Found that /etc/localtime was now a binary file.
Just as a experiment I renamed it to localtime.old and created a symlink named localtime pointing to /usr/share/zoneinfo/Europe/Tallinn and the clock in Plasma 5 system try started to show correct time at once

CC: (none) => jyri2000

The problem seems to be in Qt5 qtimezoneprivate_tz.cpp: systemTimeZoneId() tries to obtain the system time zone in several ways, and it seems for Mageia (or at least for me) it ends up parsing /etc/sysconfig/clock, but it parses the value wrong

Line 943:

if (line.left(5) == QStringLiteral("ZONE=")) {
    ianaId = line.mid(6, line.size() - 7).toUtf8();
} ...

I suppose it should be

ianaId = line.mid(5, line.size() - 5).toUtf8();

and have no idea why it's done like that instead.

CC: (none) => ita84

Nevermind, it does that because in RHEL /etc/sysconfig/clock values are in double quotes (e.g ZONE="Europe/Berlin"); in Mageia there are no quotes.

Anyway, it's an upstream issue, I believe, since none of the other ways systemTimeZoneId() uses to determine the system time zone work either: they're as follows:

- using the TZ environmental variable
- parsing /etc/timezone (Debian)
- extracting zone name from the /etc/localtime symlink, if applicable
- parsing /etc/sysconfig/clock, and assuming RHEL format with quotes
- if all else fails, set to UTC

I guess a workaround would be to set the TZ variable; I don't really see a clean way for Mageia to patch around this.

So is it that the Mageia installer issue, that here, instead of creating a /etc/localtime symlink to /usr/share/zoneinfo/continent/city file it copies that city file to /etc and names it to localtime?

I don't know what choices were made by Mageia regarding timezone settings, but having a copy of the timezone file in /etc/localtime instead of a symlink seems to be a legitimate configuration (I believe Gentoo uses it, probably because of the chance of a separate /usr partition; in that case, though, the timezone string is found in /etc/timezone instead, and that's accounted for in QTimeZone).

On the other hand, I believe that systemd's timedatectl assumes /etc/localtime is a symlink, so maybe doing that would be a better alternative, assuming we don't want to screw over users with a separate /usr partition.

Again, I have no idea why things are like they are in Mageia now, so I don't know if any changes to the timezone configuration could affect other parts of the system.

I really think we should just ensure that /etc/localtime is a symlink. Other things rely on this too.

The only possible argument would be if /etc is on a separate partition to /usr and /usr was mounted late but this really shouldn't affect us in Mageia (/usr is mounted early in initrd).

So yeah, myself and Angelo noticed this the other day too for other things.

I think Angelo was working on making our tools not write a symlink. Can you confirm?

CC: (none) => anaselli

@Colin well i have a patch for the mana tools, that it's easy to port to drakxclock, i.e. unlink localtime and link it with pointing to the new file.
Tell me if you want me to commit it, we can then move the discussion on dev mailing list if something is not wrong or can be implemented better

Angelo, you can port it and attach, so Colin can review it :)

Created attachment 5784 [details]
patch to unlink and symlink new localtime

This patch should work, but i cannot test it atm. I will add the similar one to mana tools if this works.

Colin, can you review please? So we can get it fixed before release :)

Before reviewing could you test if using manaclock, the original problem is solved? Or at least linking localtime... Because if it isn't there is no reason in porting this patch but using dbus correctly...

Made a test:
1. after installation corrected the time by creating sysmlink /etc/localtime pointing to /usr/share/zoneinfo/Europe/Tallinn
2. changed timezone using drakclock to /Europe/Helsinki (actually the same timezone) - clock started to show wrong time - probably UTC (bios time).
3. changed timezone using manaclock back to /Europe/Tallinn - clock started to show correct time according to timezone.

Had to run manaclock as root. Running as user it showed that the imezone was changed while in reality it was'nt.

> Had to run manaclock as root. Running as user it showed that the imezone was 
> changed while in reality it was'nt.
yes i'm moving lemme lemme to dbus as far as i can so all can be used bu users and
password is promped when needed. In mana clock i probably forgot to check the user id because once user could change the setting as in drakclock... anyway the test seems to work

Jüri i made some tests and there is no need to use manaclock as root, but to run it properly using "mana clock" or "pkexec /usr/bin/manaclock" is required atm.

