Re: including sparse headers in C++ code

[Josh Triplett <josh@joshtriplett.org>]

On Sat, Oct 16, 2010 at 06:03:53PM +0200, Tomas Klacko wrote:
> On Wed, Oct 13, 2010 at 2:37 AM, Christopher Li <sparse@chrisli.org> wrote:
> > On Tue, Oct 12, 2010 at 3:45 PM, Tomas Klacko <tomas.klacko@gmail.com> wrote:
> >> diff --git a/parse.h b/parse.h
> >> index 6b21e23..f2193e7 100644
> >> --- a/parse.h
> >> +++ b/parse.h
> >> @@ -35,10 +35,12 @@ struct statement {
> >>                struct /* declaration */ {
> >>                        struct symbol_list *declaration;
> >>                };
> >> +#ifndef __cplusplus
> >>                struct /* label_arg */ {
> >>                        struct symbol *label;
> >>                        struct statement *label_statement;
> >>                };
> >> +#endif
> >
> > What is this #ifndef  for?
> 
> I use it to avoid this error:
> 
> [...]/sparse/parse.h:65: error: declaration of ‘statement*
> statement::<anonymous union>::<anonymous struct>::label_statement’
> [...]sparse/parse.h:41: error: conflicts with previous declaration
> ‘statement* statement::<anonymous union>::<anonymous
> struct>::label_statement’

Huh, interesting.  Turns out that giant union contains two different
structs of the form { struct symbol *, struct statement * }; one for
label_arg and one for labeled_struct.  They have different names for the
symbol, but the same name for the statement.  We should fix that for C
as well, either by renaming one of the statements, or by just making
both variations use the same set of fields and dropping one of the
sub-structs.  Either way, that should happen as an independent patch,
separate from the C++ compatibility changes.

Chris, what do you think?

> >>  /* Silly type-safety check ;) */
> >>  #define DECLARE_PTR_LIST(listname,type)        struct listname { type *list[1]; }
> >> -#define CHECK_TYPE(head,ptr)           (void)(&(ptr) == &(head)->list[0])
> >>  #define TYPEOF(head)                   __typeof__(&(head)->list[0])
> >>  #define VRFY_PTR_LIST(head)            (void)(sizeof((head)->list[0]))
> >>
> >> +#ifndef __cplusplus
> >> +#define CHECK_TYPE(head,ptr)           (void)(&(ptr) == &(head)->list[0])
> >> +#else
> >> +/* I don't know yet how to do this better in C++. */
> >> +#define CHECK_TYPE(head,ptr) (void)((void*)&(ptr) == (void*)&(head)->list[0])
> >> +#endif
> >
> > If you can't get CHECK_TYPE work in C++, you might just make it an empty define
> > instead of doing useless point dancing. At least it is clear that it does not
> > do any thing here.
> 
> True. How about
> #define CHECK_TYPE (head,ptr)   (void)(1)
> ?

As far as I can tell, CHECK_TYPE works just fine in C++.  I could easily
compile an invocation of CHECK_TYPE, as well as some simple examples
that called the macros which invoked CHECK_TYPE.  When I tried
FOR_EACH_PTR, I encountered *other* warnings (related to assigning (void
*) to some other type without a cast), but those warnings didn't come
from CHECK_TYPE.

What warning do you encounter about CHECK_TYPE?

- Josh Triplett
--
To unsubscribe from this list: send the line "unsubscribe linux-sparse" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html