layer.effect causes warnings on reload for an unknown reason which
seems to be ownership or destruction time related. This commit uses
an alternate strategy to create the shader which does not show this
warning.
ShaderEffectSource as a property not parented to an item does not
update its sourceItem's QQuickWindow when its own is changed. This
lead to use after frees and broken effects when using ClippingRectangle.
This change requires more QtPrivate usage but eliminates generation or
cleanup related window incubation controller bugs. Additionally it
enables async loads prior to rendering windows.
- Use `copy_file_range(2)` over `sendfile(2)` which has wider
compatibility.
- Special case pam on freebsd and document `configDirectory`
incompatibility.
- Disable jemalloc for FreeBSD by default as it is the system allocator.
- Disable breakpad by default on FreeBSD as breakpad is not supported.
For bluez audio streams and potentially other types of synthetic
device, volume controls are managed via the node and persistence is
handled by the service.
The openwindow event format is ADDRESS,WORKSPACE,CLASS,TITLE but the
handler was parsing args.at(2) as title and args.at(3) as class,
which is reversed.
This caused windows to display their class name instead of their
actual title when the openwindow event arrived after windowtitlev2,
since updateInitial would overwrite the correct title with the class.
Nodes referencing a device can be bound later than the device is
bound. If this happens, the node will not receive an initial route
device volume change event. This change caches the last known route
device volume and initializes the device with it if present.
Fixes a bug introduced in 3e2ce40 where a directory imported with a
"../name" path import would be passed to scanDir as ending in '/' which
created an invalid duplicate scan entry.
Previously a hardcoded 0.0001 offset was used to determine if a volume
change was significant enough to send to a device, however some
devices have a much more granular step size, which caused future
volume updates to be blocked.
This change replaces the hardcoded offset with the volumeStep device
route property which should be large enough for the device to work with.
Fixes #279
