Re: [1003.1(2008)/Issue 7 0000074]: Pointer Types Problem

[Wojtek Lerch <wojtek@xxxxxxx>]

I don't think it's a good idea to introduce unusual restrictions on what 
the void* pointer can be converted to.  Can I convert it to a "void 
const*"?  What about a "void *const"?  Why would it need to be undefined 
behaviour to convert it to a char*, even though standard C intends to 
allow converting any valid void* pointer to char*?

Also, a few details are missing from the wording in question:

* Do these conversion work with null pointers?  Are they guaranteed to 
turn null pointers into null pointers?  (If not, then one needs to test 
the result of dlsym() before converting it to a function pointer, rather 
than after.)

* Is it possible for the converted void* pointer to compare equal to the 
address to some object?  (It can happen on hardware with separate 
address spaces for code and data -- does POSIX want to support such 
hardware?)

Here's my attempt at putting this into words:

Any valid pointer to a function type can be converted to a pointer to 
void [or maybe also to a character type].  The result can be converted 
back to a function type, producing a value that compares equal to the 
original value.

If the original pointer is a null pointer, then the converted pointer to 
void is a null pointer too.  If the original pointer points to a 
function, then the result is not a null pointer, but it is unspecified 
whether it points to any particular object or not (and therefore the 
results of dereferencing it are undefined), or whether it is correctly 
aligned for any type other than a character type.   The void* pointer 
may happen to compare equal to a pointer to some object or to one past 
the end of some array, but nevertheless is not considered to point to 
either.


Geoff Clare wrote:
> Wojtek Lerch <wojtek@xxxxxx> wrote, on 01 Jul 2009:
>
> > > (0000129) msbrown (manager) - 2009-06-29 02:27
> > > http://austingroupbugs.net/view.php?id=74#c129
> > > --------------------------------------------------------------------
> > > After the sentence ending on line 18890, add the following:
> > >
> > > If a pointer to an object or incomplete type is converted to a
> > > pointer to void and then to a pointer to a function type the
> > > behavior is undefined.
> > I don't think this is right. In C, an assignment or an initialization
> > is specified to always performs a conversion -- even a conversion from
> > a type to the same type is still considered a conversion. Doesn't that
> > mean that the value returned by dlsym() cannot be assigned to anything
> > without invoking undefined behaviour?
> Good point.  I think it needs to say:
>
>     If a pointer to an object or an incomplete type other than void is
>     converted ...
>
>
> > > If a pointer to a function type is converted to a pointer to
> > > void and then to a pointer to an object or an incomplete type the
> > > behaviour is undefined.
> Likewise this should say "... or an incomplete type other than void ..."
>
>
> > Doesn't that imply that it's safe to convert a pointer to function
> > to a pointer to void, as long as the pointer is not subsequently
> > converted to some other pointer type? Is it meant to imply that
> > it's OK for a portable POSIX application to take the address of an
> > arbitrary function and convert it to void* and back?
> That's pretty much exactly what XSH 2.12.3 says, so yes that's the
> intention.  Otherwise we would have worded 2.12.3 completely
> differently (with explicit reference to dlsym()).