On Wed, 28 May 2025 at 10:33:26 +0200, Julian Andres Klode wrote:

gnupg Depends: gpg (>= 2.4.7-19) overlaps Depends: gpg (<< 2.4.7-19.1~)
- solutions: ['gpg=2.4.7-19']
- eliminated: [] ['gpg-from-sq=0.13.1-3']

Where gpg-from-sq would be considered a valid solution for the
<< depends as it provides 2.2.something, but not the greater
depends and apt may decide to upgrade gnupg by installing gpg-from-sq.

This seems like a good improvement. But:

A complete list of mergeable dependencies is attached.

I'm not sure this merging is helpful where the two dependencies are
virtual packages that happen to be provided by the same real package
(and others), especially in the presence of "|". For example:

gnome-remote-desktop Depends: systemd | systemd-tmpfiles overlaps Depends: systemd | systemd-sysusers
- solutions: ['systemd=257.5-2', 'systemd=257.5-2']
- eliminated: ['systemd-standalone-tmpfiles=257.5-2'] ['systemd-standalone-sysusers=257.5-2', 'opensysusers=0.7.3-4.1']

I believe the intention here was that using g-r-d with systemd is
preferred (by both upstream and Debian's GNOME team), but users of
non-default init systems can swap it out for another -tmpfiles
implementation and another -sysusers implementation, for example systemd-standalone-tmpfiles and systemd-standalone-sysusers, with a
command like one of these:

apt install gnome-remote-desktop systemd-standalone-tmpfiles systemd-standalone-sysusers
apt install gnome-remote-desktop sysv-rc
apt install gnome-remote-desktop systemd-

But does the new merging make this impossible, by eliminating the
alternatives and forcing the preferred version to be taken?

Perhaps disabling this merging when there is a "|" would be helpful? In
this situation the package's maintainer is specifically accepting the possibility that a non-preferred alternative will be chosen.

Or perhaps this merging should be restricted to situations where the
package name being requested is the same, like the "gpg" in the
motivating example?

Similar situations:

libsphinxbase3t64 Depends on any implementation of the liblapack.so.3
ABI, and any implementation of the libblas.so.3 ABI, but the new code
doesn't allow them to come from separate binary packages:

libsphinxbase3t64 Depends: liblapack3 | liblapack.so.3 overlaps Depends: libblas3 | libblas.so.3
- solutions: ['libopenblas0-serial=0.3.29+ds-3', 'libopenblas0-pthread=0.3.29+ds-3', 'libopenblas0-openmp=0.3.29+ds-
3']
- eliminated: ['liblapack3=3.12.1-2', 'liblapack3=3.12.1-2'] ['libblas3=3.12.1-2', 'libblas3=3.12.1-2', 'libblis4-se
rial=1.1-2', 'libblis4-pthread=1.1-2', 'libblis4-openmp=1.1-2']

network-manager-applet Depends on any polkit agent implementation, and Recommends any fdo Notifications implementation, but the new code
disregards solutions that aren't an integrated desktop environment that
happens to provide both sets of functionality from the same binary
package:

network-manager-applet Recommends: notification-daemon overlaps Depends: policykit-1-gnome | polkit-1-auth-agent
- solutions: ['phosh=0.46.0-3', 'gnome-shell=48.1-1', 'gnome-flashback=3.56.0-1', 'cinnamon=6.4.10-1']
- eliminated: ['notification-daemon=3.20.0-5', 'xfce4-notifyd=0.9.7-2', 'python3-jarabe=0.121-1', 'plasma-workspace=4:6.3.4-1', 'notify-osd=0.9.35+20.04.20191129-2', 'mate-notification-daemon=1.26.1-1+b3', 'mako-notifier=1.10.0-1', 'lxqt-notificationd=
2.1.1-1', 'lomiri=0.5.0-1', 'dunst=1.12.2-1', 'awesome=4.3-7+b2'] ['ukui-polkit=1.2.2.2-1.1+b1', 'polkit-kde-agent-1=4:6.3.4-1', 'mate-polkit=1.26.1-4+b1', 'lxpolkit=0.5.6-2', 'lxqt-policykit=2.1.0-1']

(In particular KDE Plasma users would prefer to have plasma-workspace as
their Notifications implementation and polkit-kde-agent-1 as their
polkit agent, both of which are part of a desktop environment that
integrates this functionality, even though KDE happens to have chosen to implement them in different binary packages behind the scenes!)

smcv

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

On Wed, May 28, 2025 at 10:20:34AM +0100, Simon McVittie wrote:

On Wed, 28 May 2025 at 10:33:26 +0200, Julian Andres Klode wrote:

gnupg Depends: gpg (>= 2.4.7-19) overlaps Depends: gpg (<< 2.4.7-19.1~)
- solutions: ['gpg=2.4.7-19']
- eliminated: [] ['gpg-from-sq=0.13.1-3']

Where gpg-from-sq would be considered a valid solution for the
<< depends as it provides 2.2.something, but not the greater
depends and apt may decide to upgrade gnupg by installing gpg-from-sq.

