add pristine-tar

[Reinhard Tartler]

On Do, Mai 27, 2010 at 17:50:50 (CEST), Jonas Smedegaard wrote:

> On Thu, May 27, 2010 at 10:35:02AM -0400, Felipe Sateler wrote:
>>On Thu, May 27, 2010 at 10:34, Felipe Sateler <fsateler at gmail.com> wrote:
>>> On Thu, May 27, 2010 at 03:54, Jonas Smedegaard <dr at jones.dk> wrote:
>>>> On Thu, May 27, 2010 at 09:09:15AM +0200, Reinhard Tartler wrote:
>>>>>
>>>>>> Yeah - tests should be composed so that _failure_ triggers alternative
>>>>>> action:
>>>>>>
>>>>>>        test ! -d .git || dh_quilt_unpatch
>>>>>
>>>>> oh right, thanks for pointing this out.
>>>>>
>>>>> Is this a basis for consensus? Shall we agree on this variant for the
>>>>> clean rule?
>>>>
>>>> No.
>>>>
>>>>  a) I believe it has not been verified to work with build daemons
>>>
>>> There has to be patching too in the build rule. dpkg-buildpackage
>>> calls clean before build, so at clean it would unpatch, and at build
>>> it would fail or silently build broken packages. That is why csound
>>> failed to build when it was uploaded with a similar snippet
>>
>>Hmm, soryy, that is not correct (the csound part). The csound snippet
>>was not protected by .git directory testing.
>
> Correct.  When applying the snippet, I changed my mind regarding that: I
> felt it bad to do an exception based on the exisence of a .git dir or
> not: I believe packages should always behave the same, and conditionally
> applying patches using quilt goes against that principle.

I'd really prefer to have the consistency on the developer/user side
than on the buildd side. The unpatching in a git tree establishes the
'compatbility' with non Format 3.0 in our team. I therefore don't think
that we need to require that package *always* have to behave the same as
a principle.

> If possible to apply the patches using dpkg I would be less (but still)
> hesitant.  Yes, dpkg currently apply patches same way as quilt, but
> there is no promise of quilt and dpkg staying both keeping same
> behaviour.

sorry, but this is simply wrong. Both the format name (3.0 (quilt)) and
the development history show that Raphael tries really really hard to
simulate the exact behavior of quilt, even if quilt is not installed in
the build environment.

-- 
Gruesse/greetings,
Reinhard Tartler, KeyID 945348A4