const static initializers versionitis

const static initializers versionitis

[Gianni Mariani]

The code below compiles fine with egcs-2.91.66 but fails with 2.95.2 and

2.95.3 .   Am I missing somthing ?  Any way of getting the egcs-2.91.66
behaviour in 2.95.3 ?

g++2.95.2 -c yy.cpp
yy.cpp:7: uninitialized const `bar'
yy.cpp:9: uninitialized const `tar'
--------- yy.cpp -----------
struct foo {
        const struct foo * y;
        char    who[ 10 ];
};

const static struct foo bar[];

const static struct foo tar[];

const static struct foo bar[] = {
        tar, "bar"
};

const static struct foo tar[] = {
        bar, "tar"
};

const char * zefuc()
{
        return tar->y->who;
}

Re: const static initializers versionitis

[Gianni Mariani]

help - please ...

Why were forward declarations of const arrays removed from g++
since 2.95.2 ?

Isn't this a gcc bug ? i.e. not being able to forward declare an
array instance ?

The code still compiles fine in gcc so I would have some level
of expectation that "const"ness should not change the behaviour
of the compiler.

Is there a compiler switch that reinstates the desired behaviour ?

> 
> 
> The code below compiles fine with egcs-2.91.66 but fails with 2.95.2 and
> 
> 2.95.3 .   Am I missing somthing ?  Any way of getting the egcs-2.91.66
> behaviour in 2.95.3 ?
> 
> g++2.95.2 -c yy.cpp
> yy.cpp:7: uninitialized const `bar'
> yy.cpp:9: uninitialized const `tar'
> --------- yy.cpp -----------
struct foo {
        const struct foo * y;
        char    who[ 10 ];
};

const static struct foo bar[];

const static struct foo tar[];

const static struct foo bar[] = {
        tar, "bar"
};

const static struct foo tar[] = {
        bar, "tar"
};

const char * zefuc()
{
        return tar->y->who;
}

Re: const static initializers versionitis

[Alexandre Oliva]

On Aug 22, 2000, Gianni Mariani <gianni_@uluru.mariani.ws> wrote:

> Why were forward declarations of const arrays removed from g++
> since 2.95.2 ?

`extern' is the keyword to forward-declare data in namespace scope.

> Isn't this a gcc bug?

Nope.

> The code still compiles fine in gcc so I would have some level
> of expectation that "const"ness should not change the behaviour
> of the compiler.

const POD must be initialized.

> Is there a compiler switch that reinstates the desired behaviour ?

-fpermissive, maybe?

-- 
Alexandre Oliva   Enjoy Guarana', see http://www.ic.unicamp.br/~oliva/
Red Hat GCC Developer                  aoliva@{cygnus.com, redhat.com}
CS PhD student at IC-Unicamp        oliva@{lsd.ic.unicamp.br, gnu.org}
Free Software Evangelist    *Please* write to mailing lists, not to me

Re: const static initializers versionitis

[Gianni Mariani]

Thx Alexandre

-fpermissive seems not to be the solution ...
$ g++ -fpermissive -c yy.cpp
yy.cpp:7: uninitialized const `bar'
yy.cpp:9: uninitialized const `tar'

> const POD must be initialized.  ....

So how does one forward declare a const POD ?  As you can see in the
code the arrays point to each other !

I have 2 applications I want to move to the later C++ compiler and both
applications generate code with statically initialized arrays of various
kinds of objects which should be declared const.

So it seems as though the rule "const POD must be initialized" would
preclude initializing these kind of data structures - kind of weak
wouldn't
you say ?

So the alternative would be to not make them const and not make them
static and then we pollute the name space - which is also bad - .

Do you know what the motivation is behind this rule ?  Is it just silly
or
is there a reason for the madness ?

G


Alexandre Oliva wrote:

> On Aug 22, 2000, Gianni Mariani <gianni_@uluru.mariani.ws> wrote:
>
> > Why were forward declarations of const arrays removed from g++
> > since 2.95.2 ?
>
> `extern' is the keyword to forward-declare data in namespace scope.
>
> > Isn't this a gcc bug?
>
> Nope.
>
> > The code still compiles fine in gcc so I would have some level
> > of expectation that "const"ness should not change the behaviour
> > of the compiler.
>
> const POD must be initialized.
>
> > Is there a compiler switch that reinstates the desired behaviour ?
>
> -fpermissive, maybe?
>
> --
> Alexandre Oliva   Enjoy Guarana', see http://www.ic.unicamp.br/~oliva/
> Red Hat GCC Developer                  aoliva@{cygnus.com, redhat.com}
> CS PhD student at IC-Unicamp        oliva@{lsd.ic.unicamp.br, gnu.org}
> Free Software Evangelist    *Please* write to mailing lists, not to me

Re: const static initializers versionitis

[Alexandre Oliva]

On Aug 22, 2000, Gianni Mariani <gianni@mariani.ws> wrote:

> So how does one forward declare a const POD?

> Alexandre Oliva wrote:

>> `extern' is the keyword to forward-declare data in namespace scope.

-- 
Alexandre Oliva   Enjoy Guarana', see http://www.ic.unicamp.br/~oliva/
Red Hat GCC Developer                  aoliva@{cygnus.com, redhat.com}
CS PhD student at IC-Unicamp        oliva@{lsd.ic.unicamp.br, gnu.org}
Free Software Evangelist    *Please* write to mailing lists, not to me

Re: const static initializers versionitis

[Gianni Mariani]

Forgot to mention - tried that !

