Re: New source package formats now available
Charles Plessy <firstname.lastname@example.org> writes:
> Le Mon, Nov 30, 2009 at 11:47:04AM +0100, Goswin von Brederlow a Ã©crit :
>> When someone (e.g. an NMUer) does edit an upstream file and builds the
>> package then the source do not contain those changes while the binary
>> will. That is clearly going to cause no end of pains.
>> Building the source package and unpacking it should always result in
>> the same source tree or fail to build in the first place. Simply
>> ignoring changes is too dangerous.
> Sorry Goswin, but I will be politically incorrectâ?¦ Please do not take
> I checked your Debian QA page and it does not seem that you are doing NMUs. On
> the other hand, if you check my QA page, you will see that do regular uploads
> quite frequently. Therefore, when I write that it is not convenient for quilt
> users that dpkg suddenly attempts to take control of the patch management they
> were taking care by themselves, and when you answer that it is a necessary
> feature for people who do NMU, isn't there a strong discrepancy in the
> discussion since I speak from experience and you don't?
Nowhere in my mail do I say that. I made no mention of quilt or dpkg
doing patch management. My argumentation has nothing to do with that
but only with your proposal.
Your suggestion to simply ignore all changes to upstream files makes
it dangerously simple to forget some of them and is totaly
avoidable. I have nothing against your proposed quiltless format but
then dpkg should fail if there are upstream changes. That way you can
not forget to update the patches or add a file to the patches or
forgot to commit to git or whatever is being used.
As for experience you are arguing from the experience of a good
maintainer that is carefull and methodical. I'm arguing from
experience of how badly maintainer manage to screw things up. I've
seen packages being uploaded with a syntax error in debian/rules but
still having all the binary debs build as well. Packages linked
against stable or experimental libraries. and the list goes on and on.
People do screw up. They are people.
> As I already answered (too harshly) to RaphaÃ«l, I and many developers only
> upload packages that were built in a clean chroot. Therefore, I think that your
> statement about ignoring changes being â??too dangerousâ?? is a gross overstatement
> that is not backed by facts.
And many developers do not build in a clean chroot. Also the chroot
being clean or not has no affect on building the source. Unless you
build the source first and then do a clean build starting with
unpacking the source again your "clean" build would still be broken.
Far too many people do not build source and then call pbuilder with
the dsc file for an upload (or equivalent).
And it only takes one to screw up.
> By the way, I started recently (for reasons unrelated to this thread) to
> document in README.Debian how to test the packages I maintain, to facilitate
> uploads by people not familiar with the package if needed, as well as to inform
> the users about what level of testing they should expect. I encourage other
> maintainersâ??especially those concerned with helping â??NMUersâ??â??to do so.
Don't get me wrong. Documenting ways to test your package is
great. But why should I need to test something that dpkg can far
better test automatically with 100% accuracy?