Re: [PATCH 1/1] checkpatch.pl: thou shalt not use () or (...) in function declarations

[Nick Bowler <nbowler@elliptictech.com>]

On 2012-03-22 13:17 -0400, Nick Bowler wrote:
> On 2012-03-22 17:22 +0100, Jiri Slaby wrote:
> > On 03/22/2012 04:27 PM, Phil Carmody wrote:
> [...]
> > > diff --git a/scripts/checkpatch.pl b/scripts/checkpatch.pl
> > > index a3b9782..3993011 100755
> > > --- a/scripts/checkpatch.pl
> > > +++ b/scripts/checkpatch.pl
> > > @@ -1881,6 +1881,10 @@ sub process {
> > >  				substr($ctx, 0, $name_len + 1, '');
> > >  				$ctx =~ s/\)[^\)]*$//;
> > >  
> > > +				if ($ctx =~ /^\s*(?:\.\.\.)?\s*$/) {
> > > +					# HPA explains why: http://lwn.net/Articles/487493/
> > > +					ERROR("(...) and () are not sufficiently informative function declarations\n$hereline");
> > > +				}
> > 
> > That explanation is not fully correct. C99 explicitly says (6.7.5.3.14):
> > An identifier list declares only the identifiers of the parameters of
> > the function. An empty list in a function declarator that is part of a
> > definition of that function specifies that the function has no
> > parameters.
> 
> Nevertheless, an empty identifier list in a declaration is still not the
                                              ^^^^^^^^^^^
That should obviously have said "definition".  Sigh.

> same as a parameter type list with (void).  In particular, the empty
> identifier list *is not a prototype declaration for the function*.  That
> means that arguments passed to the function are not subject to the usual
> checks/conversions implied by a prototype.
> 
> Consider:
> 
>   int foo()
>   {
>      return 0;
>   }
> 
>   int main(void)
>   {
>     return foo(1, 2, 3, 4, 5); /* this is syntactically OK; undefined
>                                   behaviour at runtime. */
>   }
> 
> GCC will not normally warn about the above (unless you pass
> -Wold-style-definition) which warns for all function definitions that
> lack a prototype.  On the other hand, changing it to int foo(void)
> provides the required prototype for the arguments to be checked, and the
> above becomes a proper error.
> 
> > So what you are trying to force here holds only for (forward)
> > declarations. Not for functions with definitions (bodies). Is
> > checkpatch capable to differ between those?
> 
> For the above reasons, non-prototype declarations of any sort should be
> avoided.  No need for checkpatch to distinguish between whether or not
> there's a function body.
> 
> Cheers,
-- 
Nick Bowler, Elliptic Technologies (http://www.elliptictech.com/)