Hi Andreas, On Sun, Dec 22, 2024 at 07:26:15AM +0100, Andreas Metzler wrote: > On 2024-12-22 Sean Whitton <spwhit...@spwhitton.name> wrote: > > On Sat 21 Dec 2024 at 06:15pm +01, Chris Hofstaedtler wrote: > > >> Uploads to the archive are not irreversible. You can just as well > >> upload a new version reverting whatever was done. > > > Yes, quite. Integers are cheap, and these are not uploads directly to > > stable. We can always just upload again. > > Many changes in a installed package are not trivially/easily revertable. > Think of moving files between pacages (Needs Replaces/Breaks), replaxing > symlinks by dirs and vice versa (needs dpkg-maintscript-helper),
True, there are deficencies in our tooling here. However I kind of wonder if inaction on a maintainer's part causing more work to revert changes isn't actually a virtous mechanism from an organizational perspective. It makes others' contributions "stick", counterbalances the natural tendency to keep the status quo and should ideally motivate maintainers to respond swiftly to avoid future work :-) I fail to see anything actually wrong with that in most cases. If a contributor truly messed something up I would expect they would also have the motivation to clean up after themselves once the error is pointed out by someone. Since IMO learning-by-doing is also one of the more effective ways for people to learn it seems especially worth it to tutor (new) contributors is such cases if needed. > changing dpkg-conffiles (not really undoable). Could you elaborate on what scenario you're thinking of here? Ideally we should endeavor to improve our tools to eliminate common sources of irrevertable changes. Can anyone think of other common cases where a revert is really, actually impossible? (No, postinst doing `rm -rf /` isn't a common case :P) --Daniel
signature.asc
Description: PGP signature
