Le 05/09/2019 à 08:42, Mattia Rizzolo a écrit : >> When recompiling all packages that use OCaml with OCaml 4.08.0, >> diffoscope failed with: >> >>> debian/rules:16: *** Version in debian/changelog does not match diffoscope >>> version. Stop. >> >> This is because my scripts generate a new changelog entry documenting >> the recompilation, with a suffixed version, but otherwise does not >> change the source package. > > OOI, what kind of suffix do you use?
I used +o1 this time, but I also intend to use +o2, +o3, ... to distinguish between different builds. >> I believe this kind of changes-only-in-debian/changelog should be >> supported by all Debian packages. > > I disagree with this generalization. Well... it's not policy so I guess you're free to disagree... And that's why I've set severity to wishlist. >> Maybe you could relax the version >> check so that the version in diffoscope/__init__.py is only a prefix >> of the version in debian/changelog? > > So, the check is: > ifneq ($(DEB_DISTRIBUTION),UNRELEASED) > ifneq (,$(filter %-backports,$(DEB_DISTRIBUTION))) > ifeq ($(DEB_VERSION:~bpo10+1=),$(PY_VERSION)) > $(info Building a Debian backport, accepting version mismatch > $(PY_VERSION) <-> $(DEB_VERSION)) > else > $(error Version in debian/changelog does not match diffoscope version or > backport policy) > endif > else > $(error Version in debian/changelog does not match diffoscope version) > endif > endif > i.e. if you use UNRELEASED you can use whatever version you want to. Oh. I didn't pay attention to that (the UNRELEASED stuff). However, I use "unstable-ocaml" as distribution for my generated changelog entries. > I think in the past we did accidentally release with a mismatching > version, and I really don't want it to happen again, so I'm somewhat > worried about relaxing the check to "just allow all suffixes". I see this is a native package. Why don't you use the version in debian/changelog to generate diffoscope/__init__.py? If this was not a native package, or if the version number would otherwise naturally split in two parts, I would suggest to compare only the first part (upstream version/major version/whatever). > That said, it has also been annoying for ubuntu, where every time they > have to to do a change they also have to tweak this check (usually it's > me so it's not that much of a hassle, but I can see other developers > being annoyed at it). I was thinking about Ubuntu, and binNMUs (I know they don't apply to arch:all packages but this kind of behaviour doesn't help). Cheers, -- Stéphane
