There have been a few questions before on exporting a class which contains stl classes in relation to visual studio warning C4251: E.g. this question or this question. I have already read the excellent explanation at UnknownRoad.
Blindly disabling the warning seems a little dangerous, though it may be an option. Wrapping all those std classes and exporting those is also not really an option. It is after all called the Standard Template Library... I.e., one wants to provide an interface with these standard classes.
How can I use stl-classes in my dll-interface? What are common practices?
Reason: The STL is a code library, not a binary library like a DLL. It does not have a single ABI that is guaranteed to be the same wherever you might use it. Indeed, STL does stand for "Standard Template Library," but a key operative word here besides Standard is Template.
In layman words: STL is part of Standard Library.
The C runtime Library (CRT) is the part of the C++ Standard Library that incorporates the ISO C standard library. The Visual C++ libraries that implement the CRT support native code development, and both mixed native and managed code. All versions of the CRT support multi-threaded development.
C++ runtime library is the library shipped with the toolset to provide standard library functionality, and probably some internal stuff the compiler might need. In fact, those terms are often interchangeable.
Keep in mind one thing before you read further: My answer is coming from the point of view of writing portable code that can be used in applications made up of modules compiled under different compilers. This can include different versions or even different patch levels of the same compiler.
How can I use stl-classes in my dll-interface?
Answer: You often can't1.
Reason: The STL is a code library, not a binary library like a DLL. It does not have a single ABI that is guaranteed to be the same wherever you might use it. Indeed, STL does stand for "Standard Template Library," but a key operative word here besides Standard is Template.
The Standard defines the methods and data members each STL class is required to provide, and it defines what those methods are to do; but no more. In particular, the Standard doesn't specify how compiler writers should implement the Standard-defined functionality. Compiler writers are free to provide a implementation of an STL class that adds member functions and member variables not listed in the Standard, so long as those members which are defined in the Standard are still there and do what the Standard says.
Maybe an example is in order. The basic_string
class is defined in the Standard as having certain member functions & variables. The actual definition is almost 4 pages in the Standard, but here's just a snippet of it:
namespace std {
template<class charT, class traits = char_traits<charT>,
class Allocator = allocator<charT> >
class basic_string {
[snip]
public:
// 21.3.3 capacity:
size_type size() const;
size_type length() const;
size_type max_size() const;
void resize(size_type n, charT c);
void resize(size_type n);
size_type capacity() const;
void reserve(size_type res_arg = 0);
void clear();
bool empty() const;
[snip]
};
Consider the size()
and length()
member functions. There is nothing in the Standard that specified member variables for holding this information. Indeed, there are no member variables defined at all, not even to hold the string itself. So how is this implemented?
The answer is, many different ways. Some compilers might use a size_t
member variable to hold the size and a char*
to hold the string. Another one might use a pointer to some other data store which holds that data (this might be the case in a reference-counted implementation). In fact, different versions or even patch levels of the same compiler may change these implementation details. You can't rely on them. So, MSVC 10's implementation might look like this:
namespace std {
template<class charT, class traits = char_traits<charT>,
class Allocator = allocator<charT> >
class basic_string {
[snip]
char* m_pTheString;
};
size_t basic_string::size() const { return strlen(m_pTheString;) }
...Whereas MSVC 10 with SP1 might look like this:
namespace std {
template<class charT, class traits = char_traits<charT>,
class Allocator = allocator<charT> >
class basic_string {
[snip]
vector<char> m_TheString;
};
size_t basic_string::size() const { return m_TheString.size(); }
I'm not saying they do look like this, I'm saying they might. The point here is the actual implementation is platform-dependent, and you really have no way of knowing what it will be anywhere else.
Now say you use MSVC10 to write a DLL that exports this class:
class MyGizmo
{
public:
std::string name_;
};
What is the sizeof(MyGizmo)
?
Assuming my proposed implementations above, under MSVC10 its going to be sizeof(char*)
, but under SP1 it will be sizeof(vector<char>)
. If you write an application in VC10 SP1 that uses the DLL, the size of the object will look different than it actually is. The binary interface is changed.
For another treatment of this, please see C++ Coding Standards (Amazon link) issue # 63.
1: "You often can't" You actually can export Standard Library components or any other code library components (such as Boost) with a fair amount of reliability when you have complete control over the toolchains and the libraries.
The fundamental problem is that with source code libraries the sizes and definitions of things can be different between different compilers and different versions of the library. If you are working in an environment where you control both of these things everywhere your code is used, then you probably won't have a problem. For example at a trading firm where all the systems are written in-house and used only in-house, it might be possible to do this.
If you love us? You can donate to us via Paypal or buy me a coffee so we can maintain and grow! Thank you!
Donate Us With