🚀 go-pugleaf

martelli@cadlab.cadlab.it (Alex Martelli) wrote:

>st@mlc-ratingen.de writes:
>	...
>>> Note that you still need a definition of X::fifty somewhere:
>>> 
>>> 	int X::fifty;
>
>>This one has always puzzled me. What is the rationale behind *still* 
>>requiring that X::fifty be defined somewhere if it is initialized in
>>the class declaration?
>
>Suppose the program, in various different source files, takes the address
>&X::fifty, and saves it in int* variables; then, at some later points,
>those variables are compared for equality of their contents.  We want
>to ensure that such equality holds.  That the _contents_ of X::fifty
>are defined in the header seen by each such source file has obviously
>no relevance to this -- the _location_ of it must also be defined,
>and, clearly, it will have to be in exactly one place to guarantee
>the equality holds...

Why must it be defined in only one place.  Consider

	class A
	{
	  public:
	    static const int c = 1;
	    void f()
	    {
	      static const int i = 1;
	      ...
	    }
	}

in a header.

The compiler must create only one copy of i no matter how many
translation units include the header.  Why can't the mechanism that
does this also be used for c?

Michael M Rubenstein


---
[ comp.std.c++ is moderated.  Submission address: std-c++@ncar.ucar.edu.
  Contact address: std-c++-request@ncar.ucar.edu.  The moderation policy
  is summarized in http://dogbert.lbl.gov/~matt/std-c++/policy.html. ]