Test merge qt6 to main#1
Open
bernie-laberge wants to merge 45 commits into
Open
Conversation
### Feature - Qt 6 port # 1 ### Linked issues AcademySoftwareFoundation#367 ### Summarize your change. This is the first phase of the Qt 6 port. It was mostly tested on Rocky Linux 8. There are a lot of generated files under `src/lib/mu/MuQt6`, and some modifications but you don't need to go over that folder. There are known issues that will be addressed in future PR: **All platforms:** - Refactor _QtAudioRenderer_ class. Currently, The _pulse_ audio output device is not enabled, and they are loud white noise issues on MacOS. - Remove the usage of _Core5compat_ and refactor what is deprecated. I don't think there is a lot to do here, but it is mostly about refactor the usage of _QTextCodec_. **MacOS:** - The _Open File dialog_ does not open anymore. - Loud white noise audio (Should be related to _QtAudioRenderer_) - The Central widget is rendered in the bottom left corner, but as soon as you resize the windows, it goes back to normal. **Windows:** - Issue with the compilation of Shiboken6 ### Describe what you have tested and on which operating system. - [ ] Rocky LInux 8 - [ ] MacOS ### Add a list of changes, and note any that might need special attention during the review. --------- Signed-off-by: Cédrik Fuoco <cedrik.fuoco@autodesk.com>
### Rebasing Qt6 feature branch to main Rebasing the feature branch to main to keep it updated. --------- Signed-off-by: Bernard Laberge <bernard.laberge@autodesk.com> Signed-off-by: Cédrik Fuoco <cedrik.fuoco@autodesk.com> Signed-off-by: Will Wira <4399853+williamwira@users.noreply.github.com> Signed-off-by: Éloïse Brosseau <eloise.brosseau@autodesk.com> Signed-off-by: Cédrik Fuoco <105517825+cedrik-fuoco-adsk@users.noreply.github.com> Co-authored-by: Bernard Laberge <117092886+bernie-laberge@users.noreply.github.com> Co-authored-by: williamwira <4399853+williamwira@users.noreply.github.com> Co-authored-by: Éloïse Brosseau <54746458+eloisebrosseau@users.noreply.github.com>
It seems like it was missing some commit on the last rebase. They should be there now. --------- Signed-off-by: Cédrik Fuoco <cedrik.fuoco@autodesk.com> Signed-off-by: Will Wira <4399853+williamwira@users.noreply.github.com> Signed-off-by: Bernard Laberge <bernard.laberge@autodesk.com> Signed-off-by: Éloïse Brosseau <eloise.brosseau@autodesk.com> Co-authored-by: williamwira <4399853+williamwira@users.noreply.github.com> Co-authored-by: Bernard Laberge <117092886+bernie-laberge@users.noreply.github.com> Co-authored-by: Éloïse Brosseau <54746458+eloisebrosseau@users.noreply.github.com>
…xcept Windows) (AcademySoftwareFoundation#618) ### Qt6 feature branch - Various fix for all platforms and VFX 2024 CI (Except Windows) ### Linked issues AcademySoftwareFoundation#366 ### Summarize your change. Here's are the changes: - Added VFX 2024 CI (except for Windows) without ci-openrv from aswf-docker for now (Linux only, not tested yet) - Code changes to fix issues on Windows - Code changes to fix issues on MacOS - Added OpenSSL 3 for VFX2024 - Added some logic to all `.wrapper` to detect which VFX platform is active (using Python version) and then use the right OpenSSL version. - FIx issue with Python not finding Tkinter on MacOS. The issue was with VFX2024, but the fix make sense for VFX2023 too. **There are a lot of reformatting because of the pre-commit hook. Windows is not included in the VFX 2024 CI because of issues with PySide6 in the CI.** ### Describe what you have tested and on which operating system. - [x] Rocky linux 8/9 - [x] MacOS - [ ] Windows (partial, can compile with my windows, but not in CI because of pyside6 issue) --------- Signed-off-by: Cédrik Fuoco <cedrik.fuoco@autodesk.com>
