Re: APNG iDNS Requirement

[C C Magnus Gustavsson <mag@lysator.liu.se>]

A few opinions:

On Tue, 4 Jan 2000, James Seng wrote:

> The system must
> 
> 1. be compatible with current DNS servers, clients and protocols.

Aside from ASCII-only domain names, no.


Linus Torvalds wrote in a flame a few years ago:

 "Anybody who has a mail gateway that isn't 8-bit clean these days should
  be shot. They _deserve_ to be broken. I do the same thing when I notice
  that something is badly broken in the kernel: I try to avoid breaking
  user programs, but if something needs to be done it really needs to be
  done.

  Quoted-printable isn't the answer. QP is the question, and the answer is
  NO."

 "The problem is that QP is _evil_. And the reason that QP is evil is that
  it only helps perpetuate a broken standard. Instead of fixing the
  standard, QP cludges around it, and that is why I personally really
  dislike QP.."

(Available at http://www.lysator.liu.se/åttabitars/linus-on-8bit.txt btw.
The URI should work encoded as either ISO 8859-1 or UTF-8.)


Adopting a solution as UTF-5 for domain names is once again an unnecessary
cludge around a broken standard. QP has caused and is still causing a lot
of inconvenience and I would hate to see history repeating itself with
DNS. Better do it right once and for all and reduce QP and UTF-5 to
historical curiosities long ago forgotten about.

The price we have to pay for using UTF-8 instead is of course that all
nameserver software has to be replaced to handle multilingual domains. I
vote for paying that price now when we still can and when it's still
cheap, as an investment for the future that will surely pay off. All
nameserver software will eventually have to be replaced anyway...

> 2. co-exists with current English domain name space.

Yes.

> 3. provides a mechanism for DNS to adopt multilingual characters

Yes.

> 4. retains compatibility with all existing Internet protocols (e.g. 
>    HTTP, SMTP, FTP etc)

No.

Each protocol need to be reviewed and fixed. Everything else (to use
Torvalds words again) only leads to more and more cludges, and more and
more work that is totally unnecessary.

Let's not create yet another broken standard but instead do it right from
the beginning. Let's aim at using UTF-8 instead of UTF-5 and let's design
the requirements from the point of view that even standards sometimes need
to be fixed.

/Magnus