question about static member

Quote:

---

Originally posted by stober
You're right. Such a simple thing to prove -- this compiles ok with VC6...

---

The implementation of the standard has to be done by the compiler. Visual C++ is known as the worst one regarding the implementation of the C++ specification. Nevertheless I guess even your first example will give you a warning on level 4. But what about another compiler? Besides that I can force every warning to become an error in Visual C++...

But the discussion was not about the intelligence of any compiler, it was about a simple specification issue. And as I and other already said it is pretty easy. It MUST be defined. Always. As stated in the book 'The C++ Programming Language'. Written by the inventor of C++. Period.

I could also ask another question...Why would I declare a static member variable without using it??? It would be just like buying a car without having a driver's license...

Quote:

---

Originally posted by Andreas Masur

The implementation of the standard has to be done by the compiler. Visual C++ is known as the worst one regarding the implementation of the C++ specification. Nevertheless I guess even your first example will give you a warning on level 4. But what about another compiler? Besides that I can force every warning to become an error in Visual C++...

---

Yes, I thought about that too just after I hit the "Submit Reply" button.

Quote:

---

I could also ask another question...Why would I declare a static member variable without using it??? It would be just like buying a car without having a driver's license...

---

I've seen a lot of code where the programmer declared something in a class then later did not use it. Seems like pretty sloppy coding to me, but if they HAD to be declared it might break a lot of lagecy programs.

So much for my 2 cents worth. I'm not an expert on this, so I'll just shut up and let you all battle it out.

Andreas, please read after me the exact sentence Paul quotes from the C++ standard:

Quote:

---

From the ANSI specification, section 9.2.4.5
quote:
--------------------------------------------------------------------------------
There shall be exactly one definition of a static data member that is used in a program.

There shall be exactly one definition of a static data member that is used in a program.

There shall be exactly one definition of a static data member THAT IS USED in a program.

---

Got it?

As for your accusation that I did not read Paul's origial answer, on commenting my word "Had the standard said any thing like "It must be defined". My answer is I did notice Paul quoted a "must" word from Stroustop's book. Please note that Stroustrup's book DOES NOT EQUAL TO THE C++ STANDARD, even though he himself is in the standard committee.

There is not a "must" in the exact sentence in the standard C++ document. It says very clearly that WHEN it is USED, the static class member SHALL be defined exactly once. There is a conditional modifier in that sentence, the condition is when the class member is referenced or used.

There is no point defining something that is never used at all. And the C++ standard does NOT force you to define something that you don't use, even if it is declared.

To stobe, there are legitimate cases where something is declared but never used. You may have one header file, shared by two different applications, if you have a static class member which is used in one application but not the other, you have to declared in in the header file, but you do NOT need to define it any where in the second application where it is never referenced.

Andreas wonders why would one buy a car without a driver license. There are no less than a trillion legitimate reasons who one would buy a car even without a driver license. It's a pity that Andreas's intelligence doesn't allow him to even think of one of those reasons. Try a car rendal company? You don't need a driver license to operate a car rental company, since you do not drive the cars yourself :-)

Quote:

---

Originally posted by Andreas Masur

The implementation of the standard has to be done by the compiler. Visual C++ is known as the worst one regarding the implementation of the C++ specification. Nevertheless I guess even your first example will give you a warning on level 4. But what about another compiler? Besides that I can force every warning to become an error in Visual C++...

But the discussion was not about the intelligence of any compiler, it was about a simple specification issue. And as I and other already said it is pretty easy. It MUST be defined. Always. As stated in the book 'The C++ Programming Language'. Written by the inventor of C++. Period.

I could also ask another question...Why would I declare a static member variable without using it??? It would be just like buying a car without having a driver's license...

---

I made futher reading into this. The standard states that "no diagnostic is required" if the static member's definition is omitted. This is whether you use the static member or not. The code that praxda originally posted could fail to link on another compiler, since the standard doesn't state that a diagnostic "shouldn't be given", only that one is not required. Conversely, if the definition is left out and the static member is used, a compiler (more than likely, the linker) doesn't have to issue a diagnostic, since, again, one is not required. This is according to my reading of the standard, clause 9.4.2.5.

Here is clause 1.4.2 concerning what is meant by "no diagnostic is required":

Quote:

---

If a program contains a violation of a rule for which no diagnostic is required, this International Standard
places no requirement on implementations with respect to that program.

---

Therefore to avoid this, the coder should define the static members. Stroustrup, in all likelihood, states that static members must be defined so as to avoid implementations that do not issue diagnostic messages, leaving your program open to undefined behavior.

Regards,

Paul McKenzie

We all agree that when a static class member is USED, it needs to be defined just once. We disagree with what happens when it is NEVER used any where, in that case whether it should be defined or not.

The C++ standard leaves it open, it does NOT require diagnostics and leaves it to the compilers. But that doesn't mean compilers will implement it in any arbitrary unreasonable way.

Let me make an analog here. The law does NOT require you to report loss of your own private property, nor does it say you shouldn't report your property loss. It leave it open to your own decision. Does that mean you can abuse the system? The law doesn't prohibit you from reporting a loss, however small that may be. But it would be totally ridiculous to report to the police that you have lost a dollar, although it is perfectly legal to do do. On the other hand, also the law does NOT force you to report it to police if you have lost a brand new luxury car, there is hardly any idiot who would see his vehicle stolen without reporting it.

Back to the issue of omitted static class member definition. If the static member is never used at all, there is really no difference whether it is defined or not, except in the case when it is not defined, you save some code size. I hardly see any reason why a compiler should complain. Actually the vast majority of compilers will happily omit even the stuff that you explicitly define in your code, when it deems that the defined stuff is never referenced any where. Can a compiler issue a warning of unreferences and undefined static class member? Sure it can, just as you are completely entitled to the civil right to report a one dollar loss. But it's so silly that no one does it.

In the case where the static class member is not defined, but you do use it in the code, the standard does NOT require the diagnistics, either. But it is almost automatic that all compilers will give an error message. It simply can not link. It is used, hence it must be linked, and if it is not defined, there is nothing to link. Can a compiler ignore the problem and does not issue a diagnistics? It surely is perfectly legal to do so, just like you are not legally required to report loss of your vehicle. But it makes no sense not to report the error or not to report loss of your vehicle.

Another silimar thing is unreferenced local variables. Does un-referenced local variables break any C++ rules? It doesn't. It is perfectly legal standard C++ code. But virtually all compilers will issue a warning about unreferences local variables. They are useless, but probably they don't hurt much except taking up un-necessary stack space, and they do not violate any C++ standard. But they are worth a warning nevertheless.