Re: + wireless-fix-iwlwifi-unify-init-driver-flow.patch added to -mm tree

Re: + wireless-fix-iwlwifi-unify-init-driver-flow.patch added to -mm tree

[Andrew Morton]

On Tue, 13 May 2008 21:39:27 -0700 (PDT) David Miller <davem@davemloft.net> wrote:

> From: akpm@linux-foundation.org
> Date: Tue, 13 May 2008 21:05:50 -0700
> 
> > Subject: wireless: fix "iwlwifi: unify init driver flow"
> > From: Andrew Morton <akpm@linux-foundation.org>
> > 
> > drivers/net/wireless/iwlwifi/iwl-core.c: In function 'iwlcore_init_geos':
> > drivers/net/wireless/iwlwifi/iwl-core.c:323: error: implicit declaration of function 'iwlcore_init_ht_hw_capab'
> > 
> > This (or something like it) should be folded into the base patch to avoid
> > breaking bisection, please.
> > 
> > Cc: Ron Rindjunsky <ron.rindjunsky@intel.com>
> > Cc: Tomas Winkler <tomas.winkler@intel.com>
> > Cc: John W. Linville <linville@tuxdriver.com>
> > Cc: Stephen Rothwell <sfr@canb.auug.org.au>
> > Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
> 
> Andrew, you have to figure out what we're supposed to do here.

Well someone does ;)

> If I "fold in the patch" to avoid breaking bisection, I have to
> completely rebase my tree screwing up everyone of my downstream
> developers.
> 
> Or is this some patch I'm supposed to remember to fold in several
> months from now, to some random changeset out of thousands, when the
> merge window opens?
> 
> Neither option is tenable, and the headaches of neither are
> worth it purely for the sake of bisection.
> 
> My solution to the bisection problem is to wait a day before pushing
> out usually, it's a best effort thing.  I do as many sanity builds as
> I can, and we also hope that someone during that day might solve the
> problem independantly and post a fix.  That way I can fix it in my
> tree locally before the tree goes public.
> 
> And I think this is the most reasonable approach.
> 
> Once I push something to my public tree, quite frankly, it's the real
> deal, it's staying there, and it's a part of the permanent record.
> And therefore, we'll put fixes on top.

What goes into Linus's tree is there for ever and I do think that all
the short-term things we do should be built around making the permanent
record as good as possible.

This is a(nother) case where a toolchain/process problem is forcing us
to do something which we don't want to do.  In an ideal world we should
tell the git developers "we want x, please" and hopefully they can give
it to us.  Because right now, we're having to work around shortcomings
in git and we are producing a lesser product as a result of this.  A tool
should follow the way in which humans want to work, not vice versa.

Short-term...  dunno.  Perhaps you could have a two-weekly
broadly-announced rebase in which you integrate all these dribs and
drags back into their proper place?  Commit them with some well-known
identifier in the title so that they can all be located when that time
comes?

If you announce such a rebase a day or so beforehand then all the guys
who feed into you could get their stuff merged up into your tree to
minimise their pain when the rebase happens, perhaps.

(That being said, this particular no-compile isn't a huge problem - it
can be worked around with a Kconfig change.  But that isn't generally
the case)

Re: + wireless-fix-iwlwifi-unify-init-driver-flow.patch added to -mm tree

[David Miller]

From: Andrew Morton <akpm@linux-foundation.org>
Date: Tue, 13 May 2008 21:57:37 -0700

> This is a(nother) case where a toolchain/process problem is forcing us
> to do something which we don't want to do.  In an ideal world we should
> tell the git developers "we want x, please" and hopefully they can give
> it to us.  Because right now, we're having to work around shortcomings
> in git and we are producing a lesser product as a result of this.  A tool
> should follow the way in which humans want to work, not vice versa.

This has beaten like a dead horse a thousand times.  Bringing
it up again isn't likely to cause further progress. :)

> Short-term...  dunno.  Perhaps you could have a two-weekly
> broadly-announced rebase in which you integrate all these dribs and
> drags back into their proper place?  Commit them with some well-known
> identifier in the title so that they can all be located when that time
> comes?
> 
> If you announce such a rebase a day or so beforehand then all the guys
> who feed into you could get their stuff merged up into your tree to
> minimise their pain when the rebase happens, perhaps.

This, along with the idea of taking care of all of the "dribs" right
before the real merge, is error prone.

It means I have to play with large collections of patches all at one
time.  The reason I use GIT is because I'm stupid and make mistakes,
therefore I don't like playing with patches.

I used to play this game, it's a lot of work and it sucks.  One
"drib" can require fixing up 200 patches down the chain.  And
I've had this happen to me all the time in the past when I was
rebasing all the time.

Not this specific case, mind you, but it is a real concern in general.

Re: + wireless-fix-iwlwifi-unify-init-driver-flow.patch added to -mm tree

[John W. Linville]

On Tue, May 13, 2008 at 10:15:29PM -0700, David Miller wrote:

> I used to play this game, it's a lot of work and it sucks.  One
> "drib" can require fixing up 200 patches down the chain.  And
> I've had this happen to me all the time in the past when I was
> rebasing all the time.

I have to agree with Dave.

Moreover, I used to get regular complaints about the old "regular
rebase" process.  We switched to a "pull and merge" process for 2.6.25,
and in that period nearly all of the process-related complaints
disappeared for me.

To some degree this is a "pick your poison" issue, and for most people
rebasing seems like the deadlier poison.

John
-- 
John W. Linville
linville@tuxdriver.com

Re: + wireless-fix-iwlwifi-unify-init-driver-flow.patch added to -mm tree

[Andrew Morton]

On Wed, 14 May 2008 08:34:32 -0400 "John W. Linville" <linville@tuxdriver.com> wrote:

> On Tue, May 13, 2008 at 10:15:29PM -0700, David Miller wrote:
> 
> > I used to play this game, it's a lot of work and it sucks.  One
> > "drib" can require fixing up 200 patches down the chain.  And
> > I've had this happen to me all the time in the past when I was
> > rebasing all the time.
> 
> I have to agree with Dave.
> 
> Moreover, I used to get regular complaints about the old "regular
> rebase" process.  We switched to a "pull and merge" process for 2.6.25,
> and in that period nearly all of the process-related complaints
> disappeared for me.
> 

Well.  Have you ever been an hour and a half into a bisection at 3AM
then hit a massive oops deep in the TCP code which was spread across a
large number of commits?  I have and it wasn't fun.  iirc I gave up and
went to bed.

> To some degree this is a "pick your poison" issue, and for most people
> rebasing seems like the deadlier poison.

Well yes.  We'd like the best of both worlds, only we cannot have it. 
And the sole _reason_ we cannot have it is due to restrictions in git
<stimulate, stimulate>.