novnc/docs/notes

Some implementation notes:

There is an included flash object (web-socket-js) that is used to
emulate websocket support on browsers without websocket support
(currently only Chrome has WebSocket support).

The performance on Chrome is great. It only takes about 150ms on my
system to render a complete 800x600 hextile frame. Most updates are
partial or use copyrect which is much faster.

When using the flash websocket emulator, packets event aren't always
delivered in order. This may be a bug in the way I'm using the
FABridge interface. Or a bug in FABridge itself, or just a browser
event delivery issue. Anyways, to get around this problem, when the
flash emulator is being used, the client issues the WebSocket GET
request with the "seq_num" variable. This switches on in-band sequence
numbering of the packets in the proxy, and the client has a queue of
out-of-order packets (20 deep currently) that it uses to re-order.
Gross!

The performance on firefox 3.5 is usable. It takes 2.3 seconds to
render a 800x600 hextile frame on my system (the initial connect).
Once the initial first update is done though, it's quite usable (as
above, most updates are partial or copyrect). I haven't tested firefox
3.7 yet, it might be a lot faster.

Opera sucks big time. The flash websocket emulator fails to deliver as
many as 40% of packet events when used on Opera. It may or not be
Opera's fault, but something goes badly wrong at the FABridge
boundary.

Javascript doesn't have a bytearray type, so what you get out of
a WebSocket object is just Javascript strings. I believe that
Javascript has UTF-16 unicode strings and anything sent through the
WebSocket gets converted to UTF-8 and vice-versa. So, one additional
(and necessary) function of the proxy is base64 encoding/decoding what
is sent to/from the browser. Another option that I want to explore is
UTF-8 encoding in the proxy.


Building web-socket-js emulator:

cd include/web-socket-js/flash-src
mxmlc -static-link-runtime-shared-libraries WebSocketMain.as