+ Since SCI adds filters, but init() isn't run till later, that means
the context attribute for the WebSocketUpgradeFilter isn't present
during jsr356 runs. Added ability for manual filter creation to call
setToAttribute() as a pre-init step, allowing the init() itself to
bypass the set to attribute for that specific filter instance.
+ This also means ServletException is now thrown out from the
various configureContext() static methods.
+ After talking it through with Simone, swapping out 'global' init-param
with a bit more robust 'contextAttributeKey' to handle the automatic
context.setAttribute() of the filter itself.
As simply having a filter in the web.xml makes it alive, but nothing
is wired up into it, and accessing the filter instance via the
context metadata seems impossible. So we made the init-param for
'contextAttributeKey' important and required, but with defaulting
and validation checks.
+ Making key also work inside of WEB-INF/web.xml via context params
+ Making WebSocketUpgradeFilter generic enough to be used in a
web.xml descriptor
+ Adding global={bool} init-param on WebSocketUpgradeFilter to aid
library developers and end users more ways to tweak the filter
order
+ Fixing onOpen context classloader to be that of the context
that started the WebSocketUpgradeFilter (which will be the
same as the WebAppContext in most cases)
* Using Servlet 3.1's ServletContext.addFilter() mechanisms with its
FilterRegistration to encourage the WebSocketUpgradeFilter to be
first in the filter chain.
Note: this is possible, but can be overridden by another call
to the same ServletContext.addFilter() using the same techniques
to put something in front of the WebSocketUpgradeFilter.
+ Improved test case handling of frame reading (to prevent future
false failures)
+ Improved AbstractWebSocketConnection handling of close -> disconnect
to only trigger on successful write (or failure) of close frame flush
attempt.
+ Flusher close on all forms of disconnect, even half-closed.
+ Fixing bad assumptions in unit tests where the client would send a few
frames to test the protocol behavior followed by the close frame.
But the test expected the server to initiate the close, but this setup
of the tests would mean that the client initiated the close.
Jesse McConnell
Simone Bordet
Joakim Erdfelt
Greg Wilkins