[ANNOUNCE] xorg-server 1.18.2

[Adam Jackson]

On Mon, 2016-03-28 at 17:35 +0900, Michel Dänzer wrote:

> I can't help feeling that this went a bit overboard for a stable
> release. There are many changes in there which seem clearly or at least
> borderline features instead of fixes. If it's taking too long for new
> features to reach users, that would be better addressed by more frequent
> major releases instead.

I agree it was larger than I probably should have done. I did avoid
merging features that I thought were likely to be disruptive or that
would involve ABI changes, but I could have been stricter.

One challenge I face is trying to fill the role of both release manager
and reviewer of last resort. Absent much feedback about the quality of
merged patches, it can be difficult to distinguish "this looks sane
enough for my machine" and "this looks sane enough for everyone's
machine".

> Unfortunately, some of the backported fixes caused problems for users as
> well, though some of those were not regressions as much as a fix
> exposing pre-existing bugs elsewhere. (BTW,
> https://patchwork.freedesktop.org/patch/78155/ is still pending to fix
> such an issue)

Merged:



> Maybe there are some lessons to be learned from the Mesa release
> process, e.g.:
> 
> * There's a streamlined process for nominating changes to be backported
> from master to stable branches (via a specially formatted Cc: tag in the
> commit log which is parsed by a script to generate a list of changes to
> be backported; this scheme was inspired by the Linux kernel). The
> developer making a fix is usually in the best position to judge whether
> the fix should be backported to stable branches or not.
> 
> * There is soaking time (usually at least a week) between backporting
> changes to a stable branch and making a release. In contrast, for the
> xorg-server 1.18.2 release, I was already in the weekend when you first
> talked about backporting changes on IRC, and by the time my weekend was
> over, the release was already out.

I'm happy to adopt the Mesa process, sure. Honestly, I'd be happy if I
got _any_ comments about what might be appropriate for stable, or even
for master. The patch you cite above, for example, doesn't mention that
it addresses a regression (though admittedly the bz does). The number
of patches I've been cc'd on at all after review is probably less than
20. In the absence of that feedback I'm basically trusting my own
judgement. So yes, please do help me do better!

Compare the review stats for recent history for mesa and xserver:

dmt:~% topten() { grep ^....Review | sort | uniq -c | sort -rnk1 | head -10 }
dmt:~% cd git/mesa
dmt:~/git/mesa% git log --oneline 11.2-branchpoint..master | wc -l          
755
dmt:~/git/mesa% git log 11.2-branchpoint..master | topten
     81     Reviewed-by: Marek Olšák <marek.olsak at amd.com>
     68     Reviewed-by: Dave Airlie <airlied at redhat.com>
     53     Reviewed-by: Ilia Mirkin <imirkin at alum.mit.edu>
     41     Reviewed-by: Nicolai Hähnle <nicolai.haehnle at amd.com>
     34     Reviewed-by: Michel Dänzer <michel.daenzer at amd.com>
     33     Reviewed-by: Matt Turner <mattst88 at gmail.com>
     33     Reviewed-by: Jordan Justen <jordan.l.justen at intel.com>
     32     Reviewed-by: Ian Romanick <ian.d.romanick at intel.com>
     32     Reviewed-by: Iago Toral Quiroga <itoral at igalia.com>
     31     Reviewed-by: Kenneth Graunke <kenneth at whitecape.org>
dmt:~% cd ../xserver
dmt:~/git/xserver% git log --oneline xorg-server-1.18.0..master | wc -l        
252
dmt:~/git/xserver% git log xorg-server-1.18.0..master | topten
     79     Reviewed-by: Adam Jackson <ajax at redhat.com>
     26     Reviewed-by: Dave Airlie <airlied at redhat.com>
     25     Reviewed-by: Colin Harrison <colin.harrison at virgin.net>
     20     Reviewed-by: Michel Dänzer <michel.daenzer at amd.com>
     18     Reviewed-by: Kenneth Graunke <kenneth at whitecape.org>
     18     Reviewed-by: Eric Anholt <eric at anholt.net>
     16     Reviewed-by: Keith Packard <keithp at keithp.com>
     12     Reviewed-by: Peter Hutterer <peter.hutterer at who-t.net>
      9     Reviewed-by: Rémi Cardona <remi at gentoo.org>
      6     Reviewed-by: Chris Wilson <chris at chris-wilson.co.uk>

Of those 252 xserver patches I wrote only 34, and only four people
reviewed more than one of those 34. I have another 25 sent but
unreviewed. Patchwork has something like 130 patches open since 1.18
with (approximately, it doesn't quite know series r-bs) 32 reviews
total; of those 32 reviews only eight people reviewed more than one
patch (again, possibly an undercount due to series rbs).

Perhaps, in addition to a more formalized stable process, it would be
useful to get a report of the review status of various proposed patches
on a (say) weekly basis. That would be fairly easy to generate, and
might be a good context for people to ask questions about how reviews
should proceed: whether the concept is reasonable, how it interacts
with the rest of the server, etc.

- ajax