Re: integrating husni's ipv6 patch into squid-2

> On Mon, Feb 04, 2008, Guido Serassio wrote:
>
>> I don't like so much the addition of IPv6 to Squid-2 for two main
>> reasons:
>>
>> - This could be a very dangerous thing for the Windows
>> compatibility, see the impact of the IPv6 merge into Squid-3, even
>> with the C++ help.
>
> Looking at it, XP and later has a sockaddr_storage of sorts:
>
> http://msdn2.microsoft.com/en-us/library/ms740504(VS.85).aspx
>
> Hm, the earliest appearance of this is Windows XP SDK; which means
> either faking it (and having to provide local versions of all the
> library functions that support it) or abstract it out enough so
> a v4-only build only affects a small area of the codebase (ie,
> somewhat what Amos has done in -3.)

WinVista _almost_ supports the linux way of things. There are port-level
options that are defaulted differently, but otherwise the app needs to be
dual-stack capable.

WinXP was faking it, you need to hold and open two of every listening port
and hope for the best when opening client-facing ones, particularly
re-connects to an old client (ie FTP, SSH, pconn, maybe others).

Win2K has NO support so all the required generic libraries need to be
emulated internally in squid. That had me and Guido spending several weeks
of go-between locating and adapting code we could use in GPL squid.

> If we wanted to target very old NIX platforms (ie, ones without
> any v6 support in the C libraries) then we'd have to do this anyway.

You mean like all those setups we are trying to convert off 2.5?
Linux <2.6, MacOSX <10.2, Solaris <8, FreeBSD <6.0, NetBSD<2.0,
OpenBSD<3.0 are almost all in that category or have large know problems
with their v6 support.

Didn't even bother considering older unices in squid-3, since anything
able to compile C++ templates is recent. Squid-2 being C has an older
range of hardware to track.

>
> Hm, it shouldn't be all that hard. Let me think about it a little.
>
>
> Adrian
>

Amos
Received on Mon Feb 04 2008 - 21:11:14 MST