>Not necessarily. The `locale module' that one loaded could just have a list
>of (original English message, translated message) tuples, with some hashing
>function to let it do quick lookups. printk() doesn't happen very often, so
>performance wouldn't be that much of an issue, and you'd just have to accept
>the memory drain. Then if there was no match in the locale it would just
>return the original string.
>> defaults for others. How could we implement this on a per-user basis?
>You can't. Kernel messages aren't directed to a user. Per-machine is all the
>granularity you can get.
The reason that they aren't (and can't be) directed to a user is that they
all end up in some text file in /var/log. However this doesn't necessarily
have to be the case. However if we didn't store them in plain text format then
we wouldn't have this problem. What if we had a binary format for logging data
which contained reference numbers for the messages. So to display the
following message: bsd_decomp1: bad sequence # 1305, expected 1300
We could call:
klogmsg("bsd_comp", DECOMP_1, 1305, 1300);
This would log that an error had occurred in bsd_comp and the error number
was DECOMP_1, then the system could look at the file
/msg/bsd_comp.language-code and drag out the text for message DECOMP_1 (this is
for displaying messages directly on the console). Then when the klogd/syslogd
daemons are running they could store the data ("bsd_comp", DECOMP_1, 1305,
1300) in a binary file which could be displayed by any user in any language.
Of course this idea might not be well accepted as it involves changes to the
kernel, klogd, and syslogd, and most people don't like change. However I think
it's the only way for Linux to progress in this area. Under NT it's possible
to change languages and view all the old system log entries under the new
language. I think that we should go one step further and allow several users
who are simultaneously logged in to view the same log file in different
languages.
>Frankly, I'm not sure that the idea is particularly worthwhile, but we'll
>see...
I think that the original idea of "jive" isn't worth-while, and some of the
later ones offered little benefit that a simple "#define LANGUAGE" in one of
the base kernel header couldn't achieve. But I think that being able to have
multiple people read the same log file in different languages offers great
potential benefits.
Russell Coker