Greetings from London

[Lars Wirzenius]

On ma, 2007-05-21 at 11:13 +0200, Matthieu Moy wrote:
> Yes, but that's cheating ;-). In Unix, a filename is a sequence of
> bytes, not characters. This is the application's responsibility to
> display it correctly (and therefore take care of encoding).

Yes, that is what Unix does. I don't think it's cheating, I think it's
the only sensible way to do it. That means the kernel doesn't need to
change every time there's a new character set.

> What you probably expect from bzr is that a file named "éèâ" that you
> commit on a UTF-8 machine will be checked-out as "éèâ" on a latin1
> machine. And _that_ is non-trivial.

That's not what *I* expect at all. I expect that bzr (or any other part
of the system) doesn't change my filenames willy-nilly, or be
unnecessarily picky about what they are. If I want my filenames to be
portable between different environments, I'll happily make sure of that,
by using a character set that works well (ASCII, traditionally, and
increasingly UTF8 these days). If I don't care about the portability, I
don't want bzr getting in my way by insisting I do. I don't want bzr to
protect

The inability to use bzr on any random Debian source package is one
scenario where bzr's insistence on UTF8 filenames gets in my way. Most
of the packages work fine, but it's impossible (or was, when I ran my
test) to use bzr on a few of them. 

One of the thing I'd like try some day is set up a system to import
every revision of every Debian source package into a version control
system. This could then be used to do things like find out what has
changed in each revision, when, and who did it. Further, people working
on big library transitions could perhaps more easily do that, by
branching the affected packages, and merge new revisions of the packages
into their transition branches, and then when everything finally works,
easily give the necessary changes to each package maintainer.

A system that insists on anything more strict for filenames than what
the kernel does isn't good enough for that. Such as system can be a very
good tool, but it's not a power tool.

(There are, obviously, other requirements for being a power tool. :)

-- 
I am a werehuman.