Re: use of locale in "ls" again (Re: Japanese expression of date)

[Paul Eggert]

> Date: Sat, 22 Dec 2001 21:35:03 +0900
> From: Tomohiro KUBOTA <address@hidden>
> 
> Anyway 'ls' already has .po files.

It has them, but they don't work.  At least, they don't work for time
strings in the latest test versions of GNU 'ls'.  They haven't worked
since gettext 0.10.40 was merged into fileutils (more than 3 months
ago).  No user has complained about this, as far as I know.  Fixing
this is on my long list of things to do, but this stuff will be a
continuing maintenance hassle for some time.  I'd rather spend my (and
other people's) limited time elsewhere.

Thankfully, this part of the maintenance hassle is lower priority now,
since the default is now an ISO-style format that seems to be
acceptable to most users.

While we're on that subject, I still don't sense that we have a good
consensus to switch from the current ISO-style format to one that is
all-date or all-time.  Bruno Haible said that if "MM-DD" goes, the
result would be acceptable to him, but I worry that someone else will
noisily dislike whatever alternative format we come up with.

That being said, here's my latest idea: use "  HH:MM:SS" for a file
in the recent past, if the file is is dated today OR if the file is
less than six hours old.  Otherwise, use "YYYY-MM-DD".  This would be
more straightforward than the nonstandard notation with leading "-"
that I suggested earlier, and it also avoids much of the problem of
recent timestamps viewed just after midnight.


> > But GNU Emacs doesn't invoke 'ls' with LC_ALL=C.  It uses whatever
> > date format 'ls' uses, and displays that format to the user.
> 
> GNU Emacs must invoke 'ls' with LC_ALL=C if needed.

Emacs doesn't use LC_ALL=C, so I don't understand this point.


> If you really think the cost is the problem, abolish all .po files....
>
> If you say about possibility of other softwares, we will also need
> to stop using gettext at all, because this problem is not limited
> to date expression.

It's not a black-and-white situation like that.  It's merely an
engineering problem: what is the lowest-cost way for 'ls' to present
time stamps to users?  Even if we ignore the savings in implementation
hassle, a world-standard time format provides some cost advantages to
end-users.

There is ample precedent for using a single worldwide format for 'ls'
output.  For example, POSIX requires that "ls -l" must output "FOO ->
BAR" for symbolic links in all locales, including locales where "->"
is not an appropriate arrow symbol or where BAR should really come
before FOO.  Here, the advantage in having a worldwide standard is
considered to outweigh any locale-specific awkwardness disadvantages.
There's no a priori reason why 'ls' date formats can't be like "FOO ->
BAR", outside the POSIX locale anyway.