This test setup utilizes the "mocked" configuration so far only used for
manual testing. This allows adding unit tests for models without having to
run Syncthing itself. The test is pretty basic itself at this point.
This is the same as the feature already present in the launcher and
connection settings. Not sure how much sense it makes for systemd services
as they run rather independently from the UI but maybe it is useful.
Lazy-initialize the network information backend so warnings it possibly
shows don't disturb the output of the CLI (which does not need the network
information backend anyway).
* Allow to pause all devices on metered network connections (devices will
be automatically resumed when network is no longer metered)
* Allow to stop Syncthing when it was started via the built-in launcher
on metered network connections (it will be automatically started again
when the network connection is no longer metered)
* See https://github.com/Martchus/syncthingtray/issues/231
* Unify code paths for compiling expected SSL errors so in any case the
expected errors are including the error types emitted by the Schannel
backend
* See https://github.com/Martchus/syncthingtray/issues/223
Don't go though the list of dirs/devs for every dir/dev change event. Just
go though the list once after receiving new events. This makes also the
code a little bit easier to follow.
Apparently the default before introducing the request timeout was not 30000
(which is the value of QNetworkRequest::TransferTimeoutConstant). There was
actually no timeout. So let's restore this behavior for now as the main
intention was to make the timeout configurable but not change any defaults.
When connecting again, no data is cleared. Since we now keep
track of event IDs and not just timestamps it makes sense to not
forget what the last event IDs were. To still run into the
"limit" case the flags for having events at all can be used
(which is also reset on just `connect()`).
I have observed that Syncthing Tray can get stuck thinking a remote device
still needs data. Likely the update got lost. The code contains certain
conditions in which folder completion events and requesting completion are
supressed. Those conditions are based on timestamps. That is not ideal as
the accuracy is only one second (so different timestamps might be
considered equal as they are rounded to be the same). Additionally, it
makes no sense to assume a timestamp upon receiving a response as the
information might be older than the time it was received.
This change avoids those conditions. It rather uses the event ID to
decide what event/reply is newer.
This change also uses quint64 instead of int for event IDs to avoid running
into an overflow (or rather conversion error) when deserializing the ID
which might be bigger than int. (Not sure how big the ID can become; this
is to be on the safe side.)
Martchus
John Veness