* Compensate beat note length when stretching
We allow stretching beat notes to normal notes but the length starts
from -192 so there is a lag for one whole note before any change is
seen. Compensate by setting the oldNote value to 1 when stretching
if the note is 0 or below in length.
Co-authored-by: Spekular <Spekular@users.noreply.github.com>
This PR fixes issues on systems where `QCursor::setPos()` has no effect
or is not reliable. These issues included knobs moving to fast on some
operating systems. Affected widgets are `Knob` and `LcdSpinBox`.
With this PR, on all operating systems, the `setPos` calls are removed and
the cursor is not hidden anymore, so the mouse keeps moving normally
when changing values of one of the widgets.
As now the previous pointer position keeps moving (instead of being reset
to the original position using `QCursor::setPos`), the mathematics that
translate pointer pixel distance to `Knob`/`LcdSpinBox` value increase
have to be changed:
* The `Knob` transition function is now linear and uses a new factor.
* `LcdSpinBox` now uses float values and saves the current float remainder
(this is actually a separate issue revealed by this fix), leading to a fluent,
non hanging movement.
This atomically unsets the MidiJack reference in AudioJack right before
MidiJack is destroyed. This avoids AudioJack using a destroyed MidiJack
object.
This patch
* makes `m_stopped` atomic
* initializes `m_stopped` correctly to `true`
* moves the initialization of `m_stopped` to the point where jack ports
are already connected
When selecting a Piano Key to mark semitones in the Piano Roll we
select key from the y position of the pop-up menu and not the mouse.
Incidentally these two are most often the same as the menu builds
from the mouse y positon and down. If there is room for it. If there
is no room downward it will create the menu so the lower part of the
frame aligns with the mouse y position.
Fixed by creating a variable to hold the pressed key before creating
the menu.
Knob::friendlyUpdate() can be called after the model is deleted
due to signal-slot connections.
Adding a check for the model fixes a crash due to null pointer dereference.
`PianoRoll::mouseDoubleClickEvent` wasn't forwarding the event to the base class when not acting on the event. The base class calls `mousePressEvent`.
Fixes#3005
Calling via QMetaObject::invokeMethod should be thread safe.
Crash callstack:
QWidget::move
SongEditor::updatePosition
Song::stop
Song::stopExport
ProjectRenderer::run
QThreadPrivate::start
right place when it not played from the begining.
That has created a difference between the ticks and the metronome and
the sample track.
The cause of the problem was that the calculation of the frame to play
was wrong: we had calculated `framesPerTick` according to the current
engine's sample rate instead of the SampleBuffer's sample rate.
To avoid a race condition between the gui thread which destroys the
samplebuffer and the mixer thread which increases the buffer's ref-
count, we'll make sure to touch the ref-count only when wh're synced
with the mixer.
NotePlayHandleManager::acquire uses a read lock unless the pool is empty.
If two threads try to acquire NotePlayHandle simultaneously
when the value of s_availableIndex is 1, one thread will try to read s_available[-1].
If the acquire action and the release action are done at the same time,
NotePlayHandleManager::acquire may try to read data
before NotePlayHandleManager::release actually writes.
This commit prevents them by always using the write lock when acquiring a NotePlayHandle.
