Re: keep in sync with -mm tree?

[Reuben Farrelly <reuben-lkml@reub.net>]

On 8/11/2005 4:00 p.m., Andrew Morton wrote:
> Coywolf Qi Hunt <coywolf@gmail.com> wrote:
>> Hello,
>>
>> We can always keep in sync with the current Linus tree through his git
>> tree. But from where can we keep in sync with the current -mm tree?
>> ie, when somethings added to -mm, how do we get that too?
> 
> You can't.  The patches in -mm spend 90% of their time in an untested,
> often-doesn't-compile state.  It's only in the 24-48 hours preceding a
> release that I actually start build- and run-time testing it all.

Would it be at all useful if a small (and for feedback purposes only possibly 
known) subset of users were to give the -mm releases a basic compile and build 
before a proper -mm is released?  ie in the 24 hours preceding a formal -mm 
release, to shake out the most obvious brown paper bag problems?  I know -mm 
is a testbed but still, it must surely be better if the big bulk of those 
(unknown number x) people who try it don't need to further patch it to get it 
to at least build.  It also allows comments like "EDAC is known to not compile 
on i386 with SMP, we are bugging Alan already about it" to be clearly stated 
at the time it is released.

Personally I seem to have my share of compile, symbol and oopses that are 
fairly obvious and are visible within 5 mins of booting up, and I'd be willing 
to spare 15 mins once every couple of weeks or so to do a primitive regression 
test on this and report back.  I can build on i386 and possibly x86_64 if need be.

Andrew, what do you think?

>> The only way now seems to check the mm-commits list. Is it possible to
>> expose akpm's working folder somewhere for convenience?
> 
> Well I suppose I could upload stuff to
> ftp://ftp.kernel.org/pub/linux/kernel/people/akpm/mm/ daily.  Then it's
> trivial to install mm-of-the-day as a quilt series.
> 
> <does crontab -e>
> 
> Let me know how it goes..

Last time I tried a broken-out- tarball I ended up watching a heap of fuzzy 
apply's then a part of the patch which was entirely rejected, leaving only the 
first half of the broken-out mm patch applied and my tree in a busted half 
patched state.  Not terribly surprising, but at that point I wasn't sure 
whether the feedback was useful or not...  (and sure enough the next day 40 
more patches were applied so I figure it wasn't 8) ).

reuben