Thierry, can you please review the patch as you are the maintainer.

Assignee: bugsquad => thierry.vignaud

get rid of unlink() and just use symlinkf().
Also I think we should keep the eval {} :
    eval { symlinkf($tz_prefix . '/' . $t->{timezone}, "$::prefix/etc/localtime") };

interesting i always thought to eval to trap exceptions that would crash the program more than functions returning 0, is that a case in which eval could work correctly (such as unlink and symlink do)? 

BTW symlinkf is just a subroutine that contains a call to unlink and then one to symlink, so i don't see a real difference there :).

I tried to use eval without any real benefit, so please explain why i should use it.
A problem in that code is that if the link (or old copy) fails the program goes on anyway in my opinion, despite on 'eval', 'if' or any other checks, it just log atm.

commit f6d83149036c800310fb307b79a477c614141056
Author: Colin Guthrie <colin@...>
Date:   Sun Feb 8 21:33:24 2015 +0000

    drakclock: Ensure that /etc/localtime is a symlink (mga#14888)
    
    This is actually just a followup to d4be53b62 to use the wrapper symlinkf() function
    which seems a little bit neater (subjective) and to update NEWS file.
---
 Commit Link:
   http://gitweb.mageia.org/software/drakx/commit/?id=f6d83149036c800310fb307b79a477c614141056

This commit is going to horribly break things.  /etc/localtime is a copy of the timezone file, as established by the timezone package and the update-localtime script that comes with it.  If we're going to change this to a symlink, it needs to be adapted there.  Otherwise, there will be horrible issues due to the different file types (regular file vs. symlink).

CC: (none) => luigiwalser

Haven't seen any problems yet, and I did that same symlink change on my mga4 system too. If timezone package is doing some other changes then yeah, this package must be fixed too. But this commit is a change in the right direction.

In fact the whole point of *copying* the zoneinfo file becomes obsolete with a symlink anyway, so that script can die and we can just add a post script to transition to a symlink if it's a file. Should be trivial.

OK, after Sander's update, I did a one-time transition to a symlink. Does that look OK David? Can you think of any other areas where this would cause problems?

Actually thinking about this, I think it's better to use triggerun here as we don't really want this to happen all the time - overwriting the current symlink with the potentially stale data in /etc/sysconfig/clock.

e.g. under GNOME, if you have automatic timezone adjustment on, then the symlink will be updated for you (via timedated), and if you upgrade your timezone package, you don't want the (now stale) info in /etc/sysconfig/clock kicking in.

So let's just transition once.

Gah. That won't actually be needed will it as it's alreayd a symlink then :s Pah. No harm tho' IMO.

Looks OK to me.  We'll have to test it and see how it goes.

Just to be clear, I do think symlinking is better.  Fedora already does this, and I like the self-documenting nature of being able to tell which zone it is by reading the link.

:)

FWIW, /etc/sysconfig/clock is now pretty much totally obsolete. ZONE= is in the symlink target name, UTC=true/false is encoded in the third line of /etc/adjtime.

I don't know what ARC is and the only reference I can find is in perl-install/timezone.pm which always sets it to false, so think it's pointless. This then tallies up with how timedated does things. So the next step is to kill off the parsing/use of this sysconfig file.

dbus api seems to work fine: 
http://gitweb.mageia.org/software/adminpanel/tree/lib/AdminPanel/Shared/TimeZone.pm
if our DEs work fine we could close this issue....

I don't know if I am on the right bug, but...

After a clean install of Beta 3, my /etc/localtime was set to 

/mnt/usr/share/zoneinfo/Australia/Melbourne

The zone was right, but the /mnt looks like a leftover from the installation process.  An error like this could explain the bug that KDE couldn't set the time, etc.

(I wish there was a quick way of looking up that bug number without leaving this screen.)

CC: (none) => laidlaws

Yes, bug was introduced by commit f6d83149036c8003 above

The first symptom was that Linux was seeing hardware time as UTC, and Windows was seeing it as local time.  A bit of Googling led me to the fix.

