Re: [PATCH] Fix const pointer warning in gthr-win32.h

["Aaron W. LaFramboise" <aaronavay62 at aaronwl dot com>]

Kaveh R. Ghazi wrote:
> From: "Ian Lance Taylor" <iant@google.com>
>
> > "Aaron W. LaFramboise" <aaronavay62@aaronwl.com> writes:
> >
> > > OK, I discovered this file is also used in other places outside of GCC
> > > such as in libgfortran, so using tsystem.h is not OK there.  What is
> > > the  proper way to make this error go away then, as I understand the
> > > union trick is the only way to silence it?
>
> Is it possible to fix (i.e. const-ify) the relevant function prototypes, or
> do you end up calling a system interface that's non-const and get stuck?

No, the system interface is a void *.  And I think that is correct; it 
should not be const void *; the GTHR is incorrect in wanting it to be a 
const void *, because __gthread_getspecific is returning void *, and 
this is the same pointer.  If we're returning a non-const pointer from 
TLS, we need to be saving a non-const pointer, otherwise this doesn't 
make any sense.

I would try just taking out the 'const' but I think this will break a 
lot of platforms I have no way of testing.  But I suppose I could try 
anyway and just deal with the breakage.

So I'm not sure what the proper way to resolve this is.  Advice?

> > > By the way, I think this prototype is actually incorrect.  The ptr
> > > parameter really should not be const, because __gthread_getspecific
> > > returns a non-const void *.  However, I'm not sure if there's some
> > > other reason this should be const; does anyone know about this?
> >
> > What code are you compiling when you see this warning?  As far as I
> > know code that uses gthr-win32.h is mostly C code which should be
> > compiled with -Wc++-compat.

It comes from anything that includes gthr-win32.h.

> I think the warning comes from -Wcast-qual, not -Wc++-compat.