Re: stable commit notes

Re: stable commit notes

[Willy Tarreau]

Hi Sasha,

On Tue, Aug 23, 2016 at 01:54:40PM -0400, Levin, Alexander wrote:
> I've decided to automate the bits I had for tracking stable mailing list
> discussion and wrap it into the git notes interface to make it easy for
> stable maintainers to track discussions related to stable commits.

I actually think it's a good idea. I already use notes locally when I
know I have to take care of certain commits, or when I want to mark that
I already tested them. I don't know how practical this will be over the
long term but it can definitely help.

> Dave has a very valid point: we don't actually investigate the history and
> correctness for patches that look sane as much as we should. The volume of
> commits is so big that we might miss comments given by reviewers that we
> should consider as well.

In fact I think that in the ideal case we would propagate fixes from most
recent versions to oldest ones. This normally completely avoids all such
issues. But the reality is different as the older kernels we're maintaining
don't have the same release cycles so it's hard to expect that your 4.1
fetches from 4.4, then feeds your 3.18 which then feeds Ben's 3.16, then
Greg's 3.14, then Jiri's 3.12, then my 3.10 etc... So the reality is that
we're re-doing some part of the work on our respective sides and checking
in other branches if we find anything relevant. Thus I pick from 3.14, and
check for the equivalent 3.12 patches in my mbox to see if I notice any
particular comment regarding the backport. That obviously doesn't mean it's
riskless, just that the risk is reasonably low. At worst I'll pick a bug
and its fix like I did this time without noticing it.

My scripts do look for the Fixes tag, indicate whether the patch in question
is in my branch or not, and carry the mainline commit ID so that it's easy
to look for a "Fixes:" in mainline referencing it and grep for it in other
branches. Thus I think that it should integrate nicely with your notes
since I could easily check for each individual patch if there are some
particular notes and for example refuse to merge it to force to read the
notes.

(...)
> To verify it works, try looking at the original commit that Dave's rant was
> all about and make sure you see the mailing list correspondence:
> 
> 	git log b1438f477934f5a4d5a44df26f3079a7575d5946
> 
> Updating is just a matter of fetching again. I'll make sure to push updates
> on a daily basis.

I'll definitely give it a try ASAP. If you can add a link to the
discussion thread based on the message ID, it would be really cool!

Cheers,
Willy

Re: stable commit notes

[Willy Tarreau]

On Wed, Aug 24, 2016 at 09:15:26AM -0400, Levin, Alexander wrote:
> On Wed, Aug 24, 2016 at 02:34:07AM -0400, Willy Tarreau wrote:
> > Hi Sasha,
> > 
> > On Tue, Aug 23, 2016 at 01:54:40PM -0400, Levin, Alexander wrote:
> > > I've decided to automate the bits I had for tracking stable mailing list
> > > discussion and wrap it into the git notes interface to make it easy for
> > > stable maintainers to track discussions related to stable commits.
> > 
> > I actually think it's a good idea. I already use notes locally when I
> > know I have to take care of certain commits, or when I want to mark that
> > I already tested them. I don't know how practical this will be over the
> > long term but it can definitely help.
> 
> What's the concern about long term? My view is that if this will encourage
> reviews in the long term.

I'm not saying I have a concern, all I'm saying is that ideas which sound
good initially have to prove good over time, that's all. You can ask Greg,
the stable maintenance process has evolved quite a bit over the last 8 years
(and that's for good).

> > In fact I think that in the ideal case we would propagate fixes from most
> > recent versions to oldest ones. This normally completely avoids all such
> > issues. But the reality is different as the older kernels we're maintaining
> > don't have the same release cycles so it's hard to expect that your 4.1
> > fetches from 4.4, then feeds your 3.18 which then feeds Ben's 3.16, then
> > Greg's 3.14, then Jiri's 3.12, then my 3.10 etc... So the reality is that
> > we're re-doing some part of the work on our respective sides and checking
> > in other branches if we find anything relevant. Thus I pick from 3.14, and
> > check for the equivalent 3.12 patches in my mbox to see if I notice any
> > particular comment regarding the backport. That obviously doesn't mean it's
> > riskless, just that the risk is reasonably low. At worst I'll pick a bug
> > and its fix like I did this time without noticing it.
> 
> Ideally, if the process is automated enough it could work this way. It also
> means that we'll need just 1 maintainer rather than 5 :)

Sure, and we are the ones causing the mess and making Greg's life harder
sometimes.

