[Python-Dev] IPv6 in Windows binary distro

["Martin v. Löwis"]

Thomas Heller wrote:

> - Should the next binary release (2.3.3, scheduled for the end of 2003)
> include this support?

I'm leaning actually somewhat towards requesting this, although the
"no changes in Micro releases" is a strong point.

I would not at all be concerned if this was a pure add-on feature.
However, there might be minor changes to existing behaviour:
- python23.dll would now require winsock2.dll. I'm unsure whether
   Win95 was already providing this library.
- the getaddrinfo implementation would now be the Microsoft
   "native emulation", instead of the Python one. It is a native
   emulation because it detects proper getaddrinfo dynamically
   if available, and falls back to emulation otherwise. This might
   cause minor semantic changes over our emulation code.
   [there would also be significant semantic changes in case a host
    has an IPv6 address - but that would be the whole point of
    making the change]

Given that the feature is going to be requested more and more, and
given that Microsoft's getaddrinfo emulation is likely more correct,
thread-safe, etc. than our own, I'm still leaning towards
"include IPv6".

> - Should there be any attempts to detect this support in the header
> files automatically (black magic to me), or should the platform sdk be
> required to compile Python with IPv6?

If this is implemented for 2.3, I think it should be an easily-tunable
option, defaulting to "on" - anybody building Python without the SDK
would have to turn it off.

For 2.4, I hope we will move to VC7, in which case the SDK is not 
required anymore for IPv6.

Regards,
Martin