On Fri, 20 Jan 2012 15:14:52 +1100 Stephen Rothwell [off-list ref] wrote:

Hi all,

On Thu, 19 Jan 2012 15:58:28 -0800 Linus Torvalds [off-list ref] wrote:

quoted

(Stats for those that like them: 20% arch updates (arm, power, mips,
x86), 60% drivers (networking - wireless in particular, staging,
media, dri, sound, misc - including getting rid of 'struct sysdev'),
and 20% random stuff: filesystems, networking, perf etc)

More stats for the bored:

(I don't count merge commits below and everything is relative to v3.2)

Of the 8899 commits in v3.3-rc1, 6918 were in next-20120106 (the first
-next based on v3.2).  A further 792 commits have the same subject line as
commits in next-20120116 and a further 16 have the same patch-id.

This leaves 1174 commits (13%) in v3.3-rc1 that were not in next-20120106
for some reason (not too bad really, I guess).

That's a lot.  Please name names!

I was impacted by several busted patches which had not appeared in
-next.

Also, I saw numerous patches which were significantly altered during
their trip from -next to mainline, which is cheating.  These showed up
as a massive reject storm when I attempted to git-merge linus-now with
next-from-12-hours-ago.  I went in and checked.  tools/perf was a major
culprit.

Some will clearly be bug
fixes, of course.  Some will be quilt trees (probably rebased before
being sent to Linus).  Some will be patches that depend on work by others.

The quilt trees would have been eliminated by the "commits have the
same subject line" test?


I don't think this is all a huuuge problem - we sort this stuff out
fairly quickly.  But things could be improved a bit.