(...)
> > > Updating is just a matter of fetching again. I'll make sure to push updates
> > > on a daily basis.
> > 
> > I'll definitely give it a try ASAP. If you can add a link to the
> > discussion thread based on the message ID, it would be really cool!
> 
> I can look into that. Right now I get the mails from my offlinemap dir, so
> I'll need to figure out how to do that.

Often it's a matter of matter of just appending the message ID you have
in the Message-ID header (without the brackets) to certain MID services like
mid.gmane.org (dead) or marc.info. For example :

    http://marc.info/?i=1471979939.13300.168.camel@decadent.org.uk

Cheers,
Willy

Re: stable commit notes

[gregkh]

On Wed, Aug 24, 2016 at 03:38:39PM +0200, Willy Tarreau wrote:
> On Wed, Aug 24, 2016 at 09:15:26AM -0400, Levin, Alexander wrote:
> > On Wed, Aug 24, 2016 at 02:34:07AM -0400, Willy Tarreau wrote:
> > > Hi Sasha,
> > > 
> > > On Tue, Aug 23, 2016 at 01:54:40PM -0400, Levin, Alexander wrote:
> > > > I've decided to automate the bits I had for tracking stable mailing list
> > > > discussion and wrap it into the git notes interface to make it easy for
> > > > stable maintainers to track discussions related to stable commits.
> > > 
> > > I actually think it's a good idea. I already use notes locally when I
> > > know I have to take care of certain commits, or when I want to mark that
> > > I already tested them. I don't know how practical this will be over the
> > > long term but it can definitely help.
> > 
> > What's the concern about long term? My view is that if this will encourage
> > reviews in the long term.
> 
> I'm not saying I have a concern, all I'm saying is that ideas which sound
> good initially have to prove good over time, that's all. You can ask Greg,
> the stable maintenance process has evolved quite a bit over the last 8 years
> (and that's for good).

Hah, if only it were "8", it's been 11 from my count.  The "first"
stable kernel I released was 2.6.11.1 back on March 4, 2005.

time flies...

greg k-h

Re: stable commit notes

[Willy Tarreau]

On Wed, Aug 24, 2016 at 04:16:38PM -0400, gregkh@linuxfoundation.org wrote:
> On Wed, Aug 24, 2016 at 03:38:39PM +0200, Willy Tarreau wrote:
> > On Wed, Aug 24, 2016 at 09:15:26AM -0400, Levin, Alexander wrote:
> > > On Wed, Aug 24, 2016 at 02:34:07AM -0400, Willy Tarreau wrote:
> > > > Hi Sasha,
> > > > 
> > > > On Tue, Aug 23, 2016 at 01:54:40PM -0400, Levin, Alexander wrote:
> > > > > I've decided to automate the bits I had for tracking stable mailing list
> > > > > discussion and wrap it into the git notes interface to make it easy for
> > > > > stable maintainers to track discussions related to stable commits.
> > > > 
> > > > I actually think it's a good idea. I already use notes locally when I
> > > > know I have to take care of certain commits, or when I want to mark that
> > > > I already tested them. I don't know how practical this will be over the
> > > > long term but it can definitely help.
> > > 
> > > What's the concern about long term? My view is that if this will encourage
> > > reviews in the long term.
> > 
> > I'm not saying I have a concern, all I'm saying is that ideas which sound
> > good initially have to prove good over time, that's all. You can ask Greg,
> > the stable maintenance process has evolved quite a bit over the last 8 years
> > (and that's for good).
> 
> Hah, if only it were "8", it's been 11 from my count.  The "first"
> stable kernel I released was 2.6.11.1 back on March 4, 2005.

You're right, I only started to count with the first LTS, I must
be a bit tainted :-)

> time flies...

Oh yes, I just noticed that my first "official" 2.4.33.1 was exactly
10 years ago!

Willy