteLine ( s1 );

const __wchar_t __pin* pstr = PtrToStringChars(s1);
 
  for ( int i = 0; i < wcslen(pstr); i++ )
    (*const_cast<__wchar_t*>(pstr+i))++;
 
  Console::WriteLine ( s1 );

PtrToStringChars() returns a const __wchar_t* to the underlying string which we need to pin down as otherwise the garbage collector might move the string in memory while we are manipulating its contents.

Using string classes with printf-style formatting functions

You must pay careful attention when using string wrapper classes with printf() or any function that works the way printf() does. This includes sprintf() and its variants, as well as the TRACE and ATLTRACE macros. Because there is no type-checking done on the additional parameters to the functions, you must be careful to only pass a C-style string pointer, not a complete string object.

So for example, to pass a string in a _bstr_t to ATLTRACE(), you must explicitly write the (LPCSTR) or (LPCWSTR) cast:

_bstr_t bs = L"Bob!";
 
  ATLTRACE("The string is: %s in line %d\n", (LPCSTR) bs, nLine);

If you forget the cast and pass the entire _bstr_t object, the trace message will display meaningless output, since what will be pushed on the stack is whatever internal data the _bstr_t variable keeps.

Summary of all the classes

The usual way of converting between two string classes is to take the source string, convert it to a C-style string pointer, and then pass the pointer to a constructor in the destination type. So here is a chart showing how to convert a string to a C-style pointer, and which classes can be constructed from C-style pointers.

Class	string type	convert to char*?	convert to const char*?	convert to wchar_t*?	convert to const wchar_t*?	convert to BSTR?	construct from char*?	construct from wchar_t*?
_bstr_t	BSTR	yes, cast1	yes, cast	yes, cast1	yes, cast	yes2	yes	yes
_variant_t	BSTR	no	no	no	cast to _bstr_t3	cast to _bstr_t3	yes	yes
string	MBCS	no	yes, c_str() method	no	no	no	yes	no
wstring	Unicode	no	no	no	yes, c_str() method	no	no	yes
CComBSTR	BSTR	no	no	no	yes, cast to BSTR	yes, cast	yes	yes
CComVariant	BSTR	no	no	no	yes4	yes4	yes	yes
CString	TCHAR	no6	in MBCS builds, cast	no6	in Unicode builds, cast	no5	yes	yes
COleVariant	BSTR	no	no	no	yes4	yes4	in MBCS builds	in Unicode builds
1 Even though _bstr_t provides conversion operators to non-const pointers, modifying the underlying buffer may cause a GPF if you overrun the buffer, or a leak when the BSTR memory is freed. 2 A _bstr_t holds a BSTR internally in a wchar_t* variable, so you can use the const wchar_t* converter to retrieve the BSTR. This is an implementation detail, so use this with caution, as it may change in the future. 3 This will throw an exception if the data cannot be converted to a BSTR. 4 Use ChangeType() then access the bstrVal member of the VARIANT. In MFC, this will throw an exception if the data cannot be converted. 5 There is no BSTR conversion function, however the AllocSysString() method returns a new BSTR. 6 You can temporarily get a non-const TCHAR pointer using the GetBuffer() method.