On Sat, 28 Mar 2020 00:32:24 +0100
Mattias Andrée <maand...@kth.se> wrote:
Dear Mattias,
> This sounds absolutely horrible. Non-pre-composed characters are not
> widely well support and are often rendered terribly, some software
> (like the Linux VT) cannot even rendering them.
yes, the Linux VT is a good example. To really do the rendering
properly, you need a font-library that basically has infinite context
to draw characters.
This is not possible in a terminal, but you can at least "reserve" one
block for one grapheme cluster.
To put it another way, though: It's not the problem of the application,
but of the font renderer, and with complications in the TTF it only
gets more and more complicated.
> Why is even the kernel getting into encoding issues?, that should be
> an application issue, not a kernel issue. A kernel should only know
> bytes. Is it really a security issue?
I like to compare it to IDN homograph attacks, where you replace
characters like the letters a and e with homographs, in this case those
from the cyrillic alphabet а and е (they are not the same, even though
it looks like it!).
It doesn't take much creativity to see that it's enough to register a
domain
https://аmаzon.com/
and trick people into visiting it. In Firefox, you can tell it to
properly display all URLs with expanded non-ASCII forms, in this case
as
https://xn--mzon-43db.com/
and I can only recommend that to people.
In the case of file systems, I would probably go with a comparable
approach and in the kernel only work byte-wise, but when listing a
directory with to equivalent file names, I would print both in such an
expanded form. This would keep both sides happy.
With best regards
Laslo
