This page started with a copy of [this email] sent by Andy Green to the kernel list.
A fair amount of "administrivia" has built up, here's the executive summary:
~ 1) *-tracking branches in git have a history now ~ 2) pending-tracking created ~ 3) how it is decided what goes on which branch ~ 4) all patches end up in stable-tracking ~ 5) we will fork the new stable from andy-tracking ~ 6) target andy-tracking for patches
Rest of the email is explaining how or why.
1) *-tracking branches in git have a history now
The last few days I have been using a new system in git where I hide the hard rebases that are happening on my local tree for the -tracking branches and instead make "MERGE" commits on the upstream branches. It seems to be working out so far so I explain what's going on.
Committed patches now only appear directly on the branch they were committed on. So if a patch goes into stable-tracking it will appear there as normal, but what appears in the downstream branches rebasing off that is a single MERGE commit that contains whatever the diff is between the unrebased and rebased version of that branch. Since I rebase very often, it won't be unusual that the MERGE version used downstream is just the one added upstream patch content, but it can also be typical that the MERGE commit will contain the changes from dozens or hundreds of patches when the Linus tree updates or example.
The reason for this new way is that it gives the public -tracking branches a linear history that you can rewind, despite that they are being hard rebased constantly.
2) pending-tracking created
It's become clear we can end up with a bunch of patches from several directions heading into various upstreams if we're lucky. So I have introduced a "pending-tracking" that fits inbetween stable-tracking and andy-tracking, to temporarily hold all the patches that are expected to come from upstream eventually. Balaji-tracking feeds into this now, but there's other stuff like Matt's patches headed for s3c64xx upstream too. ~ This also allows me to publicly rebase Balaji's patches without writing in balaji-tracking, and it also allows me to track changes from balaji-tracking and elsewhere in one place too.
3) How is it decided what goes on which branch
~ - Patch belongs in upstream, eg, Linus or Ben Dooks tree --> pending-tracking. It's kept segregated because we expect it to turn up by upstream rebase path; if it changes on the way we can adapt it in pending-tracking and rebase
~ - Patch bases on stuff in pending-tracking, eg uses Balaji regulator things ---> andy-tracking. It's kept segregated because "it lives in the future" where the pending upstream patches arrived upstream.
~ - Neither of the above ---> stable-tracking
4) All patches end up in stable-tracking.
Stuff temporarily in pending-tracking will eventually turn up from upstream pulls, at that time it's no longer pending and is part of what stable-tracking rebased against. The now duplicate patch in pending-tracking is then removed. The patch ended up in stable-tracking.
Once the pending-tracking patch became part of stable-tracking, there is no reason for the patches modifying or depending on it held until now in andy-tracking not to merge back into stable-tracking either. So patches in andy-tracking gravitate to stable-tracking as soon as possible too.
5) We will fork the new stable from andy-tracking
Because andy-tracking is "living in the future" with latest pending upstream patches included, it makes sense to take the advantage from these even at the risk the patches will be rejected upstream.
6) Target andy-tracking for patches
If you simply target andy-tracking for patches, the resulting patch will automatically be ready for either andy-tracking, pending-tracking (usually) or stable-tracking depending on what it touched.