Re: the " 'official' point of view" expressed by kernelnewbies.org regarding reiser4 inclusion

[Olivier Galibert]

On Mon, Jul 24, 2006 at 09:39:39AM -0400, Theodore Tso wrote:
> On Mon, Jul 24, 2006 at 01:35:34PM +0200, Olivier Galibert wrote:
> >Ext patches don't get reviewed much
> > outside of the developpers, and they go in pretty much without
> > discussion in any case, except when Linus blows a fuse.
> 
> Um, you're kidding, right?  We certainly don't make the assumption
> that we can violate CodingStyle willy nilly and stuff in yacc grammers
> into ext3 and assume that no one will push back.

I'm not kidding.  I do recognise that the ext* maintainers do a very
good and clean job.


> In fact we did a lot of work to make sure the patches were clean and
> mostly ready to be accepted to mainline even before we made the first
> proposal to push extents to LKML.

I'm no talking about extends only.  Ext3 now is very, very different
than the ext2+journal it was at the start, with backwards-incompatible
format changes added all the time[1].  These changes went in
no-questions-asked.


> > I think there is something of a problem currently, tough.  It is
> > getting too hard to get code in if you're not a maintainer for an
> > existing subsystem (reiser4, suspend2...), and too easy when you're a
> > maintainer (ext4, uswsusp...).  
> 
> It's not fair to assume that the only reason why non-maintainers have
> a harder time getting changes is because their changes are getting
> more intensive review.

"only", no, definitively not.  The impact is non-negligible though.


> (Although it is the case that we probably do
> need to get better at reviewing changes that go in via git trees.)

Ohh yes, a lot better.  Just look at ALSA, most of SNDRV_HWDEP* should
never have gotten in in the first place, especially some recent ones
like SNDRV_HWDEP_IFACE_SB_RC, and even some (SNDRV_HWDEP_IFACE_USX2Y*)
are security holes the size of Cleveland.  I need to continue
documenting the Alsa kernel interface, but I need a new bucket, the
first one is overflowing.


> A much more important effect is that non-maintainers aren't familiar
> with coding and patch submission guidelines.  For example, in
> suspend2, Nigel first tried with patches that were too monolithic,
> and then his next series was too broken down such that it was too
> hard to review (and "git bisect" wouldn't work).

All his submissions since 2004 or so?  It's a little easy to limit
oneself to the last two ones.


> And of course, there are people who assume that the rules shouldn't
> apply to their filesystem...

It may be a little hard to remove XFS at that point though...  And,
while it's not a filesystem, I'd love to be pointed to the technical
discussion deciding whether uswsusp is a good idea.

  OG.

[1] All optional, I know.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/