Currently translated at 85.8% (1981 of 2308 strings)
Co-authored-by: Anthony V <anthonyv156@outlook.com>
Co-authored-by: Hosted Weblate <hosted@weblate.org>
Translate-URL: https://hosted.weblate.org/projects/mumble/mumble-client/el/
Translation: Mumble/Mumble Client
Convert our network protocol docs from rst (ReStructuredText) to md (Markdown) format.
---
This changeset is a followup to PR #6623. In that PR, which first integrated these docs into this repository, the comments indicated consensus that the docs should be in Markdown format to match our other docs.
This changeset is based on PR #6714, which does some baseline corrections, and which should be merged first.
---
Two tables are not converted to Markdown tables because they are not flat tables, which is hard, if not impossible, to represent in Markdown. Preformatted code blocks are being used for them instead.
Otherwise, the conversion seems straightforward.
The changeset includes individual commits by change type, so each kind of change can be traced. The first commit is purely a rename to ensure Git follows the content history across the file conversion.
* Replace non-breaking space ("nbsp") characters with normal space characters
* Fix rst cross-document reference that worked on readthedocs but does not work on GitHub rendering
These changes can be considered a follow-up or fixup to PR #6623, which integrated the network protocol docs into this repository.
This is done as a separate commit to retain Git change file history association. (Git does not store renames. It assumes renames according to content similarity.)
The cross-document reference apparently worked on readthedocs (our previous docs), but does not work on GitHub.
The cross-document figure reference is replaced with a file reference.
Sinks:
- OutputDebugString() (for debugger on Windows)
- Standard output stream (stdout)
- File
- Developer console (Qt widget)
Improvements compared to our old logging system based on Qt:
- Duplicate messages are skipped if less than 5 seconds have passed. When that happens, an informative message is printed.
- The previous format/pattern is retained, but the log level (e.g. "<W>") is colored based on the severity.
- The log is now written to a file on all platforms, not only Windows and macOS.
- When the log file's size reaches 5 MiB, the sink renames it and creates a new one. Up to 4 files are kept:
Console.txt -> Console.1.txt
Console.1.txt -> Console.2.txt
Console.2.txt -> Console.3.txt
Console.3.txt -> <delete>
- We can eventually create multiple loggers, ideally one for each facility (Audio, ServerHandler, etc.).
That would allow us to quickly filter log messages and know right away which subsystem is responsible for them.
- Modern C++ string formatting. This of course only applies to the new functions such as log::debug(), not qDebug().
Only one "feature" is not reimplemented: showing a dialog in case of a fatal error, right before exiting the program.
However, that only worked on Windows and macOS using their native API, to avoid depending on a Qt event loop.
A user reported that sometimes the client log gets spammed with repeated messages such as:
<W>2025-01-10 10:24:10.638 ALSAAudioInput: No such device: No such device
The error corresponds to ENODEV (19), which we treat like any other.
Since it's unrecoverable, the code gets stuck in a full speed loop until the audio engine is stopped.
This commit:
1. Switches from snd_pcm_prepare() to snd_pcm_recover(), which is specifically designed to recover streams.
2. Breaks the loop if the error is not one of the only three recoverable (EINTR, EPIPE, ESTRPIPE). Example log:
<W>2025-01-14 06:16:44.494 ALSAAudioInput encountered unrecoverable error: File descriptor in bad state -> exiting...
Please note that also ALSAAudioOutput doesn't explicitly handle fatal errors in its loop, but it shouldn't run "infinitely" at full speed because poll-based.
Previously, the Register ACL was required to get extended user
statistics (which includes used Mumble version, IP address etc.).
However, the Register ACL was deemed to be a rather arbitrary choice for
this. Instead, the Ban ACL was chosen as access to information such as
packet loss, IP address and used Mumble version and OS seem much more
relevant in the case of banning clients than it is for registering them.
Also, Ban permission is likely to be a better proxy for whether or not
someone is a moderator/admin on a given server than Register privilege.
Fixes#6697
BIO_set_close with BIO_NOCLOSE argument leads to OpenSSL not (fully)
deleting the allocated BIO struct under the assumption that the user
code has taken ownership of it. However, in our case, this is not the
case and therefore OpenSSL should do the deletion as usual.
The flag was probably introduced under the assumption that the component
that either is or isn't deleted by OpenSSL was the externally provided
buffer that is wrapped into a BIO object via BIO_new_mem_buf. However,
this is not the case. OpenSSL doesn't take ownership of the provided
buffer and therefore also doesn't delete it.
Closes#6603
cd068fe3ee added `RestrictAddressFamilies=~AF_PACKET AF_NETLINK`
which was sourced from ba167269f8/trunk/murmur.service
where it was added in d609d47916
described as "hardening" without further reasoning or information.
The disallow of `AF_LINK` leads to continuous spamming of `Address family not supported by protocol` in the systlog. Dropping this resolves that.
Resolves#6180
If a plugin is providing positional audio data and it becomes unloaded,
such as by disabling the plugin in the settings GUI, the plugin manager
may still interact with the plugin through m_activePositionalDataPlugin.
This looks unintentional and, in debug mode, this fails an assertion in
the plugin fairly quickly: `assertPluginLoaded(this)`.
