Forward declarations

[Marcin Kalicinski]

Hi All,

In some headers meant to work with both C and C++ the following is often
found:

typedef struct tagMyStruct { /*some members*/ } MyStruct;

Can I forward-declare MyStruct somehow?

Or more generally, can I forward declare a typedef or an enum? Also, is
there any reason why there should be difference between forward declarations
of classes and structs (i.e. why I need to explicitly write keyword 'struct'
or 'class' when forward declaring). In some cases I don't immediately know
what to write, because, for example, the type definition in original header
file is generated with some macro, such as this:

STDINTERFACE(IBlahBlah, IBaseBlahBlah) {
/* Members */
}

Now I need to search for the definition of STDINTERFACE to see how it works.
Even if I find it, it still may silently change in some future version of
the header file, and cause my compilation to fail. Wouldn't allowing a
generic form of forward declaration, without introducing a new keyword:

typename IBlahBlah;

be beneficial to C++?

best regards,
Marcin

 

[Leor Zolman]

Hi All,

In some headers meant to work with both C and C++ the following is often
found:

typedef struct tagMyStruct { /*some members*/ } MyStruct;

Can I forward-declare MyStruct somehow?

Sure, you can say:

typedef struct tagMyStruct MyStruct;

and complete the definition later on, if necessary.

Or more generally, can I forward declare a typedef or an enum?

typedefs, I don't think so. Under Comeau, in C mode, if I did something
like this:
typedef b;
it took it to mean as if I'd said:
typedef int b;
(and emitted a suitable warning).

With enums, Comeau tells me that forward-declaring them is nonstandard. So
it may be supported, but isn't portable.

Also, is
there any reason why there should be difference between forward declarations
of classes and structs (i.e. why I need to explicitly write keyword 'struct'
or 'class' when forward declaring).

Well, you began this post by saying "In some headers meant to work with
both C and C++", so there's an obvious reason to prefer struct right there.

In some cases I don't immediately know
what to write, because, for example, the type definition in original header
file is generated with some macro, such as this:

STDINTERFACE(IBlahBlah, IBaseBlahBlah) {
/* Members */
}

Now I need to search for the definition of STDINTERFACE to see how it works.
Even if I find it, it still may silently change in some future version of
the header file, and cause my compilation to fail. Wouldn't allowing a
generic form of forward declaration, without introducing a new keyword:

typename IBlahBlah;

be beneficial to C++?

I would hope that any such header file gives you a documented way to /use/
the facilities it generates, so that such use is immune to future
implementation changes. If you think you're using something that's going to
morph out from under you in the next release, you'll have to get creative
with, perhaps, some preprocessor (or typedef) hackery of your own. Don't
hold your breath waiting for changes in the language to support you here.
-leor

 

[Jake Montgomery]

Hi All,

In some headers meant to work with both C and C++ the following is often
found:

typedef struct tagMyStruct { /*some members*/ } MyStruct;

Can I forward-declare MyStruct somehow?

Yes:
struct tagMyStruct;
should work.

Or more generally, can I forward declare a typedef or an enum? Also, is

The lack of a forward declaration of enum is my pet peeve also.
VC6.0,7.0 & 7.1 all support it as a language extension, and it was only
when I began writing cross platform code that I discovered that it was
not legal in c/c++. (It can help immensely in avoiding header spaghetti)
I realize there are issues regarding knowing the underlying type of
an enum, but if these compilers can handle it, then it is clearly
possible. Nonetheless, I am not holding my breath.

[Snip]