Re: [idn] What happens to 8bit DNS requests

[Keith Moore <moore@cs.utk.edu>]

> I appreciate that there are different issues associated with UTF-8 or other
> non-ASCII approach, my main point is that even if we do go to IDNA, chances
> are that users unknowingly will still be sending multilingual dns requests
> through existing products that are non-IDNA aware.  

agreed.  I think such a document could be useful, with the following caveats:

- it would be very hard to make it sufficiently comprehensive that 
  it could serve as guidance for the design of an IDN solution, or
  even to serve as a guide for users wanting to know which products
  to buy.

- it could take a large amount of time and energy


so I wouldn't mind at all seeing such a document produced, but I wouldn't 
want to see such an effort divert resources from the primary work on IDNs, 
or delay the primary work on IDNs.

regarding the document you referenced (http://www.openidn.org/issues.html)

- there are many more kinds of components affected than those that you
  list (web browser, kernel, proxy server)

- there are many more implementations of each of these kinds of components
  than those that you list.  

- there are other platforms than those you list, with different issues
  than these.

all of which just illustrates that it's difficult to make this kind of
list comprehensive.

- it's not clear exactly what is meant by "kernel" or how it comes into
  play in the interaction between applications.

- it appears to be difficult to isolate or even document the nature of 
  certain problems, especially given that the behavior may change 
  depending on program preferences, presense of proxies, choice of 
  platform, etc.

- the issues you cite are good examples of how 8-bit characters get mangled 
  by some existing programs, but it's hard to know how to use this 
  information except to demonstrate that 8-bit characters get mangled
  in a variety of ways. 

- the "solution" it recommends for squid is somewhat dubious:
  it oversimplifies the problem of making IDNs work for the web.

  e.g. even if we adopted a UTF-8 on-the-wire solution for URLs in the 
  HTTP protocol (something that is not within the scope of this group
  to decide), it's not sufficient to simply change these programs to be 
  8-bit transparent for typed-in URLs.  there are other interfaces 
  which contain URLs (web pages, config files, APIs used to control  
  the browser, cut-and-paste), nameprep would certainly have to be 
  applied to URLs in the browser so that local caching would work 
  (even if proxies and remote servers were also taught to do nameprep)