This seems like a good improvement. But:

A complete list of mergeable dependencies is attached.

I'm not sure this merging is helpful where the two dependencies are virtual packages that happen to be provided by the same real package (and others), especially in the presence of "|". For example:

gnome-remote-desktop Depends: systemd | systemd-tmpfiles overlaps Depends: systemd | systemd-sysusers
- solutions: ['systemd=257.5-2', 'systemd=257.5-2']
- eliminated: ['systemd-standalone-tmpfiles=257.5-2'] ['systemd-standalone-sysusers=257.5-2', 'opensysusers=0.7.3-4.1']

I believe the intention here was that using g-r-d with systemd is preferred (by both upstream and Debian's GNOME team), but users of non-default init systems can swap it out for another -tmpfiles implementation and another -sysusers implementation, for example systemd-standalone-tmpfiles and systemd-standalone-sysusers, with a command like one of these:

apt install gnome-remote-desktop systemd-standalone-tmpfiles systemd-standalone-sysusers
apt install gnome-remote-desktop sysv-rc
apt install gnome-remote-desktop systemd-

But does the new merging make this impossible, by eliminating the alternatives and forcing the preferred version to be taken?

Perhaps disabling this merging when there is a "|" would be helpful? In this situation the package's maintainer is specifically accepting the possibility that a non-preferred alternative will be chosen.

Or perhaps this merging should be restricted to situations where the package name being requested is the same, like the "gpg" in the motivating example?

Thank you for your analysis here and on IRC.

This is now addressed in

https://salsa.debian.org/apt-team/apt/-/merge_requests/486/diffs

which restricts the matching to alternative-less dependencies with the
same name.

A future improvement should be to merge this across alternatives too
perhaps, if we have say

foo (>> 1) | bar
foo (<< 2) | bar

i.e. we are encoding "same source version of foo" in a binNMU safe
way (all this trouble is down to binNMUs, more or less); we may
want to express this as

(foo (>> 1) && foo (<< 2)) | bar

But we can't do this right now, our translation layer translates
entire or groups to SAT before merging them - we don't know which
of the solutions in the resulting clause belong to which part of
the or group so we cannot selectively merge.

With an updated analysis script that should do the same this leaves
a handful of merged dependencies in the archive.

elpa-aggressive-indent Recommends: emacs (>= 46.0) overlaps Depends: emacs
- solutions: ['emacs=1:30.1+1-5']
- eliminated: [] ['emacs-pgtk=1:30.1+1-5', 'emacs-nox=1:30.1+1-5', 'emacs-lucid=1:30.1+1-5', 'emacs-gtk=1:30.1+1-5']

elpa-clues-theme Recommends: emacs (>= 46.0) overlaps Depends: emacs
- solutions: ['emacs=1:30.1+1-5']
- eliminated: [] ['emacs-pgtk=1:30.1+1-5', 'emacs-nox=1:30.1+1-5', 'emacs-lucid=1:30.1+1-5', 'emacs-gtk=1:30.1+1-5']

elpa-diminish Recommends: emacs (>= 46.0) overlaps Depends: emacs
- solutions: ['emacs=1:30.1+1-5']
- eliminated: [] ['emacs-pgtk=1:30.1+1-5', 'emacs-nox=1:30.1+1-5', 'emacs-lucid=1:30.1+1-5', 'emacs-gtk=1:30.1+1-5']

elpa-epl Recommends: emacs (>= 46.0) overlaps Depends: emacs
- solutions: ['emacs=1:30.1+1-5']
- eliminated: [] ['emacs-pgtk=1:30.1+1-5', 'emacs-nox=1:30.1+1-5', 'emacs-lucid=1:30.1+1-5', 'emacs-gtk=1:30.1+1-5']

elpa-ert-async Recommends: emacs (>= 46.0) overlaps Depends: emacs
- solutions: ['emacs=1:30.1+1-5']
- eliminated: [] ['emacs-pgtk=1:30.1+1-5', 'emacs-nox=1:30.1+1-5', 'emacs-lucid=1:30.1+1-5', 'emacs-gtk=1:30.1+1-5']

elpa-evil Recommends: emacs (>= 46.0) overlaps Depends: emacs
- solutions: ['emacs=1:30.1+1-5']
- eliminated: [] ['emacs-pgtk=1:30.1+1-5', 'emacs-nox=1:30.1+1-5', 'emacs-lucid=1:30.1+1-5', 'emacs-gtk=1:30.1+1-5']

