Passing a class to a sprintf() and similar

Crazy Joe Devola

I am trying to pass a class I wrote to sprintf(). Not the class itself actually, but a pointer to a string the class holds. With the VC++ CString class it works fine: sprintf(t, "%s", cst); But with my class, it either crashes the program or puts garbage into the t buffer: sprintf(t, "%s", c); Unless I use casting, and then it works ok: sprintf(t, "%s", (char *)c); What is my class missing? Some kind of an operator? Here is my code:

class C1
{
public:
C1()
{
sprintf(m_Data,"C1 class");
}

operator char *() { return m_Data; }

private:
char m_Data[100];
};

int main(int argc, char* argv[])
{
char t[1000]="";
C1 c;
CString cst("Test c string");

sprintf(t, "%s", cst);
sprintf(t, "%s", c);
return 0;
}

CPallini

Crazy Joe Devola wrote:

What is my class missing? Some kind of an operator?

You are missing nothing: Try the following statement:

char * p = c;

However the cast (either the one you used or the static_cast<char *>()) is required in sprintf due to its signature (the type of its variadic arguments is established dynamically based on the fmt parameter) the char*() operator must be explicitely called.

THESE PEOPLE REALLY BOTHER ME!! How can they know what you should do without knowing what you want done?!?! -- C++ FQA Lite

Crazy Joe Devola

Thank you. Yes, char * works. However, when i use the CString class as in my example, no casting or conversion is required at all. What does the CString class have that my C1 does not? (obviously a lot :-) but specifically for what I am trying to do...)

CPallini

You are right, of course. And that is, at the moment, a mystery to me too. :-) However, have a look at this MSDN page: CString::operator LPCTSTR[^]. That's make me guess in your actual code m_Data is not the unique (and not the first) data member.

THESE PEOPLE REALLY BOTHER ME!! How can they know what you should do without knowing what you want done?!?! -- C++ FQA Lite

Lost User

I just tested OP's code (without the CString - no MFC or ATL) and it sends the actual C1 object as the last parameter, so the pointer will be rubbish. I also would like to know how CString manages to coerce the () operator without a cast.

CPallini

It is not rubbish, in my test (VS2012). It is pointer to the C1 object, having the same address of its first data member (m_Data) in this simple case, so everything goes right (of course if you insert, before m_Data, another member variable, for instance an int, the sprintf gets some garbage). In the linked MSDN page they indded suggest to explicitely cast the CString object when used as sprintf parameter.

THESE PEOPLE REALLY BOTHER ME!! How can they know what you should do without knowing what you want done?!?! -- C++ FQA Lite

Lost User

Interesting. I used VS2010 Express, and copying OP's code, it copies the entire object onto the stack, thus the 'pointer' is garbage. When I use a cast it pushes the address of the object onto the stack and works as it should. More testing required ...

Lost User

OK, I finally figured it out. CString is a bit more of a sophisticated class, and will not use a fixed length buffer for its data. It uses a dynamically allocated one, and as long as that is the first data item in the class definition, the sprintf call will work; although it is luck rather than anything else. You can check this out with the following code:

class C1
{
public:
C1()
{
// allocate space for the text and then copy to the dynamic buffer
m_Data = (char*)malloc(100);
sprintf(m_Data,"C1 class");
}

operator char \*() { return m\_Data; }

private:
char* m_Data; // make buffer a pointer

};

When you now use the object without the cast, it just copies the object's data onto the stack, and since this data is a pointer to the actual text it all works as you expect. But be aware that this may have unintended consequences, so always use the cast when you want this behaviour.

Lost User

See my answer to OP below.

CPallini

However, I don't see the difference on my system (that is it works with m_Data as provided in the OP).

THESE PEOPLE REALLY BOTHER ME!! How can they know what you should do without knowing what you want done?!?! -- C++ FQA Lite

Lost User

Definitely different for me, it only worked with the modified code.

CPallini

And apparently the 'strange case' is mine.

THESE PEOPLE REALLY BOTHER ME!! How can they know what you should do without knowing what you want done?!?! -- C++ FQA Lite

Lost User

Well it can't be just you, there must be something else that is not obvious that happens in your compiler. I only managed to work out what was happening on my system by getting the assembler listing from the compiler, so maybe your version of the compiler is different from the two I used.

jschell

Richard MacCutchan wrote:

although it is luck rather than anything else

The class has been that way for a very long time because I remember being mystified by exactly the same thing a long, long time ago. I think when I looked at source for the class it was also set up in such a way that there was in fact basically one data member (there was some tricky code going on.) So even if you used it in the middle everything still works. Because of that I wonder if there has been a deliberate effort to insure that behavior (versus luck.)

Albert Holguin

I thought this was done on pupose on CString for this very reason... (not sure if you mean the luck part referring to CString)

Crazy Joe Devola

Cool! thank you very much Richard. :thumbsup:

Bram van Kampen

Well I have noticed, VC 2012 has different parsing and scanning rules than earlier versions. The question is: Did the rules change in the standard, or was it incorrectly implemented by Microsoft. :)

Bram van Kampen

jschell

Albert Holguin wrote:

I thought this was done on pupose on CString for this very reason... (not sure if you mean the luck part referring to CString)

Yes that is what I meant - I was supposing that it was possible that it was planned that way from the start.