Re: [evolution-patches] ipv6 patch

[Not Zed <notzed ximian com>]

On Mon, 2004-09-13 at 12:34 -0400, Jeffrey Stedfast wrote:
> looks similar to what I started writing on friday (tho you've gotten a
> lot farther than I did).
Ahh didn't realise, sorry.
> I'll play around with this a bit. is this for proposal into 2.0.1? or
> 2.1?
I don't know.  There's a need for it in 2.0.1 I think.  Although its a fairly big patch.  This is up to JPR I guess.

I also fixed camel_freeaddrinfo to use a custom free'r in the appropriate case if you notice that issue in the patch.

> Jeff
>
> On Mon, 2004-09-13 at 18:11 +0800, Not Zed wrote:
> > 
> > This rather big patch addresses a major issue with network connections
> > from camel, that is, only 1 type of address is supported (ipv6 or
> > ipv4) for a given host at a time, and only one physical address for
> > that host will ever be tried, even if multiple addresses of the same
> > format exist.  The only other way to address this issue would be to
> > have all of the service connect calls perform two lookups using an
> > extended gethostbyname2_r interface, which is only available on GNU
> > anyway, where we already have getaddrinfo.  It would be a messier and
> > even less portable solution.  What we have now though is 'really
> > wrong'.
> > 
> > Most of the patch is just changing the api's to suit the changes and
> > to clean up some of the related apis too.  And to remove a lot of ipv6
> > vs ipv4 specific code from most of the places where it doesn't need to
> > know about the difference (like, just about everywhere, even the stuff
> > in stream-ssl could probably just use memcpy).
> > 
> > The main change is the new hostname resolution interfaces, which are
> > async/cancellable like the old ones, and use a getaddrinfo/getnameinfo
> > interface rather than a gethostby* interface.  So it emulates
> > getaddrinfo when it isn't available, and forces ipv4 in that case.
> > Sure it could also handle some ipv6 stuff, but i just don't care.
> > 
> > I'm inclined to simply drop support for all systems which don't
> > support getaddrinfo/getnameinfo anyway, because its too much pain to
> > deal with different address formats using the old-style interfaces.
> > But for now the patch has a very limited emulation which mostly seems
> > to work.
> > 
> > Some configure changes mean that ipv6 is simply enabled by default if
> > your system has the necessary interfaces, and it always uses those
> > interfaces if available even if you force it to use ipv4 (not that i
> > can see there's actually a reason to want to do this).
> > 
> > I would bet my house on there being lots of portability issues with it
> > all though.  e.g. redefining struct addrinfo on half/broken systems.
> > 
> > It hasn't been tested a lot, so there are probably other bugs too.
> > 
> > -- 
> > 
> > Michael Zucchi <notzed ximian com>
> > "born to die, live to work, it's
> > all downhill from here"
> > Novell's Evolution and Free
> > Software Developer