Hi!

Currently, service(8) is passing LANG (and LC_* for the version in
experimental) to the init script. However, this makes starting the
init script into a variable environment. When a user using
LANG=de_DE.UTF-8 will restart a service, the service will switch to
German. When someone with LANG=fr_FR.UTF-8, the service will switch to
French.

Another possible consequence is a service that will expect different
inputs. For example, when using fscanf(3) with LC_NUMERIC set to
fr_FR, floating-point numbers will expect a comma instead of a
dot. This may trigger difficult to diagnose bugs in some software. I
remember getting that with snmpd.

Therefore, I think this is better to use a fixed value from
/etc/default/locale.

- -- System Information:
Debian Release: jessie/sid
  APT prefers unstable
  APT policy: (500, 'unstable'), (500, 'testing'), (500, 'stable'), (101, 'experimental')
Architecture: amd64 (x86_64)
Foreign Architectures: i386

Kernel: Linux 3.14-1-amd64 (SMP w/4 CPU cores)
Locale: LANG=fr_FR.utf8, LC_CTYPE=fr_FR.utf8 (charmap=UTF-8)
Shell: /bin/sh linked to /bin/dash

Versions of packages sysvinit-utils depends on:
ii  libc6        2.18-6
ii  libselinux1  2.3-1
ii  startpar     0.59-3

sysvinit-utils recommends no packages.

Versions of packages sysvinit-utils suggests:
pn  bootlogd  <none>
pn  sash      <none>

- -- Configuration Files:
/etc/init/startpar-bridge.conf changed [not included]

- -- no debconf information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=Q6UY
-----END PGP SIGNATURE-----

[Vincent Bernat]

Hi.

This was introduced with bug #534300 when service was introduced in
2009 with LANG passed to init.d scripts, and #606875 which is a
request to also pass the rest of LC_* when LANG is passed on.

I believe either we pass LANG and LC_*, or we do not.  Do not have
strong opinions either way, but only passing LANG without passing the
rest of LC_* seem like an obvious bug to me.

Personally, I believe the init.d script should fix the locale settings
(for example using LC_ALL=C) if the service change behaviour based on
locale settings.

Is it possible to come up with a setup that pleases everyone?  Or are
the truly conflicting needs here?

 ❦ 15 mai 2014 16:17 +0200, Petter Reinholdtsen <pere@hungry.com> :

It is something difficult to know before hand.

Let's see if we can get input from people needing to keep LANG and LC_*.

Hi,

FTR, SLES is setting LANG to POSIX and CentOS/Fedora is not (re)setting
it at all in the environment before calling the initscript.

You must either pass neither, or pass them all.

The requirements are in conflict.  "service" is an user-level interface, it is
supposed to be subject to i18n and l10n.

OTOH, any initscript that requires, e.g. LC_COLLATE=C or LC_ALL=C ought to
set it.  Really.

control: reassign -1 init-system-helpers

service(8) now belongs to init-system-helpers

[2014-05-15 11:57] Henrique de Moraes Holschuh <hmh@debian.org>