- Package:
- release.debian.org
- Source:
- release.debian.org
- Submitter:
- Martin-Éric Racine
- Date:
- 2025-08-10 16:33:01 UTC
- Severity:
- normal
- Tags:
Please unblock package dhcpcd [ Reason ] As someone found out during the soft freeze, a daemonized dhcpcd competes with network-manager and other network configuration tools for control over the network interfaces and resolver. Back when I took over maintenance of the package, I ended up spliting the init.d script and systemd unit into a separate bin:dhcpcd package precisely to avoid this. Apparently, that wasn't enough. I therefore just added a "Conflicts: network-manager" to bin:dhcpcd to further drive the point that having two networking daemons on the same host is a bad idea. This is the only change. Meanwhile, bin:dhcpcd-base remains perfectly harmless on hosts running systemd-networkd or network-manager, since it doesn't come with startup scripts and is instead executed via ifupdown or some other network configuration framework as per administrator configuration. [ Impact ] Without this fix, bug reports similar to #1107683 are likely to be filed after Trixie is released. [ Tests ] None needed. It's only a dependency change (Conflicts: network-manager). [ Risks ] No code change. [ Checklist ] [x] all changes are documented in the d/changelog [x] I reviewed all changes and I approve them [x] attach debdiff against the package in testing unblock dhcpcd/1:10.1.0-12
[network-manager maintainer speaking here] Am 14.06.25 um 06:52 schrieb Martin-Éric Racine: In the past, network management systems did not declare conflicts against each other. If what you say is true, and dhcpcd interferes with networkd, NetworkManager, ifupdown and its variants, why was only a Conflicts against the network-manager added? This feels like a very ad-hoc workaround.
su 15.6.2025 klo 18.26 Michael Biebl (biebl@debian.org) kirjoitti: Indeed, but it if even an experienced Debian developer ended up filing a bug wondering why 2 network configuration daemons compete for the same interfaces, it strongly indicates a need for idiot-proofing dependencies. Because, as the bug report that caused this change showed, it's the most obvious case. Ideally, I'd also have a Conflicts against networkd, but it's not a separate package, and it needs to be explicitly enabled, so we'll manage fine. Please also note that the dhcpcd binary itself doesn't compete for control of the interfaces and resolver; it's as passive as dhclient was, hence why dhcpcd-base (it could just as easily been called dhcpcd-bin) remains perfectly harmless even on a host that uses networkd or network-manager. In the later case, IIRC, we even have a netowrk-manager backend that can use the dhcpcd binary? Btw, it's interesting that you phrased this as dhcpcd interfering with others, and not the other way around. Biased, much? Martin-Éric
Hi Martin, Have you thought the implications of this this change all the way through to user impact? What's about bookworm users who currently have both dhcpcd and network-manager installed? Is the observed breakage in #1107683 so bad it's unlikeley there are any? What if some exist and are relying on NM config for conectivity? Would this break their systems on upgrade? I'm honestly not sure how apt tiebreaks with conflicts. Which package would get removed in that case? I'm happy to do some testing with debvm if we agree on this being a concern.
Hi Martin, Have you thought the implications of this this change all the way through to user impact? What's about bookworm users who currently have both dhcpcd and network-manager installed? Is the observed breakage in #1107683 so bad it's unlikeley there are any? What if some exist and are relying on NM config for conectivity? Would this break their systems on upgrade? I'm honestly not sure how apt tiebreaks with conflicts. Which package would get removed in that case? I'm happy to do some testing with debvm if we agree on this being a concern.
Hi Martin-Éric, Daniel, What's the answer here? I'd like to see some reply to the concerns raised. Daniel, if you could try and reproduce the scenario you're worried about that would be appreciated. Paul
to 3.7.2025 klo 20.12 Paul Gevers (elbrus@debian.org) kirjoitti: As already stated, there is no impact. The Conflicts merely is idiot-proofing the dependencies to avoid two networking daemons fighting for control of the same interfaces. I've filed a bug against network-manager to have a Conflicts against bin:dhcpcd to address the same case, in reverse. I'll happily address genuine issues. Daniel's "concerns" are whataboutisms. Martin-Éric
to 3.7.2025 klo 20.27 Martin-Éric Racine (martin-eric.racine@iki.fi) kirjoitti: Sure enough, neither of the two people who expressed concerns ever got around demonstrating the negative impacts they foresee as requested by Paul. Meanwhile, Trixie is now released, making this bug a moot point. Closing. Martin-Éric