[U-Boot] [PATCH v6 1/7] powerpc: Extract EPAPR_MAGIC constants into processor.h

[Stefan Roese]

Hi Wolfgang,

On 10/30/2012 12:05 PM, Wolfgang Denk wrote:
>> As you know this patch is part of a patch-series. And this is the first
>> time that this patch has a change. So this summary covers the complete
>> history for this patch.
> 
> But exactly this is information which I do not have, and which is not
> included in your patch.  As is, I can only intepret this to be version
> 6 of this specific commit, and I wonder which changes were made in the
> previous 5 versions.

*If* we agree upon a per patch series versioning (see below), then this
would be enough. To only list the changes that have been made to this
patch. Your suggestion from below is even better. To document that no
changes have been made:

	V2: no changes
	...

I'm pretty sure that Simon (or other people with a bit of python
knowledge) can easily add this to patman.

>> In this version of the patch series, I only made this small change to
>> this patch 1/7. I wanted to spare the list a resending of the complete
>> patchset for such a small change.
>>
>> So what is the recommended way to do this? Is it really
>> recommended/required to repost the complete patch series upon a small
>> change in only one patch? No problem, I can do this. patman makes it
>> very easy. :)
>>
>> Should I repost the complete series again?
> 
> No, not at all!

Okay.

> I understand you are using patman for patch management.  So I added
> Simon on Cc: to have his oponion, too.
> 
> I see two options:
> 
> 1) Versioning is done on a per-patch base.  In this case, this patch
>    should have been submitted as "[PATCH v2 1/7]", in which case it
>    would have been clear to everybody that this is the first and only
>    change compared to previous submission(s).
> 
>    I don't dare to say "most", but at least some people have worked
>    like this when submitting patch series (manually) in the past.
>    
>    I can understand if somebody argues that it is not exactly easy to
>    collect the correspondign patches to a series if individual patches
>    contain different version numbers.  Correct threading of the
>    messages is essential here.

Yes, this is my main concern about option a). Very hard to match the
single patches (and its versions) to the patch series version. Without
proper threading. And I personally don't use threading in my mail client
(my problem, I know).

> 2) Versioning is done on a per-series base.
> 
>    One problem here is that it becomes difficult to keep track if
>    what is what when only single patches of the series change and get
>    reposted - on the other hand it has always been a major PITA when
>    people repost whole series after only changing a line of two in on
>    of their many patches, so we strongly encourage posting of only the
>    changed patches.  Once more, proper threading appears to be
>    essential.
> 
>    Another problem is what we are running into here: after severl
>    versions of the patch series one patch that has been untouches
>    previously gets changed.  Now it gets posted as "V6", and it's
>    impossible to know how many previous versions of this patch might
>    have been posted before - one? 2? 3? 4? or 5?
> 
>    When the version ID refers to the patch series rather than to the
>    individual patch, then I think it is mandatory to take this into
>    consideration in the patch history, whih then must refer to all
>    versions of the _series_.  In the present case, the patch history
>    should have looked like this:
> 
> 	V2: no changes
> 	V3: no changes
> 	V4: no changes
> 	V5: no changes
> 	V6: Fix compile warning: release.S:354:0: warning: "EPAPR_MAGIC" redefined
> 
> 
> Is there any clear majority of preferences for patch versioning?
> My gut feeling is that more people would like version IDs on a
> per-series base, but I would like to see some confirmation for this,
> and the we should document such expectations.

As you have already guessed, I'm in favoring the 2nd option, versioning
on a per-series base.

What do other developers have to say? What's the recommended way to do
this in the Linux world? Even if we don't need to do everything in the
same way as done in Linux development, it is much easier to do it in a
similar fashion for users working in both projects (U-Boot & Linux)
regularly.

> It appears that patman is oriented toward using a single version ID
> per series.  Simon - would it be possible to automatically add such
> "no changes" information when generating the patches?

A little motivation: Simon, you could earn yourself another beer the
next time we meet! ;)

Thanks,
Stefan