Additionally, in case Mumble shuts down while there exists an active
positional data plugin, Mumble will (now) try to log that the plugin has
lost link by calling the respective function on the global Log instance.
However, at the point where the plugin manager object is destroyed, the
log has already been deleted, making this produce a segmentation fault.
To prevent this from happening, an explicit check for whether the Log
instance still exists has been added to the logging logic.
Fixes#6549
In a Qt application, the QApplication object should be the first QObject
that is created. However, the `Meta` class used to have a static member
called `mp`, which means that this member gets initialized _before_
main() runs and therefore before the QApplication is created.
This has caused an "Invalid nullptr in QObject::connect" warning
somewhere in Qt's internals (since the move to Qt 6). The impact of this
warning is unclear at this point.
This commit makes the mp parameter a std::unique_ptr that gets
explicitly initialized in the main function (more or less right after
the QApplication object is created). This guarantees that the MetaParams
object does not get created before the QApplication object, fixing the
observed warning.
It is worth noting that we do have a couple of other static QObject
variables in the main translation unit, but these seem to be
inconsequential (at least they don't seem to trigger a similar warning).
Fixes#6669
This commit introduces the "--hidden" cli option which prevents
Mumble and the ConnectDialog to show up on startup.
This is especially useful for users who want to automate the
startup process without human interaction.
Fixes#3879
Previously, only "window highlighting" was supported
by Mumble. However, when Mumble was minimized to tray,
there was no way to observe that highlighting.
As requested in #4584, this commit introduces highlighting
to the tray icon. When a highlight message is received, the
tray icon will flash the "information icon" every two seconds
until the MainWindow of Mumble receives focus again.
To enable this for a specific message type, the user needs to
check the "highlight" box in the "Messages" configuration dialog.
Fixes#4584
The old tray icon implementation was very old and contained
a lot of workaround for things that probably are no longer an issue.
Furthermore, the event loop was modified in a way such that it could
end up in an infinite loop draining CPU time and rendering Mumble
unusable.
Based on the new Qt5 implementation, this commit introduces
a complete rewrite of the tray icon. The following things should be noted:
* We assume the information in the Qt documentation [1] is valid. This
means that all versions of Windows, all Linux window managers/compositors
that implement the d-bus StatusNotifierItem specification, and all versions
of macos support the functionality of QSystemTrayIcon and its notification
system. That means we can drop the platform-specific code branches and handle
messages directly with QSystemTrayIcon::sendMessage. This should for example
also be true for recent versions of Gnome, which do not have an actual system
tray, but implement the d-bus StatusNotifierItem specification. Therefore, we
can actually merge and simplify the notification code for Windows and Unix*.
* With regards to the bullet point above, we only limit the "hide to tray"
functionality behind QSystemTrayIcon::isSystemTrayAvailable (because otherwise
you would not get the Mumble window back without binding a shortcut first).
Other code branches that were previously limited when isSystemTrayAvailable
returned false were removed. According to Qt, the QSystemTrayIcon code does not
actually care if a system tray is available and will even retroactively add itself
if a tray becomes available after the application was started.
* On (X)Wayland, the minimize button in the window frame does not trigger a
minimize change event. This means that users with such a system may only be
able to "hide to tray" by 1) pressing the close button in the window frame and
enabling "minimize instead of close" 2) clicking the tray icon or the tray icon
hide action or 3) binding a shortcut to hide the window. This is either a bug
or a deliberate decision by Qt or Wayland and we have no way to do anything
about that. (QTBUG-74310)
* The "messageClicked" event is buggy in Qt on some platforms. That means that
clicking the system notification spawned by Mumble via QSystemTrayIcon::sendMessage
will (on some systems) never trigger anything especially not showing and activating
the window. This is a long-standing bug in Qt (QTBUG-87329), but we have absolutely
no way to work around this. The event is correctly hooked up in Mumble and if
this is ever fixed in Qt, this will start working again automatically.
* The tray icon has been redesigned according to state-of-the-art tray icon
design guidelines [2]. Which basically just means: 1) d9a2d470ab has
been reverted to provide the user with a consistent menu 2) The main action of
the tray icon (toggle show/hide) is the first entry in the context menu
and the default action when the icon is clicked and 3) the TalkingUI toggle
action was added. Actions for double and middle mouse clicks were removed as
they might have contributed to infinite loops.
* There is no way in Windows to show and activate a window that is
not part of the current active process. If you have Mumble running in
the background and receive a message, we can not raise the Window without
you clicking the Mumble taskbar item or tray icon yourself. This
is deliberate by Microsoft and can and should not be circumvented. (#5701)
* This also fixes the case where the Mumble MainWindow would disappear when
pressing "OK" in the settings dialog. This happened because users would have
"minimize to tray" and "minimize on close" enabled.
[1]
https://doc.qt.io/qt-5/qsystemtrayicon.html#detailshttps://doc.qt.io/qt-6/qsystemtrayicon.html#details
[2] https://learn.microsoft.com/en-us/windows/win32/uxguide/winenv-notificationFixes#1486Fixes#3028Fixes#3722Fixes#3977Fixes#3999Fixes#5012
