IPv6

[Leif Hedstrom <zw...@apache.org>]

Hi all,

I've had to move out the IPv6 feature to v2.3.0, since there are no resources available to work on this. The tickets in question is Jira TS-18 and TS-396. The issue is that the code only "works" on Linux, and completely fails in the network stack on other platforms.

If someone is interested to work on this, there are a few options

1) Fix the code / patch :).

2) Maybe make the IPv6 code Linux specific, i.e. only compile / use on Linux. I'm somewhat concerned about this, the fact that we break it on all other platforms is a pretty serious issue I think.


Thoughts?

-- leif

Re: IPv6

["Alan M. Carroll" <am...@network-geographics.com>]

Sunday, September 12, 2010, 1:56:21 PM, you wrote:

>  I do however believe it's only reasonable to expect developers to support gcc in general, and follow C/C++ standards as much as possible. We should avoid making the code depend on very unique or specific gcc features unless necessary, but compiling with gcc should be adequate IMO.

The problem arises in that standards compliance improves over the life of gcc, so that one can easily write standards compliant code that won't compile on a sufficiently early version of gcc. After 4.x, though, there hasn't seemed to be that much change so it's less of a worry.

I also looked in to TR1 for gcc and it's mostly there, but you have to change compiler modes. I don't think it is going to work well for ATS. I'm a bit bummed about that.

Re: IPv6

[Leif Hedstrom <zw...@apache.org>]

For gcc we really only support v4.x due to the atomic stuff (the alternative code is not well maintained). I've personally made changes for ICC support, and Theo supports the SunCC.

 I do however believe it's only reasonable to expect developers to support gcc in general, and follow C/C++ standards as much as possible. We should avoid making the code depend on very unique or specific gcc features unless necessary, but compiling with gcc should be adequate IMO.

Sent from my iPad

On Sep 12, 2010, at 12:01 PM, "Alan M. Carroll" <am...@network-geographics.com> wrote:

> A related question is compiler compatibility. How concerned are we with code that compiles on back version of gcc or on completely different compilers

Re: IPv6

["Alan M. Carroll" <am...@network-geographics.com>]

A related question is compiler compatibility. How concerned are we with code that compiles on back version of gcc or on completely different compilers?

Re: IPv6

[Leif Hedstrom <zw...@apache.org>]

On Sep 10, 2010, at 6:12 PM, Miles Libbey wrote:

> Also seems like a pretty interesting meta question.  I'd guess that most developers use (or have access/knowledge of) a very small subset of operating systems that TS supports, and that other enhancements will run into the same issue.  If we generally follow the #1 suggestion, I wonder how this will affect or impair general progress.  If we generally follow the #2 suggestion, our 'support' for various OS's will likely be flakey.  Seems like rock meets hard place. 

Yeah, good points. I'm hoping that #1 is a matter of using the right posix / socket APIs (correctly). Generally, these APIs are the same across all platforms we support, I just don't know what it is we're doing that "breaks" things on non-linux systems (but I'm guessing we're doing something Linux specific stuff where we ought to use something portable).


-- Leif

Re: IPv6

[Miles Libbey <ml...@apache.org>]

Also seems like a pretty interesting meta question.  I'd guess that most developers use (or have access/knowledge of) a very small subset of operating systems that TS supports, and that other enhancements will run into the same issue.  If we generally follow the #1 suggestion, I wonder how this will affect or impair general progress.  If we generally follow the #2 suggestion, our 'support' for various OS's will likely be flakey.  Seems like rock meets hard place. 

miles


On Sep 10, 2010, at 10:22 AM, Leif Hedstrom wrote:

> Hi all,
> 
> I've had to move out the IPv6 feature to v2.3.0, since there are no resources available to work on this. The tickets in question is Jira TS-18 and TS-396. The issue is that the code only "works" on Linux, and completely fails in the network stack on other platforms.
> 
> If someone is interested to work on this, there are a few options
> 
> 1) Fix the code / patch :).
> 
> 2) Maybe make the IPv6 code Linux specific, i.e. only compile / use on Linux. I'm somewhat concerned about this, the fact that we break it on all other platforms is a pretty serious issue I think.
> 
> 
> Thoughts?
> 
> -- leif
>