Oddity with pointer initializer

[Smeckler]

I can't find any references to this in google's archive, but I'm sure it
must be an old chestnut...

The declaration

MyClass* p(NULL);

is fine under gcc, but not VC6. I suspect VC6 thinks this is a declaration
of a pointer to a function taking an argument of type 'NULL' ?

So is the right way?

MyClass (*p)(NULL);

Or is gcc correct to allow the above?

Cheers all.

[John Carson]

Re: Oddity with pointer initializer

"Smeckler" <smNOecklers@ho tSPAMmail.com> wrote in message
news:bg2voq$hkn $1$8302bc10@new s.demon.co.uk[color=blue]
> I can't find any references to this in google's archive, but I'm sure
> it must be an old chestnut...
>
> The declaration
>
> MyClass* p(NULL);
>
> is fine under gcc, but not VC6. I suspect VC6 thinks this is a
> declaration of a pointer to a function taking an argument of type
> 'NULL' ?[/color]


NULL is not a type, so your code cannot be a function declaration. VC 7
compiles it without a problem.


--
John Carson
1. To reply to email address, remove donald
2. Don't reply to email address (post here instead)

[John Carson]

Re: Oddity with pointer initializer

"Smeckler" <smNOecklers@ho tSPAMmail.com> wrote in message
news:bg358t$o2i $1$8300dec7@new s.demon.co.uk[color=blue][color=green][color=darkred]
> > > The declaration
> > >
> > > MyClass* p(NULL);
> > >
> > > is fine under gcc, but not VC6. I suspect VC6 thinks this is a
> > > declaration of a pointer to a function taking an argument of type
> > > 'NULL' ?[/color]
> >
> >
> > NULL is not a type, so your code cannot be a function declaration.
> > VC 7 compiles it without a problem.
> >[/color]
>
> OK, just to clarify:
>
> MyClass* p = NULL;
>
> is fine, whereas
>
> MyClass* p(NULL);
>
> gives "error C2059: syntax error : 'constant'" in VC6.[/color]

Both forms are fine in VC++ 7.
[color=blue]
> The only thing I can think is that VC is parsing it as a "ptr to
> function" declaration and thinks that NULL is supposed to be the type
> of the argument to the function.[/color]

I am not sure of the exact history here, but the traditional way to
initialise a built in type is with

BuiltInType variable = value;

The traditional way to initialise a class object is with

Class object(value);

(assuming a single argument constructor). In order to permit a more
consistent syntax, at some point it became possible to initialise built-in
types (including pointers) using the syntax for initialising a class object.
My guess is that VC++6 simply pre-dates that change or at least Microsoft
hadn't got around to fully implementing it at that stage.
[color=blue]
> I wonder if I should be writing:
>
> MyClass (*p)(NULL);
>[/color]

This also works with VC++7 but only because the brackets have no effect in
this case.


--
John Carson
1. To reply to email address, remove donald
2. Don't reply to email address (post here instead)

[Karl Heinz Buchegger]

Re: Oddity with pointer initializer



Smeckler wrote:[color=blue]
>
> I can't find any references to this in google's archive, but I'm sure it
> must be an old chestnut...
>
> The declaration
>
> MyClass* p(NULL);
>
> is fine under gcc, but not VC6. I suspect VC6 thinks this is a declaration
> of a pointer to a function taking an argument of type 'NULL' ?
>[/color]

What is the exact error message?

[color=blue]
> So is the right way?
>
> MyClass (*p)(NULL);
>
> Or is gcc correct to allow the above?[/color]

The above is fine, since NULL is no data type but either:
* an undeclared identifier
* the literal constant 0 or an equivalent expression,
due to inclusion of some header which defines the
macro NULL in that way.

and none of those 2 make sense in a function declaration, so it
can't be a function declaration but it must be a variable
declaration with an initialization.

You could write:

MyClass* p = NULL;

--
Karl Heinz Buchegger
kbuchegg@gascad .at

[Alf P. Steinbach]

Re: Oddity with pointer initializer

On Mon, 28 Jul 2003 12:00:59 +0100, "Smeckler" <smNOecklers@ho tSPAMmail.com> wrote:
[color=blue]
>I can't find any references to this in google's archive, but I'm sure it
>must be an old chestnut...
>
>The declaration
>
>MyClass* p(NULL);
>
>is fine under gcc, but not VC6. I suspect VC6 thinks this is a declaration
>of a pointer to a function taking an argument of type 'NULL' ?[/color]

It works fine in VC7.

Have you included a definition of NULL, e.g. via [cstddef]?

[color=blue]
>So is the right way?
>
>MyClass (*p)(NULL);[/color]

That's unnecessarily complicated.


[color=blue]
>Or is gcc correct to allow the above?[/color]

Yep.