Both issues were raised in review.
webmin/upgrade_desc had <a href="webmin.com">, which a browser resolves
against the Webmin server rather than the public site. lang/en has
<a href=https://webmin.com>; the Turkish now matches. This was the only
relative href in any Turkish file.
proc uses __norefs=1, so the four $stime values were displayed
literally instead of expanding. lang/en spells the word out in all four
places rather than referencing stime, so the Turkish now does the same.
A scan for the same class of defect across every module found one more
case that had not been reported: useradmin/uedit_dothers was
$udel_dothers, and useradmin also sets __norefs=1. lang/en spells that
one out too, so it is fixed the same way.
The scan also found heartbeat/conf_auto_on and conf_auto_off, which are
${conf_enabled} and ${conf_disabled} in lang/en itself - so English is
affected as well, and conf_auto_ references ${default}, a key that does
not exist in the file. That is an upstream bug rather than a
translation one, so it is left alone here and reported on the PR.
Every module now has a reviewed desc_tr and longdesc_tr in
module.info.tr - 245 strings across 125 modules. Previously 37 were
reviewed, 194 came from module.info.tr.auto and 14 were missing
entirely (grub2, hardware-info, kea-dhcp, logviewer, nftables, systemd,
xterm), so those modules showed English titles in the Webmin menu.
The reviewed strings are removed from module.info.tr.auto, which now
holds only name_tr - the same split the de and ca translations use.
Thirty-six of those files became empty and are deleted.
Some of the machine text this replaces was wrong rather than just
clumsy:
sarg "Squid Report Generator" had become "Kalamar Rapor
Oluşturucu" - the fish, not the proxy
heartbeat the product name was translated as "Kalp Atışı"
cluster-cron "Cron Jobs" had become "Cron İş İlanları" (job ads)
proc "renice" was translated as "rename"
samba "file and print shares" was split into "create a samba
file" and "print the shares"
exports, hpuxexports, sgiexports, inetd, init, mount, net
each repeated the file path mid-sentence
firewall, the "(iptables)" and "(ip6tables)" qualifiers that
firewall6 distinguish the two modules were dropped
Turkish capitalisation of English words was also wrong throughout
("İptables", "İnetd", "İscsitarget", "İnternet"), and three previously
reviewed strings were stored as double-encoded UTF-8 (cron/desc_tr,
mailboxes/desc_tr, mailboxes/longdesc_tr).
cluster-cron and ldap-server carried a stray desc_de_tr key - a German
key that had leaked into the Turkish file. It is removed.
A consistency check of every lang/tr against its lang/en turned up 58
lines that were wrong independently of the missing translations. All of
them are fixed here:
Missing or wrong $N placeholders (the value silently disappears from
the message at runtime, or the wrong value is shown):
acl/cert_eca, acl/hide_desc, adsl-client/index_stopdesc,
apache/reconfig_ever, apache/log_files, apache/log_files_l,
bind8/log_manual, dhcpd/listl_lfnotcont, fdisk/mkfs_desc1,
fdisk/fsck_desc1, fdisk/tunefs_desc, hpuxexports/save_ehost,
inittab/log_rename, mysql/index_ever, mysql/index_version,
mysql/log_perm_create, mysql/log_perm_delete, mysql/log_perm_modify,
sendmail/index_epath, sendmail/index_startmsg, sendmail/asave_ematch,
sendmail/mailers_efile, sendmail/generics_efile,
webalizer/index_ewebalizer, webalizer/index_econf
cron/env_ename carried a $1 that lang/en does not have, and its text
described a different error than the key means.
adsl-client/index_stopdesc and inittab/log_rename each repeated $1
where $2 was meant, so both values printed the same string.
HTML markup that did not match lang/en:
bind8/mass_desc, dhcpd/index_dhcpdnotfound, firewall/index_existing,
inetd/index_applymsg, mysql/index_nomod, mysql/index_nomods,
samba/esync_msg, samba/viewu_list, sendmail/trusts_desc,
sendmail/mailers_desc2, sendmail/domains_desc, squid/clear_msgclear,
status/proc_thresh
Several of these were also truncated mid-sentence, dropping a whole
sentence of the original text; those are now translated in full.
Encoding:
mailboxes/confirm_warnf was stored as double-encoded UTF-8 and
displayed as mojibake. It also used $1 twice instead of $1 and $2.
Trailing whitespace:
proc/kill_term, kill_hup, kill_stop, kill_cont
Untranslated strings:
fdisk/select_* device descriptions, mount/irix_mode0, irix_mode0cd,
shorewall/tunnels_ipsec:noah
bind8/mass_desc previously translated the literal configuration
keywords master, slave, stub and forward into Turkish; they are now
left in English, since they are values the user has to type.
Finishes the Turkish translation of system-status, logviewer, smart-status,
updown, htpasswd-file, bsdexports, dfsadmin, openslp, mailcap,
backup-config, at, proc, sgiexports, samba, lpadmin, cluster-cron,
tcpwrappers, hpuxexports, exports, cluster-shell, webmincron, time,
change-user, syslog, shell, inetd, htaccess-htpasswd, krb5, xterm, idmapd,
cluster-passwd, tunnel, inittab, passwd, man, xinetd, webalizer,
adsl-client and help, and removes their tr.auto files.
htaccess-htpasswd needed special handling: its existing lang/tr was a
verbatim copy of lang/en, so reusing it would have carried 75 English
strings into the finished file. The whole module was translated instead.
Six other strings that were still in English in acl, firewall, webalizer
and xinetd were translated at the same time.
Product and protocol names stay in English, including SMART, WebNFS,
Kerberos, LIPKEY, SPKM, OpenSLP, Winbind, TCP Wrappers, Xinetd, journalctl,
Boot Loader Specification, Trusted Platform Module and Alpine Package
Keeper.
ⓘ Avoid forcing xterm shell PTYs into IO::Stty raw/noecho mode, which can leave interactive shells with broken echo, line editing, and control-key behavior. Keep the existing stty logic for other PTY callers, but add an opt-out flag so xterm can let the shell manage terminal mode normally.
https://github.com/webmin/webmin/issues/2452
*Note: Replace `sysctl -a` with targeted queries for only the 5 needed values (hw.physmem, hw.pagesize, vm.stats.vm.v_*_count) instead of dumping thousands of kernel params. This reduces `get_memory_info()` overhead from 25% CPU to ~5% CPU when called by real-time monitoring every 1-3 seconds.
