adept-updater reports many pkgs DEFECT

Bug #181891 reported by Torsten Eichstädt
2
Affects Status Importance Assigned to Milestone
GLibC
New
Undecided
Unassigned
synaptic
New
Undecided
Unassigned
adept (Ubuntu)
Invalid
Undecided
Unassigned
apt (Debian)
In Progress
Undecided
Torsten Eichstädt

Bug Description

Binary package hint: adept-updater

WORKAROUND: Don't let the upgrade run in such a case; wait and try again later.

Today, adept-updater presented me a long list of DEFECT packages several times.
The action suggested was to remove them. The list included many essential packages, so I closed the updater each time. Imagine what for instance my mother would have done...
Contrary to the list of broken pkgs, the status bar stated 3 pkgs upgradable (I'm sure with this), 0 to download (I guess, can't reproduce, see below), and - (minus) about one GB after the upgrade.
Magic, magic, on the next start of the updater, it showed only 4 pkgs upgradable, and the upgrade succeeded.

Revision history for this message
Torsten Eichstädt (torsten-eichstaedt) wrote :

1.: It is imperative to either
    - set root's locale to 'C' and to change 'sudo' to implicitely include root's login profile when a user switches to the root role, even when not called as 'sudo -'. Don't forget to update the docs ;).
    - or switch to the 'C' locale inside various system tools, and switch back to the original locale when printing messages.
The former is a workaround, but much less work than the latter. Bugs in l18n will be found slower, but IMHO system maintenance is more important.
Rationale: Either
    - The localization of basic libraries is not yet finished; they are still containing too many bugs, probably
      "between the lines", i.e. hard to find.
    - Or many basic system routines implicitely rely on 7-bit ASCII string ordering (again, "between the lines").
See the examples below.

2.: Could you please start a project that runs tests at least on the basic (minimal) system. Rationale: Ubuntu (and derivatives) contains way too many bugs, many of them are slips of the pen. They should not make it to the users system. Currently, it's either a nightmare or simply a matter of patience to get out of a broken pkg state. For the average user who doesn't know about the internals of the pkg system this is not acceptable.
---------------------------------------------------------------------------------------------------------------------------------------------
[...] indicates some lines left out for clarification
Example 1:
# export LC_ALL=C
# apt-get check
[...]
  aqsis-libsc2a: Depends: libtiff4 but it is not installed
[...]
# export LC_ALL=de_DE.UTF-8
# apt-get check
[...]
  dirmngr: Hängt ab: libpth20 (>= 2.0.7-2) ist aber nicht installiert
[...]
# apt-get check|grep aqsis
[none; -->different pkg listing w/ unmet depencies in different locales; this is utterly wrong]
---------------------------------------------------------------------------------------------------------------------------------------------
Example 2:
[This is from a previous run; after that I ran adept-updater and fetched a new pkg list, that's why the first pkg]
[shown differs from the example above]
# apt-get -s check
[...]
  acpi: Hängt ab: libc6 (>= 2.6-1) ist aber nicht installierbar
[translated: "acpi: depends: libc6 (>= 2.6-1) but can not be installed"]
[... many more pkgs complaining about libc6]
# dpkg -s libc6
Package: libc6
Status: install ok installed
[...]
Version: 2.6.1-1ubuntu10
[...]

CONCLUSION: The pkg maintainers either
    - did not agree on the correct naming convention for pkg versions (2.6-1 vs. 2.6.1) or
    - (more likely, because later I did not get complains on libc6; I have unattended security updates turned on)
    - version comparison is (was?) broken

Revision history for this message
Torsten Eichstädt (torsten-eichstaedt) wrote :

Launchpad ate my comment I was typing. When it's allowed to click on "Also affects + .." while I'm editing a new comment, it's a severe bug (in launchpad) to send the comment to a blackhole and not giving it back to me.

Back on topic, again: I reproducibly tracked this down and now can say for sure the bug is either in APT or in libc6/strcmp() (or both). I am investigating APT's sources to fix it (most likely missing setlocale('C') before version comparison). In the meantime, s/o responsible should URGENTLY read and considering my first comment, to fix root's locale to 'C' deep inside where noone can unset it. This workaround will temporarily fix _many_ other bugs affecting computer systems consistency, stability, etc. IMHO, once you (Ubuntu folk) decide to disable the root account -- which is the right way to go -- you are responsible for the consequences. Now what we have is a single user-visible entry point with 'sudo', and another (maybe few) in the system libraries, I'm not familiar with the latter. Fix it there, please. Root's locale _must_ be kept 'C' for a while. Wait for l10n to stabilize, then this can be reverted.

Changed in apt:
assignee: nobody → torsten-eichstaedt
status: New → In Progress
Revision history for this message
Jonathan Thomas (echidnaman) wrote :

Likely not adept's fault -> invalidating the adept task.

Changed in adept:
status: New → Invalid
To post a comment you must log in.
This report contains Public information  
Everyone can see this information.

Other bug subscribers

Remote bug watches

Bug watches keep track of this bug in other bug trackers.