elpa-evil-paredit Recommends: emacs (>= 46.0) overlaps Depends: emacs
- solutions: ['emacs=1:30.1+1-5']
- eliminated: [] ['emacs-pgtk=1:30.1+1-5', 'emacs-nox=1:30.1+1-5', 'emacs-lucid=1:30.1+1-5', 'emacs-gtk=1:30.1+1-5']

elpa-fsm Recommends: emacs (>= 46.0) overlaps Depends: emacs
- solutions: ['emacs=1:30.1+1-5']
- eliminated: [] ['emacs-pgtk=1:30.1+1-5', 'emacs-nox=1:30.1+1-5', 'emacs-lucid=1:30.1+1-5', 'emacs-gtk=1:30.1+1-5']

elpa-helm-projectile Recommends: emacs (>= 46.0) overlaps Depends: emacs
- solutions: ['emacs=1:30.1+1-5']
- eliminated: [] ['emacs-pgtk=1:30.1+1-5', 'emacs-nox=1:30.1+1-5', 'emacs-lucid=1:30.1+1-5', 'emacs-gtk=1:30.1+1-5']

elpa-ido-vertical-mode Recommends: emacs (>= 46.0) overlaps Depends: emacs
- solutions: ['emacs=1:30.1+1-5']
- eliminated: [] ['emacs-pgtk=1:30.1+1-5', 'emacs-nox=1:30.1+1-5', 'emacs-lucid=1:30.1+1-5', 'emacs-gtk=1:30.1+1-5']

elpa-makey Recommends: emacs (>= 46.0) overlaps Depends: emacs
- solutions: ['emacs=1:30.1+1-5']
- eliminated: [] ['emacs-pgtk=1:30.1+1-5', 'emacs-nox=1:30.1+1-5', 'emacs-lucid=1:30.1+1-5', 'emacs-gtk=1:30.1+1-5']

elpa-parsebib Recommends: emacs (>= 46.0) overlaps Depends: emacs
- solutions: ['emacs=1:30.1+1-5']
- eliminated: [] ['emacs-pgtk=1:30.1+1-5', 'emacs-nox=1:30.1+1-5', 'emacs-lucid=1:30.1+1-5', 'emacs-gtk=1:30.1+1-5']

elpa-pkg-info Recommends: emacs (>= 46.0) overlaps Depends: emacs
- solutions: ['emacs=1:30.1+1-5']
- eliminated: [] ['emacs-pgtk=1:30.1+1-5', 'emacs-nox=1:30.1+1-5', 'emacs-lucid=1:30.1+1-5', 'emacs-gtk=1:30.1+1-5']

elpa-rainbow-delimiters Recommends: emacs (>= 46.0) overlaps Depends: emacs
- solutions: ['emacs=1:30.1+1-5']
- eliminated: [] ['emacs-pgtk=1:30.1+1-5', 'emacs-nox=1:30.1+1-5', 'emacs-lucid=1:30.1+1-5', 'emacs-gtk=1:30.1+1-5']

elpa-undo-tree Recommends: emacs (>= 46.0) overlaps Depends: emacs
- solutions: ['emacs=1:30.1+1-5']
- eliminated: [] ['emacs-pgtk=1:30.1+1-5', 'emacs-nox=1:30.1+1-5', 'emacs-lucid=1:30.1+1-5', 'emacs-gtk=1:30.1+1-5']

elpa-ws-butler Recommends: emacs (>= 46.0) overlaps Depends: emacs
- solutions: ['emacs=1:30.1+1-5']
- eliminated: [] ['emacs-pgtk=1:30.1+1-5', 'emacs-nox=1:30.1+1-5', 'emacs-lucid=1:30.1+1-5', 'emacs-gtk=1:30.1+1-5']

gnupg Depends: gpg (>= 2.4.7-19) overlaps Depends: gpg (<< 2.4.7-19.1~)
- solutions: ['gpg=2.4.7-19']
- eliminated: [] ['gpg-from-sq=0.13.1-3']

gnupg Recommends: gpgv (>= 2.4.7-19) overlaps Recommends: gpgv (<< 2.4.7-19.1~)
- solutions: ['gpgv=2.4.7-19']
- eliminated: [] ['gpgv-from-sq=0.13.1-3']

php-amqplib Depends: php-phpseclib (<< 4~~) overlaps Depends: php-phpseclib (>= 3.0)
- solutions: ['php-phpseclib3=3.0.43-2']
- eliminated: ['php-phpseclib=2.0.48-3'] []


--
debian developer - deb.li/jak | jak-linux.org - free software dev
ubuntu core developer i speak de, en

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)