Re: draft-cmetz-v6ops-v4mapped-api-harmful-00.txt and draft-itojun-v6ops-v4mapped-harmful-01.txt

[Iljitsch van Beijnum <iljitsch@muada.com>]

On vrijdag, sep 12, 2003, at 11:33 Europe/Amsterdam, Mauro Tortonesi 
wrote:

> then my credo as a developer is "ipv6 is a brand new protocol", because
> AF_INET6 sockets have many options that are not applicable to ipv4
> connections, and writing applications that make use of the ipv4-mapped
> catch-all approach will be a great problem for developers (that will 
> find
> themselves writing __a lot__ of checks/workarounds for special cases 
> and
> will become crazy during the testing/debugging process).

So what's the alternative? Ripping mapped addresses out of current 
implementations doesn't make sense for many reasons. The only question 
is whether you should/want to use them. If you want to do stuff with 
IPv6 that you can't do with IPv4, you'll have to do a lot of checking 
regardless of how you do it.

If you feel using the IPv6 socket API for IPv6 and the IPv4 socket API 
for IPv4 makes more sense, then by all means, do it that way. But for 
simple applications I think it's a huge plus to be able to talk to the 
network in a unified way without having to spend time and effort on 
making the IPv4/IPv6 distinction. Obviously this means you can't do any 
IPv6-specific stuff.

> moreover, i think that the ipv4-mapped catch-all approach in the
> development of ipv6-enabled apps may eventually become an obstacle for
> widespread adoption of production quality ipv6-enabled software (since 
> it
> makes more difficult the use of the advanced ipv6 capabilities)

I don't agree. Just tell the stack you don't want to talk IPv4.

Also, in the document I read it seems half or more of the trouble comes 
from mixing IPv4 and IPv6 calls. So don't do that.

I'm sure that from the vantage point of someone who writes large and 
complex applications things look different, but as someone who only 
writes simple stuff, I really appreciate having the mapping approach 
available.