[dpdk-stable] What kind of commits can be backported to help the process

[Nélio Laranjeiro]

On Thu, Feb 16, 2017 at 01:03:03PM +0800, Yuanhan Liu wrote:
> On Wed, Feb 15, 2017 at 11:23:45AM +0100, Nélio Laranjeiro wrote:
> > Hi stable mailing list,
> > 
> > As written in the subject, it is not fully clear on what kind of patches
> > can enter this branch.
> > 
> > Some fixes apply easily on top of other ones which may not be related,
> > the question is on those "unrelated" patches.  Is it acceptable to
> > backport them, and if yes at what point is it acceptable depending on
> > the nature of the patch:
> > 
> >  1. Re-work: just a refactor of some structure, clean-up, ...
> >  2. Behavior: it change the behavior of a part of the code, ...
> >  3. Performances: it impacts performances (positively or negatively).
> >  4. None, the patch must apply by itself.
> >  5. ...
> > 
> > What is the expectation for this branch?
> 
> Here is the typical flow I took to pick commits to a specific stable
> branch:
> 
> - firstly, I will get a list of bug fixing commits, with the help of
>   devtools/git-log-fixes.sh (as well as the "cc: stable at dpdk.org" tag
>   inside the commit log).
> 
>   Those commits fix some bugs in a former releases, thus they will be
>   applied to a specific stable branch.
> 
> - Some of them could be applied cleanly. I will then drop a note to
>   all related people (the author, the reviewer, etc) and stable list,
>   to inform that this commit will be in a specific stable release.
> 
> - And some of them could not be applied cleanly, when conflicts happens
>   (code base could be changed).
> 
>   When that happens, I will try to backport it by myself if the commit
>   is simple enough (say, just few lines of code and the conflicts could
>   be easily fixed).
> 
>   If not, I will stop (to not mess something up because I'm not familar
>   with the code), instead I will then ask the author (and even, the
>   maintainer) to do the backport. And that's how the 'request-backport'
>   email comes.

Thanks for this information, it helps to understand how you do it.

> So to answer your question. The backport should be easy (when one guy
> knows the code enough). If it invovles re-work and changes the behavior
> the commit doesn't have, it basically means it's done wrongly.
> 
> That helps?

Not really, the issue is more related to fixes which have been published
after a re-work of the code, this re-work may have changed internal
API/ABI, structures, ...  Backporting it becomes like fixing the issue
on totally different code inducing several days of work, tests and
validation.

Should those fixes be backported?

Thanks,

-- 
Nélio Laranjeiro
6WIND