[PATCH] dix: Fill in errorValue in ChangeGC for tile/stipple/clip BadMatch

[Daniel Stone]

Hi,

On Tue, Jun 21, 2011 at 09:14:30AM -0700, Alan Coopersmith wrote:
> On 06/21/11 02:52 AM, Daniel Stone wrote:
> > On Mon, Jun 20, 2011 at 06:50:47PM -0400, Adam Jackson wrote:
> >> The caveat here is that unused data in errors isn't guaranteed to be
> >> 0, which means you'd have to know to only look for extra BadMatch
> >> info for either a particular server version or (better) protocol
> >> version.  You could reasonably make that part of protocol 11.1, on
> >> the assumption that 11.0 clients wouldn't bother looking for it, but
> >> that 11.1 clients would want to be assured of it.
> > 
> > Yeah, I was thinking about protocol 11.1, but from
> > ProcEstablishConnection:
> > [...]
> > 
> > So we could bump the server to 11.1 and allow 11.0 clients to connect,
> > but 11.1 clients wouldn't be able to declare that to 11.0 servers.  Ugh.
> 
> Would it be simpler then to do this via an extension?  Perhaps Generic Event 1.1
> could return more detailed errors to clients registered with it?
> 
> Seems like it would be nice to not just return the bad value in the error, but
> some hint as to which field it was when multiple fields may contain the same
> values.

Mm, we'd then need to define GE equivalents for all core errors, and
somehow make clients aware of them.  I think I'm more or less with ajax
that we should do it in a core protocol rev, otherwise it's probably not
worth bothering with tbqh.

Julien also pointed out for the client-connection issue, that if clients
are having to retry on failing r/R vs. l/B anyway, 11.1 clients could
just try reconnecting with 11.0 if they fail connection to a 11.0
server.

Cheers,
Daniel