A modest proposal -- We need a patch penguin

I would be happy to serve as patch penguin, as I plan on collecting all patches anyway in my new duties as maintainer of www.linuxhq.com.

I am currently writing code to scan the usual places for linux patches and automatically add them to our databases. This would be really simplified by having patches sent to us. And, since we already have a functioning site, we have the hardware/network capacity to serve as a limitless queue of waiting patches for Linus. I would love nothing more than to update the site with information as to the status of these patches.

( [email protected] )

Philosophical question: Would you have a major philosophical objection to acting as Dave Jones's secretary and webmaster? (He is the de facto current patch penguin. I'm just asking for the position to be recognized. We need that before we can really move forward with anything else. If you were to queue patches for Linus and then be ignored by Linus, nothing would have been accomplished, and if somebody ELSE then takes your work and integrates it, it would be yet more pressure to fork the tree, pressure which I'm trying to REDUCE here...)

Remember minix? Way way way back? Andrew Tanenbaum had a little kernel, ran on intel hardware, came with complete source code. And he did not accept patches, due to his minix book contract and the resulting licensing issues. Collaborative development on Linux STARTED in the minix newsgroup, largely by recruiting people who were frustrated at trying to get their patches into minix.

Remember GNU? Stalled in the late 80's? For legal reasons, Richard Stallman wanted people to physically sign over their copyrights (on paper he could put in his file cabinet) to any code they submitted to the GNU project. This caused way too much friction (and Richard wasn't exactly a coalition building statesman either), and eventually people got fed up with the project and took their code elsewhere.

These are the kind of pressures that, if they build up high enough, cause projects to fork. It's all different trees with different patches in them, and if the patch pressure builds up too high forking is inevitable. (Re-integration of forks is also quite possible, they can be short lived. But that's the same integration issue, just deferred a bit.)

I'm not saying Linux is in immediate danger of forking, I'm just saying that code integration can be a serious limiting factor, and is a potentially seperable problem from being a code architect. I think an explicit full-time integration maintainer could reduce/buffer the patch pressure, and that this could be good for the project.

Rob

Please don't just accumulate stuff.

It would be useful to know which of the patches still applies against the most recent 2.2, 2.4 or 2.5 kernel, so each patch gets some status fields:

1) applies against 2.2
2) applies against 2.4
3) applies against 2.5

4) was applied to 2.2
5) was applied to 2.4
6) was applied to 2.5

7) bitrotted patch, no longer applies and wasn't
applied ... moved to 'old' queue

kind regards,

Rik

"Linux holds advantages over the single-vendor commercial OS"
    -- Microsoft's "Competing with Linux" document

http://www.surriel.com/		http://distro.conectiva.com/

Right. Accepting any patch is wrong policy. You'll be swamped. Patch must be marked "applies to 2.N.M", patch tracking system must check that automagically.

Also each patch(set) can be commented by general public and by maintainers. If there is _no_ comment from any of _maintainers_ (i.e. it is not reviewed or found too ugly to worth commenting) it is automatically dropped from the system after some time. This will force patch authors to care about code quality.

If patch is too old (several releases behind) system can mail author(s):
"Warning. Your patchset #3476346 needs rediffing. It will be dropped otherwise"

These "small" details determine whether system is useful or just turns into huge pile of patches of questionable value.
--
vda

Public patch tracking system/queue, maybe something derived from bugzilla.

(i) patches are sent to the maintainer and entered into the system.

(ii) reviewed patches are update appropriately, eg. ( "reject - untidy, please fix", "accept - expected version 2.4.18pre19" etc. )

(iii) patch versions, updates can be kept, as in mozilla's bugzilla site. And comments on that patch can also be kept right along side the code.

Regardless of wether the current system is changed or not, the linux kernel would benefit from a central, searchable, public repository of patches.

The code is available, bugzilla has all this functionality today.

So here's hoping for a patchzilla.kernel.org :)

--Kervin

http://linuxquestions.org/ - Ask linux questions, give linux help.

Followup to:  <[email protected]>
By author:    Kervin Pierre <[email protected]>
In newsgroup: linux.dev.kernel

If Linus et al signs on to the idea, I'm sure we can build it...

-hpa

<[email protected]> at work, <[email protected]> in private!
"Unix gives you enough rope to shoot yourself in the foot."
http://www.zytor.com/~hpa/puzzle.txt	<[email protected]>

Hi John,

How about extracting patches from lkml with procmail?

