Re: [RFC] [patch] drm core internal versioning..

[Ian Romanick <idr@us.ibm.com>]

Dave Airlie wrote:
> An issue raised by DRM people with the new drm core is how to stop users
> shotting themselves in the foot when upgrading drm modules from CVS and
> mixing up cores and drivers...
> 
> This patch (against DRM CVS) proposes a simple internal version that gets
> passed from the module to the core, when built in-kernel, it gets set to
> default DRM_INTERNAL_VERSION_KERNEL, when built in DRM CVS or snapshot, it
> gets DRM_INTERNAL_VERSION_EXTERNAL, a core built in one will refuse to
> load a module build in the other..
> 
> This is quite a simple solution that should stop the most obvious issue,
> it doesn't stop people updating CVS drivers on top of themselves but my
> view is anyone doing this is either following our scripts or knows what
> they are doing...

I guess I don't get it.  We want to prevent a DRM module that requires 
core API N with a drm_core that exports M, M != N, right?  So why not do 
that?  In that model don't you just need DRM_INTERNAL_VERSION?  With the 
setup you propose it seems like we could still have problems.  If a user 
installs a snapshot, then installs just a DRM from a later snapshot, etc.