Add dual stack IPv6 support.

[jgosmann]

This implements support for IPv6 with a dual stack solution supporting IPv4 and IPv6 at the same time. In theory this also allows to listen on multiple ports or network interfaces at the same time. But this is not exposed within main.py, so the nengo command only allows to listen on all localhost addresses (no password protection) or all interfaces (with password protection). I assume this to be fine in almost all cases and do not intend to change this as long as there are no such requests because it would complicate the interface without need.

Due to the current implementation of the WebSocket origin checking the GUI does not support access by IP address (e.g. 127.0.0.1 or [::]), except when started with a password.

Addresses #21.

[tcstewar]

How does this PR interact with #808? Should I merge #808 first (which I really should have done months ago), and then look at this? Or are they pretty independent?

[jgosmann]

Completely forgot about that one. I'd suggest merging this one first. They are independent for the most part, but there are a few things in #808 which I now think are not right to do (changing “localhost” to “::”, switching the whole server to AF_INET6 with no fallback to AF_INET). It's probably easier to sort those out after rebasing on this PR.

The changes in #808 should allow the access via IP address together with this PR.

[tcstewar]

For my Windows 10 laptop, this fails to allow connections. It looks like self.bindings is empty. self.server_name is localhost and then this for loop fails to find anything to use: https://github.com/nengo/nengo_gui/pull/864/files#diff-6b004fd2ea83a7abcc0443f852b14d6fR149

[jgosmann]

What does the socket.getaddrinfo(*server_address) call return? (Or as an approximation what does socket.getaddrinfo('localhost', 8080) return?

[tcstewar]

What does the socket.getaddrinfo(*server_address) call return? (Or as an approximation what does socket.getaddrinfo('localhost', 8080) return?

Here's the result:

[(<AddressFamily.AF_INET6: 23>, 0, 0, '', ('::1', 8080, 0, 0)), (<AddressFamily.AF_INET: 2>, 0, 0, '', ('127.0.0.1', 8080))]

[jgosmann]

That problem should be fixed (but there might be more).

[tcstewar]

That problem should be fixed (but there might be more).

:) Here's the new error:

Traceback (most recent call last):
  File "C:\Users\Terry\py3\Scripts\nengo-script.py", line 11, in <module>
    load_entry_point('nengo-gui', 'console_scripts', 'nengo')()
  File "c:\users\terry\documents\github\nengo_gui\nengo_gui\main.py", line 77, in main
    page_settings=page_settings)
  File "c:\users\terry\documents\github\nengo_gui\nengo_gui\gui.py", line 43, in __init__
    self.model_context, server_settings, page_settings)
  File "c:\users\terry\documents\github\nengo_gui\nengo_gui\guibackend.py", line 366, in __init__
    self, self.settings.listen_addr, GuiRequestHandler)
  File "c:\users\terry\documents\github\nengo_gui\nengo_gui\server.py", line 240, in __init__
    DualStackHttpServer.__init__(self, *args, **kwargs)
  File "c:\users\terry\documents\github\nengo_gui\nengo_gui\server.py", line 156, in __init__
    self.bindings.append(self.Binding(fam, addr))
  File "c:\users\terry\documents\github\nengo_gui\nengo_gui\server.py", line 126, in __init__
    socket.IPPROTO_IPV6, socket.IPV6_V6ONLY, 1)
AttributeError: module 'socket' has no attribute 'IPPROTO_IPV6'

[jgosmann]

But socket.AF_INET6 is defined and socket.has_ipv6 is True?

[jgosmann]

Hm, this seems to be a known Python bug?

[jgosmann]

Does the socket module have socket.IPPROTO_IPV4?

[jgosmann]

The magic value to add is 41. (I suppose it's not 42 because we start counting at 0? 😉 )

[jgosmann]

Should be fixed.

[tcstewar]

Should be fixed.

THat's now working for me in Windows 10 (py27, py35, py36)... I'm going to try it on a Windows 8 machine tonight....

[tcstewar]

This also works fine on Windows 8... :)

However, Travis-CI seems upset by something about this... is this a problem with Travis-CI or with this PR?

[tcstewar]

I can try rebasing to master (and force-pushing) and seeing if that helps....

[jgosmann]

Probably travis... I think the tests have been broken for a while? Can't verify right now ...

…

On October 30, 2017 7:11:34 PM EDT, tcstewar ***@***.***> wrote: This also works fine on Windows 8... :) However, Travis-CI seems upset by something about this... is this a problem with Travis-CI or with this PR? -- You are receiving this because you were assigned. Reply to this email directly or view it on GitHub: #864 (comment)

-- Sent from my Android device with K-9 Mail. Please excuse my brevity.

[tcstewar]

Probably travis... I think the tests have been broken for a while? Can't verify right now ...

I'll do the rebase and see if that helps...

[tcstewar]

(er, never mind, you're still assigned to it so I shouldn't do a force-push on it.... especially since I just asked you about that today! :) )

[jgosmann]

you're still assigned

Sorry, forgot to unassign myself after adding the last fix.

The tests are failing on the same way on master and the last succesful “build” has been 7 months ago ...

[jgosmann]

Related PR #839.

[tcstewar]

I'm all set with this.. thank you for your patience on this one.... :)