---cut here-->8---
:0 :
* ^sender: [email protected]
* ^subject:.*patch
{
:0 Bc:
* ^--- .*/
* ^+++ .*/
linux-kernel-patches
}
---8<--cut here---

This recipe has its limits, but it's a start.

Regards, Olaf.

Actually I was sort of thinking that maybe part of the problem with our current system is the noise-to-signal ratio of lkml itself.

Perhaps it's time we set up a specific lkml-patch mailing list, and leave lkml for discussions about the problems. Have a script that posts general details about patches on lkml when there is a post to lkml-patch if you like, so people know and can go and take a look if they want. If you get complex, it can vet the patches to see if they apply, before pushing them to the list. It also goes well with some sort of patch tracking system (who says we can't use a mailing list as a distribution mechanism), if that gets the go ahead, while not requiring it.

Another possibility (or could even be combined) is that perhaps we need to start separating the mailing list at the code tree level.

eg: The "development" tree (lkml-dev which would currently contain 2.5.x) from the "stable" tree (lkml-stable which would currently contain 2.4.x) from the "older" trees (lkml-old which would currently contain 2.2.x/2.0.x), at the mailing list level.

That way, people can concentrate on a specific tree (eg: Linus could concentrate on 2.5.x), without getting inundated with all the other stuff. This progresses easily when the next "stable" branch hits, so that the "dev" list can keep talking about what they plan to do while waiting for the stable to fork into the new development tree, and the previous stable joins the ranks of the "old" kernels, where it might possibly still get the occasional fix.

By reducing the noise (and hey, there is a reason people black-list certain subjects on lkml apart from personal/flame war issues), people can concentrate on the facts. The less noise (the less traffic?) the more likely every message will be read, patches will be checked, etc. Especially when you have other "duties" apart from maintaining kernel code, it's not always easy keeping up with lkml.

Stuart Young - [email protected]
(aka Cefiar) - [email protected]

[All opinions expressed in the above message are my] [own and not necessarily the views of my employer..]

well since most patches have a subject line starting with [PATCH] its not hard to pull them out with procmail.

I like the suggestion (most recently, of Daniel? pardon if I miscredit) of having patches-2.[45]@vger.kernel.org type addresses, which would archive patches, and have a high noise-to-signal ratio. Maybe even filter out all non-patches.

The big issue I cannot decide upon is whether standard e-mails should be

	To: torvalds@
	CC: patches-2.4@
or just
	To: patches-2.4@

(I'm guessing Linus would prefer the first, but who knows)

Also, something noone has mentioned is out-of-band patches. Security fixes and other patches which for various reasons go straight to Linus.

Jeff

Perhaps it'd be easier for patches-2.4 to actually send a copy to whoever is the relevant maintainer of a "section" (which could be worked out from the path in the patch, as long as it's made relevant to linux/) as well as the 2.4 maintainer? There is a lot of things that can be done here.

Perhaps that is a good use for my lkml-patches idea, which gives those who have no avenue a place to post patches so they get picked up.

Something that does need to be done is that various directories under the kernel tree need to have someone "who receives patches" for that part, and who forwards them onto the kernel maintainer (eg: Linus, Marcello, etc) for further review/inclusion/rejection. This way, anything that doesn't fall under a particular maintainer gets sectioned off to someone, so it does get review, and hopefully a reply.

Stuart Young - [email protected]
(aka Cefiar) - [email protected]

[All opinions expressed in the above message are my] [own and not necessarily the views of my employer..]

I'd say: cc Linus specifically if you think it's something he'd find personally interesting. Leave out the cc if it's a minor bugfix or maintainance.

Oh, as somebody suggested in this thread, there is a difference in priority between bugfixes and other kinds of patches. Should buxfixes go to [email protected] with [BUGFIX] in the subject, or would [email protected] be a better idea?

Out-of-band patches are not going to stop. The difference is, they will be duly noticed after the fact because they should be relatively few in comparison to in-band patches.

Another kind of out-of-band patch is where Linus takes the basic idea from somebody's patch and completely rewrites it, or does some hacking on his own, which he's been known to do. Somehow I wouldn't expect he'd bother emailing the results to himself.

Daniel

Complete with "Patches only against 2.4.17 through 2.4.19", "Doesn't compile con ARM with CONFIG_FOO", "Works fine on AXP"?

Plus search capability: Which files does it touch? What functions/variables change/appear/dissapear? Etc?

Looks like a _HUGE_ ammount of work...

Again, as was discussed here: PLEASE do save the complete message. Oh, BTW the following thread might have caveats, fixes, and important comments.

Horst von Brand			     http://counter.li.org # 22616