Re: [Dri-devel] Simplified DRM backwards compatibility scheme

[Keith W. <kei...@ya...>]

On Mon, 15 Oct 2001 11:36, Daryll Strauss wrote:
> On Mon, Oct 15, 2001 at 10:14:33AM -0600, jhartmann wrote:
> > If you have demenstrated that this is the case then we should remove the
> > version system then I guess.  I do want to voice my concerns though by
> > writing out my argument fully though.
>
> Having a version system is safer. If something does require an
> incompatible change, like the security fix in the Radeon, then you have
> some way of invalidating earlier versions. If it turns out that we can
> keep all the versions compatible that's even better and you just don't
> indicate that the version change was an incompatible one.

The radeon fix could be acheived in a backwards compatible way by doing a 
copy in the kernel.  Acually you would have to do this even with a 
version-setting-ioctl:  if the old version is set, you still have to 
implement the functionality in a secure way.

If a new techinque is needed for a fast *and* secure texture upload, then you 
have to add it in a new ioctl.  

But there's nothing gained by having an ioctl to tell the kernel which one 
you are going to use - the kernel needs to implement both versions and it 
doesn't really care which one you use.  

> Therefore the version system just becomes a safety net and costs us
> nothing in runtime overhead. In practice, we want to strongly discourage
> incompatible changes, so it acts like Keith is recommending.

Actually it doesn't acheive anything.  There's no pretending that you don't 
have ioctls that you really do - that's all a versioning scheme can acheieve 
(ignoring the sarea snafu, of course).

We don't want to discourage backwards-incompatible changes, we want to outlaw 
them.

Keith

_________________________________________________________
Do You Yahoo!?
Get your free @yahoo.com address at http://mail.yahoo.com