The commit didn't consider the installer, so after installation you have that wrong side effect. Anyway the KDE problem was present already, since we had no
link before @Doug on comment#31 please try drakclock and see if now the link is correctly set and KDE work right.

commit 46ce1e3284d595bd7be1ab41a79ecd67730553e4
Author: Colin Guthrie <colin@...>
Date:   Sun Feb 15 09:52:03 2015 +0000

    Fix timezone dir install prefix prepending (broken symlink during installer mga#14888)
---
 Commit Link:
   http://gitweb.mageia.org/software/drakx/commit/?id=46ce1e3284d595bd7be1ab41a79ecd67730553e4

Thierry, does the above look OK? Not sure if the API is called outside of drakx repo (always a problem when changing this code - no idea where else it's used!) but I can't think of too many places which would want this - especially in an installer context!

(In reply to Mageia Robot from comment #33)
> commit 46ce1e3284d595bd7be1ab41a79ecd67730553e4
> Author: Colin Guthrie <colin@...>
> Date:   Sun Feb 15 09:52:03 2015 +0000
> 
>     Fix timezone dir install prefix prepending (broken symlink during
> installer mga#14888)
> ---
>  Commit Link:
>   
> http://gitweb.mageia.org/software/drakx/commit/
> ?id=46ce1e3284d595bd7be1ab41a79ecd67730553e4

Yes Sir!

The Robot Apocalypse has started already!

(In reply to Angelo Naselli from comment #32)
> The commit didn't consider the installer, so after installation you have
> that wrong side effect. Anyway the KDE problem was present already, since we
> had no
> link before @Doug on comment#31 please try drakclock and see if now the link
> is correctly set and KDE work right.

I re-worked the symlink as it should have been.  Immediately the correct time was shown on GkrellM.  I run Xfce, not KDE.  I logged out and in again, and the correct time showed in the notification area.  I have my clock synchronised with NTP.  In this case setting the time with drakclock did nothing, just exited.  I didn't need to look again.

(In reply to Colin Guthrie from comment #34)
> Thierry, does the above look OK? Not sure if the API is called outside of
> drakx repo (always a problem when changing this code - no idea where else
> it's used!) but I can't think of too many places which would want this -
> especially in an installer context!

this looks a little big ugly :-(
I would have extracted a get_raw_timezone_prefix() from get_timezone_prefix() instead...

Created attachment 5920 [details]
possible patch

Could this patch sound better?
If so i'll commit later at home

yes this looks better.
Though we use the following style for options:
  my ($system_prefix) = @_;

Also, for perl_checker, we prefix boolean options with "b_"

So we would rename system_prefix to b_system_prefix.
As it's an hint, I would further rename it to $b_use_system_prefix

Last but not least, in write(), the $system_prefix variable is useless, just use:
    my $tz_prefix = get_timezone_prefix(1);

After "git revert" previous commit of course

On comment#39 yes the get_timezone_prefix($use_system_prefix) was only for readability :)

commit b26a69c7edbec529be9d325414e8cb47ca132182
Author: Angelo Naselli <anaselli@...>
Date:   Sun Feb 15 12:45:58 2015 +0100

    Revert "Fix timezone dir install prefix prepending (broken symlink during installer mga#14888)"
    
    This reverts commit 46ce1e3284d595bd7be1ab41a79ecd67730553e4.
---
 Commit Link:
   http://gitweb.mageia.org/software/drakx/commit/?id=b26a69c7edbec529be9d325414e8cb47ca132182

commit 9ff316b07c005c0e43e2a9f0546baa85a3abaaae
Author: Angelo Naselli <anaselli@...>
Date:   Sun Feb 15 12:54:28 2015 +0100

    Fix timezone dir install prefix prepending (broken symlink during installer mga#14888)
---
 Commit Link:
   http://gitweb.mageia.org/software/drakx/commit/?id=9ff316b07c005c0e43e2a9f0546baa85a3abaaae

*** Bug 15301 has been marked as a duplicate of this bug. ***

CC: (none) => junknospam

So, is this bug fixed after the latest commits in comment 42 and comment 43?

Not present in a clean install of the RC.

Fixed then.

Status: NEW => RESOLVED
Resolution: (none) => FIXED