Re: weird structure definition in header file

["Robert P. J. Day" <rpjday@mindspring.com>]

On Tue, 6 Sep 2005, Robert P. J. Day wrote:

> On Tue, 6 Sep 2005, nhorman@tuxdriver.com wrote:
>
> > On Tue, Sep 06, 2005 at 03:13:48PM -0400, Robert P. J. Day wrote:
> > >
> > >   i'm looking at some legacy code and, in a header file, i find the
> > > following (paraphrased for brevity):
> > >
> > > typedef struct {
> > > 	... stuff ...
> > > } Widgets ;
> > >
> > > extern Widgets Widget ;
> > >
> > >
> > >   huh?  i can see why a header file would want to define a structure
> > > but i'm confused why the *header* file would then refer to an external
> > > object of that type.  that's a new one on me -- typically, i'd expect
> > > a *source* file to define such a thing and other *source* files to
> > > contain the "extern" declaration.
> > >
> > >   is this some subtle programming cleverness of which i am unaware?
> > > thanks.
> > >
> > This is done quite frequently when a data structure needs to be
> > referenced from multiple locations.  They're not necessicarily
> > consecutive like that, but its rather a common practice to extern a
> > instance of a type in a header file.  The kernel source tree does
> > this very frequently, check include/net/ipv4 for lots of examples.

um, still a couple questions on this.  first, in terms of examples,
there *is* no include/net/ipv4 directory in the kernel source tree
anymore, it's just net/ipv4.  and there's only a single header file in
there, which has no example of such a practise.  were you thinking of
a different directory?  (i'm in the 2.6.12.5 source tree.)

also, i'm assuming that some source file must eventually include an
actual definition of a "Widget" for the purposes of linking, as in:

  #include "widgets.h"
  ...
  Widgets Widget ;

however, given the inclusion of the header file, doesn't this give me
both a referencing declaration and a defining declaration of that
object in the same file?  is there no problem with that?  i was under
the impression that common practise was to have a single defining
declaration and all the *remaining* be referencing declarations.

rday