support websockets in the python client

[Antoine Martin]

We already support websockets as transport in the server, see #1136 / #1134, so we could support websockets in the python client as well. (we already have ssh, ssl, tcp and unix / named-pipes!)

I don't see the point of adding this one: we try hard to avoid copying data in the network layer and using websockets will cause more of it, using more CPU.

[Antoine Martin]

Found: ​https://github.com/liris/websocket-client which supports python 2.
This may help with things like #1298.

[Antoine Martin]

Implemented basic support in r13558, hoping this will help debug #1298.
You can now connect using a websocket transport:

xpra attach ws/127.0.0.1:10000/

Looking at the traffic, it's not pretty: redundant size headers, etc... #1134 may help a bit here, but ultimately this transport is very unlikely to ever perform as well as the plain xpra protocol it wraps.

Still TODO:

• man page updates (maybe revamp the whole display connection string to avoid copying the same information everywhere)
• support parameters on the URL
• maybe expose the underlying TCP socket's details via get_info
• hook ssl options

[Antoine Martin]

Minor updates:

• r13559: added to osx moduleset
• r13560: connection closed handling, extra socket info
• r13562: we can now use this connection string syntax: ws[s]/username:password@host:port/path?arg=val&etc

[Antoine Martin]

• r13594: man page updates
• r13596 seems to allow us to use SSL connections (blocked by #1213), ie with a server configured with a self-signed CA as per wiki/Encryption/SSL:

  xpra attach wss:127.0.0.1:10000 --ssl-ca-cert=./ca.crt
  

(successfully establishes the SSL connection to the server but then fails because of the issue in ticket:1213#comment:3)

---

Excluding the wss issue above, this is ready for testing. See also #1213
@afarr: you should be able to use ws URL connection strings with the latest clients.
This is useful for testing the same transport layer used by the html5 client.

[alas]

Interesting.
Tested the permutations with 1.0.6 r15627 osx client (10.12) against a 1.0.7 r16032 fedora 25 server.

Each permutation worked without so much as a wrinkle. Good to know the clients will work over the same transport layer.

Closing.

[Antoine Martin]

See also #2121