2005-11-08 02:35:50

by Coywolf Qi Hunt

[permalink] [raw]
Subject: keep in sync with -mm tree?

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?

The only way now seems to check the mm-commits list. Is it possible to
expose akpm's working folder somewhere for convenience?
--
Coywolf Qi Hunt
http://sosdg.org/~coywolf/


2005-11-08 02:50:47

by Andrew Morton

[permalink] [raw]
Subject: Re: keep in sync with -mm tree?

Coywolf Qi Hunt <[email protected]> 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.

> 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..

2005-11-08 03:15:29

by Paul Jackson

[permalink] [raw]
Subject: Re: keep in sync with -mm tree?

> ftp://ftp.kernel.org/pub/linux/kernel/people/akpm/mm/

Cool - thanks.

--
I won't rest till it's the best ...
Programmer, Linux Scalability
Paul Jackson <[email protected]> 1.925.600.0401

2005-11-08 08:26:19

by Reuben Farrelly

[permalink] [raw]
Subject: Re: keep in sync with -mm tree?

On 8/11/2005 4:00 p.m., Andrew Morton wrote:
> Coywolf Qi Hunt <[email protected]> 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

2005-11-08 08:39:47

by Andrew Morton

[permalink] [raw]
Subject: Re: keep in sync with -mm tree?

Reuben Farrelly <[email protected]> wrote:
>

Please retain the cc line.

> On 8/11/2005 4:00 p.m., Andrew Morton wrote:
> > Coywolf Qi Hunt <[email protected]> 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?

Maybe. I'll take a look at uploading a nightly rollup as well.

> >> 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.

That would be atypicel ;)

2005-11-09 00:27:56

by Paul Jackson

[permalink] [raw]
Subject: Re: keep in sync with -mm tree?

reuben wrote:
> 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,

Usually when I see that, it is because I did not religiously get
the exact right 2.6.*-rc* version as a base for the broken-out
patch set.


--
I won't rest till it's the best ...
Programmer, Linux Scalability
Paul Jackson <[email protected]> 1.925.600.0401

2005-12-02 03:22:54

by Coywolf Qi Hunt

[permalink] [raw]
Subject: Re: keep in sync with -mm tree?

2005/11/8, Paul Jackson <[email protected]>:
> > ftp://ftp.kernel.org/pub/linux/kernel/people/akpm/mm/
>
> Cool - thanks.

I find that it is something being there when we don't need, and
missing when we do need. I suggest to let users pull from instead of
akpm push it there. It's a script invoked by users.
--
Coywolf Qi Hunt
http://sosdg.org/~coywolf/