I want a "static" - so declaring it extern is specifically not
what I (ideally) would want. i.e. can't have an "extern static"
declaration.

$ g++ -c yy.cpp
yy.cpp:7: multiple storage classes in declaration of `bar'
yy.cpp:9: multiple storage classes in declaration of `tar'

------------ new yy.cpp ---------------
struct foo {
 const struct foo * y;
 char who[ 10 ];
};

extern const static struct foo bar[];

extern const static struct foo tar[];

const static struct foo bar[] = {
 tar, "bar"
};

const static struct foo tar[] = {
 bar, "tar"
};


const char * zefuc()
{
 return tar->y->who;
}


>
> On Aug 22, 2000, Gianni Mariani <gianni@mariani.ws> wrote:
>
> > So how does one forward declare a const POD?
>
> > Alexandre Oliva wrote:
>
> >> `extern' is the keyword to forward-declare data in namespace scope.
>
> --
> Alexandre Oliva   Enjoy Guarana', see http://www.ic.unicamp.br/~oliva/
> Red Hat GCC Developer                  aoliva@{cygnus.com, redhat.com}
> CS PhD student at IC-Unicamp        oliva@{lsd.ic.unicamp.br, gnu.org}
> Free Software Evangelist    *Please* write to mailing lists, not to me

Re: const static initializers versionitis

[Alexandre Oliva]

On Aug 22, 2000, Gianni Mariani <gianni@mariani.ws> wrote:

> I want a "static" - so declaring it extern is specifically not
> what I (ideally) would want. i.e. can't have an "extern static"
> declaration.

I see.  You'll hopefully have better luck with anonymous namespaces.
AFAICT, there's no way to forward-declare static data items in C++.

namespace {
  extern const struct foo bar[];
  extern const struct foo tar[];

  const struct foo bar[] = { ... };
  const struct foo tar[] = { ... };
}

-- 
Alexandre Oliva   Enjoy Guarana', see http://www.ic.unicamp.br/~oliva/
Red Hat GCC Developer                  aoliva@{cygnus.com, redhat.com}
CS PhD student at IC-Unicamp        oliva@{lsd.ic.unicamp.br, gnu.org}
Free Software Evangelist    *Please* write to mailing lists, not to me

Re: const static initializers versionitis

[Gianni Mariani]

I don't know how to "export" a symbol from an anonymous
namespace.  But somthing like that would make this ANS's
viable.

$ g++ -c yy.cpp
yy.cpp:20: parse error before `('

------------ yy.cpp ------------
namespace {

struct foo {
        const struct foo * y;
        char    who[ 10 ];
};

extern const struct foo bar[];

extern const struct foo tar[];

const struct foo bar[] = {
        tar, "bar"
};

const struct foo tar[] = {
        bar, "tar"
};

const char * ::zefuc()
{
        return tar->y->who;
}

};

Alexandre Oliva wrote:

> On Aug 22, 2000, Gianni Mariani <gianni@mariani.ws> wrote:
>
> > I want a "static" - so declaring it extern is specifically not
> > what I (ideally) would want. i.e. can't have an "extern static"
> > declaration.
>
> I see.  You'll hopefully have better luck with anonymous namespaces.
> AFAICT, there's no way to forward-declare static data items in C++.
>
> namespace {
>   extern const struct foo bar[];
>   extern const struct foo tar[];
>
>   const struct foo bar[] = { ... };
>   const struct foo tar[] = { ... };
> }
>
> --
> Alexandre Oliva   Enjoy Guarana', see http://www.ic.unicamp.br/~oliva/
> Red Hat GCC Developer                  aoliva@{cygnus.com, redhat.com}
> CS PhD student at IC-Unicamp        oliva@{lsd.ic.unicamp.br, gnu.org}
> Free Software Evangelist    *Please* write to mailing lists, not to me

Re: const static initializers versionitis

[Alexandre Oliva]

On Aug 23, 2000, Gianni Mariani <gianni@mariani.ws> wrote:

> I don't know how to "export" a symbol from an anonymous
> namespace.

An anonymous namespace is implicitly `using'ed :-P by the enclosing
namespace.

> $ g++ -c yy.cpp
> yy.cpp:20: parse error before `('

Move the function definition out of the namespace.

-- 
Alexandre Oliva   Enjoy Guarana', see http://www.ic.unicamp.br/~oliva/
Red Hat GCC Developer                  aoliva@{cygnus.com, redhat.com}
CS PhD student at IC-Unicamp        oliva@{lsd.ic.unicamp.br, gnu.org}
Free Software Evangelist    *Please* write to mailing lists, not to me

Re: const static initializers versionitis

[Gianni Mariani]

OK - that works. Thanks.

Alexandre Oliva wrote:

> On Aug 23, 2000, Gianni Mariani <gianni@mariani.ws> wrote:
>
> > I don't know how to "export" a symbol from an anonymous
> > namespace.
>
> An anonymous namespace is implicitly `using'ed :-P by the enclosing
> namespace.
>
> > $ g++ -c yy.cpp
> > yy.cpp:20: parse error before `('
>
> Move the function definition out of the namespace.
>
> --
> Alexandre Oliva   Enjoy Guarana', see http://www.ic.unicamp.br/~oliva/
> Red Hat GCC Developer                  aoliva@{cygnus.com, redhat.com}
> CS PhD student at IC-Unicamp        oliva@{lsd.ic.unicamp.br, gnu.org}
> Free Software Evangelist    *Please* write to mailing lists, not to me