[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Breaks in lenny



Manoj Srivastava <srivasta@debian.org> writes:
> On Tue, 18 Dec 2007 13:41:17 -0800, Russ Allbery <rra@debian.org> said: 

>> I don't think we really get back the functionality of quilt until
>> we're shipping the repository with the source package.

>         This is an interesting statement. What exactly does quilt get me
>  that a distributed VCS that does not ship  the whole archive history
>  does not have?  What use cases do you have in mind?

I think my statement was too strong.  Mainly I'm disagreeing that having a
Subversion repository somewhere is sufficient.  There are other cases.

The use cases I have in mind are, for example, easily adding additional
local patches to a package and then help with merging those patches
against later upstream changes.  (git or arch does this better than quilt,
but quilt does a fairly good job once you're used to its work flow, and
Subversion is horrible.)  Easily adding a new patch on top of the existing
patches without destroying one's ability to perform the sort of later
merging and work against those existing patches that you need to do to
merge new upstream or the like.  Keeping patches conceptually separate.

If you have access to the VCS repository, you don't need to have it with
the source package, but anywhere we put the VCS repository that isn't the
source package adds additional access requirements.  It may be that
expecting people to be able to get to a public web server isn't much of a
burden these days, so in that sense shipping the history with the package
may not be necessary.  I think it's ideal, but it's not mandatory.  But
putting a Subversion repository on Alioth doesn't really help with any of
that, and certainly isn't a substitute for using quilt.

>         I must be being dense.  People have, in the past, asked me to
>  merge changes from their arch branches of devotee, and I could see the
>  diffs between their branch and various devotee branches I had lying
>  around, and cherry picking and merging from their branch was relatively
>  painless.

>         Are you suggesting that somehow dpatch/quilt would have been
>  even more effective?  How do you quickly get diffs between several
>  parrallel lines of development that I am trying out from a quilt'ed
>  patch set?

No, arch and/or git is even better than quilt at this, with the
corresponding drawback that you have to actually use arch and/or git (what
quilt does is so simple that you can duplicate it by hand with nothing
more than diff and patch available).  What wouldn't be sufficient is
having the package in Subversion somewhere.

I don't disagree with the argument that we should all move towards arch,
git, bzr, or some similar system.  I'm slowly working that direction
myself, as I find the time.  What I do strongly disagree with is the idea
of dropping quilt or dpatch while still using Subversion or CVS, which
simply cannot do the things that quilt and dpatch do.

-- 
Russ Allbery (rra@debian.org)               <http://www.eyrie.org/~eagle/>


Reply to: