As discovered by @michaelgregorious & @zonkmachine, applyRelease's
condition to determine whether the release ramp starts within the
current buffer was off by 1 frame, running the release code when the
ramp should only start in frame 0 of the next buffer. This could cause
the ramp to be miscalculated, starting at a value greater than 1.0 and
and thus actually amplifying the signal.
This fixes to bugs leading to clicks on instrument note-off in most
instruments.
The first bug was introduced as part of a refactoring done by Vesa [1]
and caused each note play handle's last buffer being dropped because
the play handles were deleted before being processed in AudioPort.
Thanks to @lleroy for pointing this out. [2]
The second bug / typo has always been there [3] and was a misplaced
parenthesis in Instrument::applyRelease breaking the calculation of the
note-off envelope's start frame, sometimes putting it outside of the
buffer.
Fixes#3086
[1] 857de8d2c8 and related commits
[1] https://github.com/LMMS/lmms/issues/3086#issuecomment-519087089
[2] 02433380c6
* Mallets - Add random knob function
Implement randomness for the instrument. When Random is applied, Hardness and Position of instrument 1 - 9, or Modulator and Crossfade on instrument 10 (Tubular Bells), are nudged on every note to liven up the sound. With the modulated knobs placed at their center values, Random at max will select from the full range of possible values.
Co-authored-by: saker <sakertooth@gmail.com>
Co-authored-by: Dominic Clark <mrdomclark@gmail.com>
lilv can return the scale points in randomized order, and so the linked
models (of stereo effects wit 2 `Lv2Proc`) may have different scale
points. This would mean that the same combo enumeration value would have
different float values in linked models. This problem is fixed by
sorting the scale values.
Also, it is more natural for users if scale points are enumerated in a
numerical order.
This displays a warning dialog if the users requests unusual
buffersizes:
- buffersizes less than 32
- buffersizes which are not a (natural number) power of 2
This commit also replaces some `setGeometry` stuff by `QBoxLayout`.
This improves the way digits are calculated for display in
`Lv2ViewProc`:
- More than 2 digits are now recognized
- Minus signs are now recognized
- Tests have been added
Adjust the classic theme so that it is consistent with the changes made to the default theme.
It was only necessary to change the font-size of the check boxes in the file browser to points because the QTreeView already does not contain any font settings.
Replace the check for a `nullptr` buffer with a check that checks if the NotePlayHandle uses a buffer. This is effectively the same condition that's used by `PlayHandle::doProcessing` to decide if it should call play with a `nullptr` in the first place. Hence it should be the most fitting check.
Remove whitespace in methods and add it to the parenthesis of some statements. Remove underscores from parameter names. Remove usage of `this` pointer.
Add `auto` keyword to shorten line length in `InstrumentPlayHandle::play`.
Move `const_cast` of `NotePlayHandle` closer to the `process` method as this method is the reason that the cast is needed in the first place. One might also add a version of `nphsOfInstrumentTrack` that returns a non-const result but it seems to be a general problem of a missing clear responsibility so I keep it as is.
Remove identical calls to `InstrumentTrack::processAudioBuffer` which appear in the overridden implementations of `Instrument::play` and `Instrument::playNotes`. Instead the call to `processAudioBuffer` has been moved into `InstrumentPlayHandle::play` and `InstrumentTrack::playNote`. These two methods call the aforementioned methods of `Instrument`. Especially in the case of `InstrumentTrack::playNote` the previous implementation resulted in some unncessary "ping pong" where `InstrumentTrack` called a method on an `Instrument` which then in turn called a method on `InstrumentTrack`. And this was done in almost every instrument.
In `InstrumentTrack::playNote` an additional check was added which only calls `processAudioBuffer` if the buffer is not `nullptr`. The reason is that under certain circumstances `PlayHandle::doProcessing` calls the `play` method by explicitly passing a `nullptr` as the buffer. This behavior was added with commit 7bc97f5d5b. Because it is unknown if this was done for some side effects the code was adjusted so that it behaves identical in this case.
Move the complex implementation for `InstrumentPlayHandle::play` and `InstrumentPlayHandle::isFromTrack` into the cpp file and optimize the includes.
