[PATCH 1/2] OS support: fix writeable client vs IgnoreClient behavior

[Aaron Plattner]

On Wed, Jul 07, 2010 at 04:03:18PM -0700, Jesse Barnes wrote:
> On Wed, 7 Jul 2010 15:54:50 -0700
> Aaron Plattner <aplattner at nvidia.com> wrote:
> 
> > On Fri, Jul 02, 2010 at 09:33:54AM -0700, Jesse Barnes wrote:
> > > On Fri, 2 Jul 2010 10:45:52 +1000
> > > Peter Hutterer <peter.hutterer at who-t.net> wrote:
> > > 
> > > > On Wed, Jun 30, 2010 at 10:55:52AM -0700, Jesse Barnes wrote:
> > > > > > > This looks good to me. I assume the client is allocated with calloc so
> > > > > > > that the count starts out at zero?
> > > > > > > 
> > > > > > > Reviewed-by: Keith Packard <keithp at keithp.com>
> > > > > > 
> > > > > > With the memset in InitClient noted:
> > > > > > Reviewed-by: Daniel Stone <daniel at fooishbar.org>
> > > > > 
> > > > > Peter, can you cherry pick this over to the 1.8 branch as well?
> > > > 
> > > > sorry, that came in too late, after I had already tagged and started
> > > > build-testing 1.8.2. Please feel free to push it to server-1.8-branch
> > > > yourself, the branch is open now.
> > > 
> > > Ok just pushed it, thanks.
> > 
> > Sorry, I was behind on email again.  This is an ABI break due to the
> > addition of 'ignoreCount' in the middle of ClientRec, and is therefore not
> > appropriate for server-1.8-branch.
> 
> Arg yeah.  Can we move it to the end or do we have to revert?

Sadly, moving it to the end doesn't look like it'll be enough:

    dix.h:extern _X_EXPORT ClientPtr clients[MAXCLIENTS];

I'd be surprised if modules access the clients array directly (ours
doesn't), so maybe you can get away with it, but it's iffy.

-- Aaron