Re: [Quilt-dev] Dude, you killed my diffstat ;)

[Andreas Gruenbacher]

On Saturday 12 June 2004 05:06, Tom Rini wrote:
> On Sat, Jun 12, 2004 at 03:44:19AM +0200, Andreas Gruenbacher wrote:
> > On Friday 11 June 2004 18:22, Tom Rini wrote:
> > > ... and I been meaning to see what it'd take to add something like that
> > > in.  Is there a reason it's not used by default (additional dependancy)
> > > ?
> >
> > The additional dependency is not the main reason. I really don't want to
> > force stuff like that onto people who have no interest in it.
>
> I see your point.  It's just that when most people learn about diffstat,
> from what I've seen the reaction has been "ooh, neat!".  But, that's not
> really a good reason to change behavior. :)
>
> > Some additional tweaking will be required to make things like the
> > "Signed-off-by" lines survive (for kernel patches; they are expected at
> > the end of patches).
>
> Er, they're expected at the end of the desc, before patches. The bits 
> that preserve existing comments (that's not %diffstat, which is possibly
> a 'good' reason to keep requiring %diffstat in the code) will catch
> Signed-off-by stuff fine. :)

Well, that's much easier to handle. Good.

Actually the approach we had was this: If during a refresh a %patch line was 
found, it was assumed that this line starts the section that contains the 
patch. If a %diffstat line was found, this started the diffstat section which 
was updated as well. Sections extended until the next %<section-name> line or 
end of file. There was a header section before the first %<section-name> line 
as well. Quilt fell back to the current behavior only if no sections were 
found.

So with diffstat (which also requires %patch) your patch would look something 
like this:

        My wonderful patch fixing everything

        Here goes a description that fully described why the patch is necessary,
        where it come from, how to test that the patch actually does what it's
        supposed to do, etc.

        %diffstat
        <diffstat output>

        %patch
        <the actual patch>

An added benefit of this format is its extensibility: You can add additional 
sections that quilt doesn't care about; there was a filter script for 
extracting/updating arbitrary sections, so this can easily be automated. I 
have a few ideas what this can be used for; some of the obvious things are a 
section identifying the %maintainer, a %changelog, or various bits of status 
tracking information.

Cheers,
-- 
Andreas Gruenbacher <address@hidden>
SUSE Labs, SUSE LINUX AG