This commit adds a signal to ServerResolver and a slot to MainWindow to handle ServerHandler errors, in order to start the reconnection timer after the first connection attempt.
Fixes#3220.
Change our checks for MUMBLE_NO_PCH to use string comparisons
instead of numerical comparisons.
When MUMBLE_NO_PCH is not set, the numerical equivalent check
fails because nothing is on the left-hand side of the -eq operator.
Fixesmumble-voip/mumble#3314
Our Tavis CI builds ended up taking a long time to complete.
This PR was made to try out whether mirroring pkg.mxe.cc ourselves would improve that situation.
It did. Build times dropped from ~50 minutes to ~15 minutes.
sss123next in mumble-voip/mumble#3294 reports that the current ping
timer logic causes problems with slow TLS handshakes, such as when
connecting to servers with large DH parameters.
This commit ensures pings are not sent before the TLS handshake has
completed.
This was tested against a Grumble instance built against a modified
crypto/tls package that sleeps during the handshake. I verified that
isEncrypted() returns false when the handshake has not finished, and
returns true when the handshake has completed.
Since AsyncNotifyWhenDone has been added to gRPC, IsCancelled is
never safe to call with the async API unless the done tag has arrived.
Now that AsyncNotifyWhenDone is used to clean up cancelled streams,
calling it isn't needed anyway as any cancelled streams are cleaned
up right after they're known to be cancelled.
Our MinGW build was left stale due to mumble-voip/mumble#3208.
Altough our current MinGW builders use Wine as their test runner
-- and were added to allow_failures exactly for that reason --
it makes more sense to get blocked from merging on build failures.
The alternative is that we actively *check* that the MinGW is not
broken, which doesn't happen in practice.
We obviously can't expect Wine to mirror the exact behavior on any
particular Windows version, but let's just look at the MinGW targets
in .travis.yml as a check for whether a) the MinGW build works on Linux,
and b) whether our tests pass using Wine.
Now that test.pri adds source files and headers by default, the .pro files
of our tests need to append to SOURCES and HEADERS to ensure everything
test.pri adds to SOURCES and HEADERS are kept around.
This change fixes AudioOutput to use the float overload of std::abs:
float std::abs(float);
instead of a non-existant template version.
Fixesmumble-voip/mumble#3281
Needs-Backport: 1.2.x
Since we now build with a macOS target of 10.8 when
building against Qt 5.10, we get deprecation warnings
when calling gestalt.
The fix, for now, is to ignore deprecation warnings in
OSInfo on macOS.
This should be addressed in a better way in the future.
For that, mumble-voip/mumble#3290 has been filed.
Qt 5.10 uses features available only on macOS 10.8, such as
std::future<void>.
See the referenced issue for more information.
Fixesmumble-voip/mumble#3288
The comment block that justifies the use of Boost in GlobalShortcut_win
accidently stated that MinGW "does not support
boost:future/boost::promise", which is incorrect.
Fix the comment to say what was intended: that MinGW doesn't
support std::future/std::promise.
This change remove support for loading external images.
That is, this change ensures Mumble doesn't load images from tags in text messages,
comments and channel descriptions via HTTP or HTTPS.
With this change in place, Mumble only supports images that are embedded
in the message via data URLs.
