[Intel-gfx] [PULL] drm-misc-next

[Daniel Stone]

Hi,

On Tue, 6 Aug 2019 at 10:58, Daniel Vetter <daniel.vetter at ffwll.ch> wrote:
> On Tue, Aug 6, 2019 at 11:55 AM Emil Velikov <emil.l.velikov at gmail.com> wrote:
> > On Tue, 6 Aug 2019 at 10:49, Daniel Vetter <daniel.vetter at ffwll.ch> wrote:
> > > The thing is, dim push shouldn't allow you to do that. And the patches
> > > have clearly been applied with dim apply (or at least you added the
> > > Link), unlike Rob who seems to just have pushed the revert.
> > >
> > Thanks, did not know about dim push. Will make sure I use it.
>
> So the intro doc isn't good enough, and we need to enforce it. I think
> Daniel's idea was to have a pre-merge hook which checks for a git
> variable using --push-option. Can you pls look into this? I guess we'd
> need the dim patch, and example premerge hook to be installed
> server-side. Should have a nice error message too ofc.

Yeah, the docs are already quite clear that you cannot push to the DRM
trees with normal git, and that you have to use dim. Not only does it
check and enforce all the rules in the documentation, but it also
rebuilds drm-tip and keeps other trees in sync, which isn't done with
a regular git push.

The idea I had a few weeks ago was to have dim use 'git push
--push-option fdo.pushedWithDim=this-was-pushed-with-dim-and-not-manually',
then have the hooks on the server side check for that option and
refuse any direct pushes. (Or at least, if people are pushing
directly, they have to _really_ try to be doing it, and can't do it by
accident.)

If someone types up the dim patch to do that and gets it committed,
after a couple of days' grace period for everyone to update I can roll
out the server-side hooks which refuse non-dim pushes.

Cheers,
Daniel