On Sat, Oct 19, 2024 at 11:14:04PM +0200, Santiago Vila wrote: > I received this report from the BTS, related with the usr-merge stuff.
Thank you for forwarding. > There is also an alternate proposal to fix the problem, by modifying > > https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1085407#10 It is correct that base-files does not provide perform versioned Provides of usr-is-merged and that as a result the real usr-is-merged package cannot be removed in many configurations as a result of versioned dependencies. At the time adding the provides, I carefully considered whether versioned provides made sense. I observed that usr-is-merged contains a pile of Conflicts and that versioned Depends could be used to enforce those Conflicts without duplicating them into other packages. As a result, I considered that adding versioned Provides would require base-files to also copy those Conflicts. It is known that Conflicts in essential packages can have bad effects for upgrades, so I opted for not copying them and - as a result - not adding versioned Provides. At this time, usr-is-merged has two reverse dependencies. One is init-system-helpers (unversioned, it could arguably drop it) and the other is dbus (versioned). This hints that my line of reasoning may not have been aligned with reality well. It also is difficult to judge how this facility may have been used in derivatives. The proposed dbus change has two benefits over modifying base-files here. For one thing, it clarifies the intention of the versioned dependency with dbus maintainers. For another, it avoids the risk of silently dropping Conflicts. I therefore recommend continuing that way and looking back here if that avenue proves unsuccessful. I earlier noticed that the usr-is-merged package could not be removed from an installation containing dbus. Given all the other workarounds and hacks around, I considered the presence of the usr-is-merged package a cosmetic detail. Do note that many of the workarounds use the usr-is-merged package name for setting up protective diversions. As a result, keeping this package has the benefit of avoiding unintentional reuse. I considered that removal of it could be reasonably be deferred into the forky cycle. Let me also use this opportunity to again express thanks to Chris and Michael. Even though I am able to spend significant amounts of time on this transition thanks to Freexian, it is no longer the case that I am the main driver of this transition. Many others and Chris and Michael in particular have expended a lot of effort into it and we have handled many recent decisions as more of a team seeking consensus from one another. I would appreciate if you (Santiago) could extend your trust regarding base-files/usr-merge matters that you place in me (thank you) to Chris and Michael as I am trusting both of them. Helmut