* FileDialog : Adding mounted volumes on Linux
* Adding some file systems types and collapsing 2 'if' statements in one
* Removing the foreach
* Correcting PREPROCESSOR directive
* fixes#6354: Sample and Hold for LFO Controller
LFO controller's "white noise" wave shape didn't respect the frequency knob at all, so
Sample-and-Hold was added to extend the functionality of the LFO Controller with this
random waveshape. The original functionallity can still be accessed by setting the
FREQ knob to minimum (0.01)
---------
Co-authored-by: Kevin Zander <veratil@gmail.com>
Co-authored-by: saker <sakertooth@gmail.com>
* Change the title for SideBarWidgets to be vertically centered related to icon and with no underlining
* Update src/gui/SideBarWidget.cpp
Co-authored-by: saker <sakertooth@gmail.com>
* Updating FileBrowser display when searching : showing matching files
* Correcting increment and removing duplicated calls
* Correcting reload tree when filter is activated
---------
Co-authored-by: saker <sakertooth@gmail.com>
* Add options to enable sanitizers
* Specify them as debugging options
* Apply suggestions made by Dom
* Fix linking issues
There were linking issues, most likely because we were
applying the sanitizer flags to unnecessary targets that
are part of the build. Now, we only focus on adding the
compiler flags to lmmsobjs as a PUBLIC dependency,
and selectively choose what targets we need to apply
the linker flags to as PRIVATE dependencies.
* Add UBSan
* Add status messages
* Remove GNU as supported compiler for MSan
* Revert to using add_<compile|link>_options again
* Fix sanitizer linkage within veal and cmt libraries
I suspect that our sanitizer flags were removed for these two CMake
targets. Instead of setting the properties directly, we call
target_compile_options and target_link_options
instead.
* Remove TODO comment
* Revert "Fix sanitizer linkage within veal and cmt libraries"
This reverts commit b04dce8d8c.
* Use CMAKE_<LANG>_FLAGS_<CONFIG> and apply fixes
* Remove quotes
* Add support for additional flags
* Disable vptr for UBSan
* Print "Debug using" in status for clarity
* Wrap ${supported_compilers} and ${status_flag} in quotes
* Replace semicolons with spaces in additional_flags
* Remove platform check for no-undefined flag
The platform check was added as an attempt
to make this branch compile with the veal
and cmt libraries. However, it seems to work
without it, so the problem must've been something
else occuring in these files.
* Undo change to FP exceptions status message
* Use spaces instead of tabs
* Remove -no-undefined flag for cmt and veal libraries
The quantPos arg of AutomationClip::setDragValue causes the node time
to be quantized before its looked up in the timeMap iterator.
This results in the node not being found and a new one being created
inside the setDragValue function even though we had found one.
Simply setting it to true causes this bug, simply setting it to false
causes a new node to be created off grid/not snapped. So the fix
is to quantize the position for the lookup only if we haven't found an
existing node under the cursor.
* Profiler rework
* Workaround for GCC bug
* Rollback QFile removal
* Use enum instead of a plain index to describe detailed stats
* Use the GCC workaround code for all compilers to avoid redundancy
* Update and fix comments
* Implement suggestions from review
* Split AudioEngine::renderNextBuffer() into separate functions, fix old formatting
* Remove QFile include
* Revert formatting changes
* Apply suggestion from review (remove unnecessary template parameter)
* Revert more formatting changes
* Convert DetailType to enum class
* DetailType enum class improvements suggested in review
* Use std::atomic for m_detailLoad
* RAII-style profiler probes
* Apply suggestions from code review
Co-authored-by: Dominic Clark <mrdomclark@gmail.com>
* Fix namespace comment
* Improve CPU load widget precision and use floats for load computations
* Atomic m_cpuLoad
* Add custom step size support for CPULoadWidget
* Apply suggestions from review (convert the profiler probe into a nested class, other small changes)
* Do not limit stored load averages to 100%
---------
Co-authored-by: sakertooth <sakertooth@gmail.com>
Co-authored-by: Dominic Clark <mrdomclark@gmail.com>
* Fix LADSPA effects memory leak
* Fix buffer overflow in PianoView
* Avoid using invalid iterators in AutomationClip
* Fix memory leaks in SimpleTextFloat
* Handle potential case where QMap::lowerBound(...) returns end iterator
* Implement suggestions from review
