Hello, On Tue 29 Mar 2022 at 09:06PM +02, Lucas Nussbaum wrote: > On 28/03/22 at 16:21 -0700, Sean Whitton wrote: >> >> I don't think it's the preferred method. I believe most of the project >> sees git histories are the preferred tool for achieving the goal you >> state. >> >> If we had only source packages for this purpose, then yes, 3.0 (quilt) >> plus DEP-3 would be preferred. But we have git, too, and indeed dgit. >> Repositories for packages contain both upstream history and Debian >> packaging history, and we have powerful git subcommands for extracting >> information. In many cases there is a good argument to be made that the >> git history is part of the preferred form of modification. > > I think there are three different use cases to consider: I agree that we should consider them separately. > A. preferred form for regular contributions to the package (typically > by the package maintainer) > > B. preferred form for occasional contributions to the package (typically > by an NMUer, or by the security team) > > C. preferred form for reviewing the packaging and Debian-specific > changes. > > > I fully agree that the git repository is the preferred form for A. > However, for B and C, I think that our lingua franca is the source > package, and thus a source package that doesn't make it hard to > understand things such as Debian-specific patches. Of course that > could change if we were able to standardize on a git workflow (or > a small set of git workflows), but I don't see this happenning anytime > soon. What you get from 'dgit clone' is designed to serve (B) and (C) well, and somewhat sidesteps precisely the problems created by our having multiple git workflows. Please consider trying out dgit-user(7) and/or dgit-nmu-simple(7) next time you need to engage in (B) or (C). It's really very nice :) -- Sean Whitton
Attachment:
signature.asc
Description: PGP signature