Re: [v6ops] IPREF as a transitioning tool

["Soni \"They/Them\" L." <[e-mail]>]

On 2023-11-12 08:50, Gert Doering wrote:
> Hi,
>
> On Sun, Nov 12, 2023 at 08:34:32AM -0300, Soni "They/Them" L. wrote:
> > We mean we still think moving the CLAT to libc would be a good idea.
>
> This encourages application programmers to stick to IPv4 sockets, because,
> why should they care?  CLAT will take care.
>
> Remember we're still talking servers here, in a v6-only DC, and server
> applications being under much better control (normally) than clients
> (or worse, IOT).  Adding CLAT functionality just adds more code paths
> that need to be tested, and get in the way - a hard and fast fail on
> "try to use an IPv4 socket" is good, not "it works most of the time".

Moving it into the libc also means removing those pesky IPv4 syscalls.

It is about pushing it on application developers to deal with.

For bonus points, move it into a separate libc than the system libc. 
Require explicitly linking with libc4 (ooh, sounds explosive!) if you 
need IPv4. This is about as close as you can get to "bring your own 
IPv4" without causing too much pain to sysadmins (after all, we need to 
mind the sysadmins if we want sysadmin adoption - and in the context of 
IPv6 deployment, sysadmins matter slightly more than developers). Tho 
the system libc should still know how to resolve IPv4-only domain names 
as IPv6 addresses - luckily the libc doesn't expose DNSSEC to the 
application.

>
> Gert Doering
>          -- NetMaster