We have a 4to6 tunnel set up on the server doing the checking so we can actually see IPv6 -only mirrors, and also end up using IPv6 for any mirror supporting them.

These pages are new and not very filled out yet:
http://www.archlinux.org/mirrors/bjtu.edu.cn/

But I can envision them showing a lot more information than what is there now. We can pull the status data in; do the required DNS lookups on the hostnames and show the A/AAAA records (which will show whether IPv4/IPv6 support is there), etc.

If you know Python and/or Django, you are more than welcome to chip in here. Steps I can think of:

1. Add a migration to add has_ipv4, has_ipv6 columns (or something like that) to the MirrorUrl model.
2. Add another management command we could run like once a day that does DNS lookups on all of the mirror URLs, and then updates the columns accordingly based on what info came back.
3. Possibly persist this DNS information in another table, or just retrieve it on the fly.
4. Display DNS info and more on the mirror details page.
5. Allow filtering of the generated mirrorlist based on IPv4/IPv6 status (not much different than the FTP/HTTP filter).

After looking at the code I get the impression that the DNS lookups could easily be added to mirrorcheck.py. (urllib looks them up anyway) Saving the results of the lookup in MirrorLog would also be sensible.
What do you mean with "DNS information" in 3.? Are you planning to save the IP-addresses in a table?

That was a thought, that way we can implement #4 without much hassle.

This works pretty well for getting IP addresses:
In [15]: [i[4] for i in socket.getaddrinfo('testmyipv6.com', 80, 0, socket.SOCK_STREAM)]
Out[15]: [('24.73.171.238', 80), ('2001:4830:2502:8002::ac10:a', 80, 0, 0)]

Not sure how much we are going to have to do, but this might be helpful:
http://google-opensource.blogspot.com/2008/10/ipaddrpy-flexible-and-easy-python-ip.html

How about that:

>>> import socket
>>> from urlparse import urlparse
>>> url = "http://archlinux.de/"
>>> families = [x[0] for x in socket.getaddrinfo(urlparse(url).hostname, None)]
>>> socket.AF_INET in families
True
>>> socket.AF_INET6 in families
True

EDIT:
The attached patch implements #1 to #4, but is work in progress.
I did not use threads for mirrorresolv.py. Are long running manage.py-commands blocking the server? If not, I see no reason to add threading.

I now also implemented #5. (Patch attached)

In my opinion the handling of the Country-input in /mirrorlist/ is not user friendly at the moment:
Choosing 'Any' should result in a list of all mirrors, and not in a list of all mirrors with unspecified ('Any') country-property.
I suggest to fix that like I did it in the patch:
If the country input is 'Any', disable filtering by country.

To make the 'Any' MultipleChoiceField-entry independent from the countries actually used by the Mirrors, it is now manually added to the form if it was not added with make_choice().

Added logging to mirrorresolv.py and fixed sorting in /mirrors/status.

I disagree with the 'Any' changes completely. I can no longer choose "Any" and "Canada" for instance and get what I was looking for. This was the idea behind the top art of the mirrorlist update page. Unfortunately, we don't get to make any other selections in that part of the page such as IP version or being able to sort by status.

If we want to do what you propose, don't reuse a value- we should either add a checkbox to the form that indicates we want to use all mirrors, or add an "All" option to the list.

Other comments:
* there is some trailing whitespace in the patch, make sure that is as clean as possible.
* "mirrorUrls" -> "mirror URLs" in log messages, no need to introduce camelcase that doesn't exist elsewhere.
* You're mixing strings and integers in your choice field- it ends up working, but it is probably better to always use strings- ('4', 'IPv4') since you compare on strings later anyway.
* "ipv4_only" is misleading as a function argument name, and the way this works seems wrong/backwards. I would expect the method to take as input "supports IPv4" or "supports IPv6", and you would pass both as true if you could support either. Passing both as false should also return no mirrors.

I did the things you suggested and generated the attached commits:

99bccbb Add 'All' choice and make it the default
145c462 Add has_ipv{4,6} to MirrorUrl
c2763c4 Add mirrorresolv manage.py command
5b941e5 Allow filtering of mirrorlist by IP-version
1699dbe Display information about supported IP-versions

"Passing both as false should also return no mirrors."
That would be nice, but it would make things more complex without adding any value.
People without neither IPv4 nor IPv6 won't need mirrorlists anyway.
At the moment there is no way to call find_mirrors with both values set to False, because the form validation prevents it.

Do you have a github account or something at any other git hosting provider? It might make it easier for me to pull your branch and check it out. I do like this being split into a few patches though, so thanks for that.

You are being shortsighted in your logic with find_mirrors(). Of course I realize that there is only one way it is currently called, but the fact that this stuff didn't even exist until 2 months ago also tells me there is a lot that can be done in the future with it, and I think your argument as to why you shouldn't fix it is a bit silly.

You convinced me to return an empty QSet when called without any IP support at all.
I created a repository at github and pushed my changes:

Github project page: http://github.com/pyropeter/archweb
Git-url for pulling: git://github.com/pyropeter/archweb.git

OK, this is mostly implemented and now live on the main site. You have an email from me about one patch, and I made a few changes along the way to make sure this thing can run in cron without needing to be watched for failures and such. Let me know what you think.

Should be done, for example: https://www.archlinux.org/mirrors/bjtu.edu.cn/