Edward H. Trager wrote on 2003-08-06:

> On Wednesday 2003.08.06 08:29:37 -0400, Chris Heath wrote:
> > > Also, I think that console-keyboard drivers may be divided to 2 versions,
> > > with kernel build option:
> > >
> > > the simple, with fixed 8bit codepage, unused legacy things like VT100 graphics 
> > > removed
> > > (for people who dont'n need i18n or work not so much in console)
> > >
> > > and the sophisticated, full-functional and well internationalized.
> >
> > Yes, I think this is the way to go.  I imagine we will meet a lot of
> > resistance if we try to add heavyweight Unicode stuff into the existing
> > console.
> >
> > The heavyweight version would need a LOT of requirements gathering
> > before we even begin programming.  It probably should include support
> > for:
> >
> > * lots of encodings, but use Unicode internally
>
> Hmmm ... If I knew how to do this stuff myself (which I don't), I
> would not bother with any encoding except Unicode UTF-8 (If people
> did not like my patch, they wouldn't have to use it, right?).  That
> would cut down on the requirements and complexity a bit ... (a BIG
> BIT, N'est-ce pas?).
>
> I (and many others ...) would argue that everyone needs to move to
> Unicode.  So when you buy that shiny new version of Linux 5 years in
> the future, it's going to support Unicode very well, and it is
> perhaps no longer going to support the 3-5 mutually incompatible
> legacy encodings of your language that you previously had to
> struggle with ...
>
This can actually work out.  Just tell other people to run luit over
the UTF-8-only console.

> > * user-space pluggability for extra-heavyweight stuff like Japanese
> >    input methods or fonts
>
> I wonder if the object oriented design of SCIM (Simple Common Input Method:
> http://ns.turbolinux.com.cn/~suzhe/scim/index.html) could support CJK and
> other IMs on the console?
>
> > * bidi text (Arabic)

There is Akka from the arabeyes project which is supposed to give bidi
over a non-bidi terminal.  IIRC, last time I tried it didn't compile
for me.  Anybody had more luck with it?

> > * variable width fonts (CJK),
> > * variable-width encodings (Unicode combining chars),
>
> Yes, it would be nice if console worked as well as (or better than)
> mlterm (http://mlterm.sourceforge.net/) does for bidi, ligatures and
> combining characters for complex layout languages.
>
How do you do ligatures and combining characters within the limited
text mode?  Or do you mean only for the framebuffer mode?

> > * absolute positioning for things like line drawing...

You mean sub-charcell pixel-level positioning?  I don't think we need
to reinvent MGR, certainly the console is the least sensible place to
experiment with non-text extensions to terminal emulators.

> > * ANSI / ECMA and VT100 compatibility
> >
> > Hmmm... I suspect I've hardly even scratched the surface and already this
> > looks like it's going to be way too big for the kernel.  Maybe the
> > entire thing should be in user space.
> >
> > How important is it to have an in-kernel console?
> >
It'd be pretty annoying if you could crash the console.  Since it's
in-kernel it's almost impossible (you can only stuck the video card in
some mode the kernel doesn't understand ;-).  But we do trust e.g.
init(1) to be a user-space program.  I don't see strong reasons why
the console can't move to user-space, except for the fact that it
currently has a lot of syscall/ioctl hooks and interfaces would have
to be added to connect them to the user-space console process.

But there is another good reason to keep the kernel console and it's
non-unicode modes: there are very minimalistic builds of linux out
there, a dumb in-kernel implementation where the buffer is in video
memory is the most effective solution possible.

-- 
Beni Cherniavsky <[EMAIL PROTECTED]>
--
Linux-UTF8:   i18n of Linux on all levels
Archive:      http://mail.nl.linux.org/linux-utf8/

Reply via email to