On Thu, Apr 04, 2019 at 01:05:29PM +0100, Ben Dooks wrote:
> iOn 04/04/2019 12:48, Luc Van Oostenryck wrote:
> > On Thu, Apr 04, 2019 at 10:53:08AM +0100, Ben Dooks wrote:
> > > On 03/04/2019 21:46, Luc Van Oostenryck wrote:
> > > > On Wed, Apr 03, 2019 at 04:35:48PM +0100, Ben Dooks wrote:
> > > > > @@ -2981,8 +3071,18 @@ struct token *external_declaration(struct token *token, struct symbol_list **lis
> > > > > if (!(decl->ctype.modifiers & MOD_STATIC))
> > > > > decl->ctype.modifiers |= MOD_EXTERN;
> > > > > +
> > > > > + base_type->ctype.printf_msg = decl->ctype.printf_msg;
> > > > > + base_type->ctype.printf_va_start = decl->ctype.printf_va_start;
> > > > > } else if (base_type == &void_ctype && !(decl->ctype.modifiers & MOD_EXTERN)) {
> > > > > sparse_error(token->pos, "void declaration");
> > > > > + } else if (base_type && base_type->type == SYM_PTR) {
> > > > > + // think this is correct to do //
> > > > > + struct symbol *ptr_base = get_base_type(base_type);
> > > > > + if (ptr_base) {
> > > > > + ptr_base->ctype.printf_msg = decl->ctype.printf_msg;
> > > > > + ptr_base->ctype.printf_va_start = decl->ctype.printf_va_start;
> > > > > + }
> > > >
> > > > I'll need to check this.
> > > > What situation is this supposed to cover?
> > >
> > > definition of pointers to functions, ie:
> > >
> > > int (*ptr)(args..) __attribute__((format...));
> >
> > But this is not the correct syntax as it applies the attribute
> > to the declared object 'ptr', a pointer type and not to the
> > underlying function type.
> >
> > The syntax used in the kernel is (cfr. kdb_printf_t):
> > __attribute__((...)) int (*ptr)(args...);
> > and GCC's manual also specifies:
> > int (__attribute__((...)) *ptr)(args...);
> > (but I'm not sure this one is correctly supported by sparse).
> >
> > It would be good to issue a warning if the 'format' attribute is
> > applied to anything that is not a function type.
>
> ok, gcc doesn't seem to issue a warning about that.
> in fact, it seems happy about:
>
> int main(void)
> {
> int (*ptr)(const char *msg, ...) __attribute__((format(printf,1,2)));
>
> (ptr)("moose %s %d\n", 1, 2);
> return 0;
> }
>
> gcc -O2 -Wall test4.c
>
> test4.c: In function ‘main’:
> test4.c:6:17: warning: format ‘%s’ expects argument of type
> ‘char *’, but argument 2 has type ‘int’ [-Wformat=]
> (ptr)("moose %s %d\n", 1, 2);
> ~^ ~
> %d
OK. I'm not very surprised, as the manual also says:
"... at present the <x> attribute applies to 'f', which causes
a warning since 'f' is not a function, but in future it may apply
to the function '****f'. The precise semantics of what attribute
in such cases will apply to are not yet specified."
I propose for the moment to just support the simple case used for the
kernel (some other changes in the handling of attributes in declarations
are anyway needed).
-- Luc
