On Thu, Apr 26, 2012 at 9:01 PM, Gert Doering <g...@greenie.muc.de> wrote:
> Hi,
>
> On Thu, Apr 26, 2012 at 08:35:54PM +0300, Alon Bar-Lev wrote:
>> At embedded system and several distros like Gentoo, the ipv6 should be
>> made enable/disabled, just like any other project.
>> It is the same #ifdef as we already have these in code.
>> Even in 2012 kernels can be compiled without ipv6.
>
> Enabling/disabling IPv6 for "just some piece of random code that can
> open up network sockets" is easy, and usually not even worth the extra
> lines of code ("just pass different options to getaddrinfo() and/or
> bind()").
>
> Given the nature of OpenVPN, and that almost everything in the code
> (except for the crypto stuff) is related to IP packets or IP addresses,
> it's not that straightforward here.
>
> Adding IPv6 to the OpenVPN code base touched well over 100 different
> places of the code, and to really make it optional (to get the space
> saving worth for an embedded system) you'd have to have #ifdef's at
> all these places, in the config data structures, and so on - this is
> lots of work, especially maintaining both code paths and testing all
> this, with dubious benefits. As I said, we got bitten by JJO's
> decision to make IPv6 transport conditonal, and decided to get *rid*
> of those #ifdefs for the sake of "less code paths, less maintenance
> effort".
>
>
> I'm not sure you can find packages that have had working IPv6 support,
> and then decided to make it conditional afterwards. Usually IPv6 was
> added at some point, and it was added conditionalized, because that
> package can also be compiled on older systems without IPv6 - since we
> have the luxury of not supporting anything that has no tun/tap interface,
> we have the same luxury of not having to bother with anything that has
> no IPv6. (Even if you compile the Linux kernel and glibc without IPv6,
> the API stuff - getaddrinfo(), inet_ntop(), etc., will still be there,
> because they are IP version agnostic - they will fail if you actually
> try to *use* them for IPv6, but you can compile and run the same
> applications just fine).
>
Right, the ipv6 was added indeed differently, without this conditionals.
Because of this I did not perform this task as part of the major build
cleanup, and left this for the next phase.
Most of the lowlevel network code already has #ifdefs, I will try my
best to find a way to do the conditional without much changes.
Alon.
