Re: Warning for unadorned 0 in varargs lists?

[Eddie Kohler <kohler at icir dot org>]

> On Thu, Aug 19, 2004 at 03:58:54PM -0700, Eddie Kohler wrote:
> > Hi all,
> >
> > I habitually use "0" for a null pointer, which causes serious problems
> > on 64-bit machines when passed to a varargs function.
> >
> > So an optional [i.e. no -Wall] GCC warning that complained when an
> > unadorned, uncast "0" was passed to a varargs function would help me a
> > lot, and probably others.
>
> You are implicitly assuming that the 0 is really intended to be a 
> pointer.
> But there's
>
>     printf("FOO = %d", FOO);
>
> where FOO is a macro that is defined to be zero.  Also:
>
>     open("unreadable_file", O_CREAT, 0);

Right (and both cases are more serious than what I had thought of).

> > Would such a warning be considered for inclusion in GCC?  Or do people
> > think it might generate too many false positives?
>
> A warning would make sense if, rather than warning for all varargs
> functions, the user only gets a warning for functions that definitely
> expect pointers: execl, for example.  It might be better to do this 
> with
> something like the format-checking attributes supported for printf-like
> functions: that is, an attribute that says a varargs function's 
> variable
> arguments are all pointers, or something like that.

This would generate the minimum number of false positives, but wouldn't 
catch a bunch of errors, or errors on most user functions.  I'd be 
happy even with a version of the warning that required me to write 
"(int) 0" when I meant the integer zero (in varargs lists only).

Working on a patch for my own use: I don't know how to distinguish the 
tree for "(int) 0" from "0".  Is there any difference?

Thanks,
Eddie