[XFS SUMMIT] Ugh, Rebasing Sucks!

["Darrick J. Wong" <darrick.wong@oracle.com>]

Hi everyone,

Many of you have complained (both publicly and privately) about the
heavy cost of rebasing your development trees, particularly when you're
getting close to sending a series out for review.  I get it, there have
been a lot of large refactoring patchsets coming in the past few kernel
cycles, and this has caused a lot of treewide churn.  I don't mind
cleanups of things that have been weird and wonky about XFS for years,
but, frankly, rebasing is soul-grinding.

To that end, I propose reducing the frequency of (my own) for-next
pushes to reduce how often people feel compelled to rebase when they're
trying to get a series ready for review.

Specifically, I would like to make an informal for-next push schedule as
follows:

 1 Between -rc1 and -rc4, I'll collect critical bug fixes for the
   merge window that just closed.  These should be small changes, so
   I'll put them out incrementally with the goal of landing everything
   in -rc4, and they shouldn't cause major disruptions for anyone else
   working on a big patchset.  This is more or less what I've been doing
   up till now -- if it's been on the list for > 24h and someone's
   reviewed it, I'll put it in for-next for wider testing.

 2 A day or two after -rc4 drops.  This push is targeted for the next
   merge window.  Coming three weeks after -rc1, I hope this will give
   everyone enough time for a round of rebase, review, and debugging of
   large changesets after -rc1.  IOWs, the majority of patchsets should
   be ready to go in before we get halfway to the next merge window.

 3 Another push a day or two after -rc6 drops.  This will hopefully give
   everyone a second chance to land patchsets that were nearly ready but
   didn't quite make it for -rc4; or other cleanups that would have
   interfered with the first round.  Once this is out, we're more or
   less finished with the big patchsets.

 4 Perhaps another big push a day or two after -rc8 drops?  I'm not keen
   on doing this.  It's not often that the kernel goes beyond -rc6 and I
   find it really stressful when the -rc's drag on but people keep
   sending large new patchsets.  Talk about stumbling around in the
   dark...

 5 Obviously, I wouldn't hold back on critical bug fixes to things that
   are broken in for-next, since the goal is to promote testing, not
   hinder it.

Hopefully this will cut down on the "arrrgh I was almost ready to send
this but then for-next jumped and nggghghghg" feelings. :/

Thoughts?  Flames?

--D