E.g. in case of a disconnected thermistor, or when it is really
really cold. Make sure our parsing doesn't get confuse by this and
thinks that
ok T:150.0/210.0 T0:-55.7/0 T1:150.0/210.0
is actually only T and T1 being reported by the firmware, causing
stuff to be wrongly canonicalized thanks to T/T1 being Smoothie's way
of reporting dual extruder setups.
Fixes#2007
Introduced a custom send queue type that actually contains two queues,
one for resends, one for regular lines. A flag indicates whether
lines should be returned from both or only resends. That way we
ensure that as soon as we have an active resend request we ignore
what was already in the queue and only send the lines we need to resend.
Also: PrependQueue => PrependableQueue
Less tracking of offsets and also more similar to how
firmware does this.
We rewrite our current position on tool change to current position
plus tool offset instead of applying it for every single
position change.
We fetch settings once explicitly after passive login but before
completed startup. Without this patch we fetched them again after
onUserLoggedIn got fired (if it got fired by the passive login)
during startup. That's not necessary because we did the passive login
before our initial settings fetch, so IF that already logged us in,
our settings fetch already was done with us logged in as well and
the onUserLoggedIn later only tells us what we already knew.
We still had some concurrent requests triggered through onStartup that
managed to overtake the passive login (and hence nuke the session
data on the client). This should now really not be possible any longer.
Also created a sequence diagram of the new process and linked it from
a lengthy comment.
It caused issues with our line number tracking in the browser
developer tools and I don't see any way to avoid this considering
that we can't adjust the arguments supplied to console.log without in
fact wrapping it :/
Under specific circumstances it could happen that the passive login
response was overtaken by a response from an earlier still
unauthenticated request which would then effectively overwrite the
session cookie immediately and hence cause the browser to use its
login session against the server, but without displaying that to the
user. See #1881 for what is probably an issue caused by exactly this
kind of scenario.
Additionally a couple of requests needed to be done a second time
after login, just moments after having done them for an anonymous
user. See #1572 for a request to change this behaviour.
This patch changes the startup order of the web interface like this:
1. connect to the websocket, postpone any events triggered by
this (e.g. "fromHistoryData") until 3
2. perform a passive login, postpone any events triggered by this
(e.g. "onUserLoggedIn", "onUserLoggedOut") until 4
3. trigger postponed events from 1
4. trigger "onStartup" (triggers postponed events from 2)
5. fetch settings
6. bind view models etc
A connection close of the websocket will disable event processing on the
socket until it is marked as initialized again by the passive login
processing, which will also be triggered immediately on server connect.
That way under normal circumstances nothing should ever get triggered
in the registered view models that might generated requests to the
server API until a passive login has been done, a server session
initialized and if necessary a user properly authenticated.