### Qt6 Feature branch - Rebase to main ### Linked issues n/a ### Summarize your change. I rebased by cherry-picking the commits since the last rebase. ### Describe the reason for the change. Rebase to main to minimize efforts --------- Signed-off-by: Bernard Laberge <bernard.laberge@autodesk.com> Signed-off-by: Cédrik Fuoco <cedrik.fuoco@autodesk.com> Signed-off-by: Éloïse Brosseau <eloise.brosseau@autodesk.com> Co-authored-by: Bernard Laberge <117092886+bernie-laberge@users.noreply.github.com> Co-authored-by: Éloïse Brosseau <54746458+eloisebrosseau@users.noreply.github.com>
…ndation#641) ### Qt6 Feature branch - Fix issue with menu on MacOS ### Linked issues n/a ### Summarize your change. The QMenuBar and QMenu has changed in Qt6. Some behaviours seems to have changed in Qt6, and the code around a memory leak (based on the comments) is not longer working. I made some changes that works on MacOS and Rocky Linux 8. There no information about how to reproduce that memory leak. **We'll need to test around that to make sure that it is not the case with Qt6 anymore.** ### Describe the reason for the change. Multiple menu were appearing on the menu bar on MacOS. ### Describe what you have tested and on which operating system. - [x] MacOS - [x] Rocky Linux 8 --------- Signed-off-by: Cédrik Fuoco <cedrik.fuoco@autodesk.com> Signed-off-by: Cédrik Fuoco <cedrik.fuoco@autodesk.com>
…Windows (AcademySoftwareFoundation#651) ### Qt6 Feature - Fix issues with Python ### Linked issues AcademySoftwareFoundation#367 ### Describe the reason for the change. I created two functions to do the install step of Python. One for Python 3.10 (VFX2023) and one for Python 3.11 (VFX2024). I think it will be easier to maintain and remove later on. The difference was only for Windows, but both functions contains the code for all platforms. Since Python 3.11, the `PLATLIBDIR` directory was renamed to `DLLs` instead of `lib`. You can changed it by modifying the Visual studio project file, but I found some place in the Python source code that was hardcoded to `DLLs`. I opted to keep `DLLs` and copy that over to the OpenRV's stage directory. It does not change anything functionally because the Python executable is looking for the DLLs library automatically. **The changes above also fixes the issue with PySide6 compilation on Windows.** ### Describe what you have tested and on which operating system. Windows ### If possible, provide screenshots. stage directory:  --------- Signed-off-by: Cédrik Fuoco <cedrik.fuoco@autodesk.com> Signed-off-by: Cédrik Fuoco <cedrik.fuoco@autodesk.com>
### Feature/Qt6 - Fixing the audio settings ### Linked issues n/a ### Describe the reason for the change. The code that was handling the audio sample rate changes was refactored in this PR because of the Qt Audio module refactor in Qt 6. In Qt 5, you could get a list of valid sample rate from the audio device, but in Qt 6, you only get a minimum and maximum value. With Qt6 OpenRV, the widget to change the sample rate is a SpinBox instead of a ComboBox. **_What is left to do for Audio:_** - [ ] There are still some memory issues (crash) when changing audio sample rate on MacOS (tested on ARM64). **This PR does not fix that.** - [ ] Other refactor to support audio format (e.g. int8) that are not available by default in Qt 6 anymore. ### Describe what you have tested and on which operating system. Windows, MacOS and Rocky. ### If possible, provide screenshots. With Qt 6, a QSpinBox is used for the sample rate setting.  --------- Signed-off-by: Cédrik Fuoco <cedrik.fuoco@autodesk.com> Signed-off-by: Cédrik Fuoco <cedrik.fuoco@autodesk.com>
…#664) ### Feature/Qt6 - Rebase to main (Jan 21 2025) ### Linked issues n/a ### Describe the reason for the change. Rebase to main --------- Signed-off-by: Bernard Laberge <bernard.laberge@autodesk.com> Signed-off-by: Cédrik Fuoco <cedrik.fuoco@autodesk.com> Signed-off-by: Éloïse Brosseau <eloise.brosseau@autodesk.com> Signed-off-by: mcoms <mastercoms@tuta.io> Signed-off-by: Mirco Tornow <mirco.tornow@outlook.com> Co-authored-by: Bernard Laberge <117092886+bernie-laberge@users.noreply.github.com> Co-authored-by: Éloïse Brosseau <54746458+eloisebrosseau@users.noreply.github.com> Co-authored-by: mcoms <mastercoms@tuta.io> Co-authored-by: Mirco Tornow <mirco.tornow@outlook.com> Co-authored-by: Bernard Laberge <bernard.laberge@autodesk.com>
### Feature/Qt6 - Format and rebase ### Linked issues n/a ### Summarize your change. Reformatted and rebased branch to main. ### Describe the reason for the change. Reformatted the code base manually in the branch to prevent a lot of conflicts when this branch will be merged in main. --------- Signed-off-by: Cédrik Fuoco <cedrik.fuoco@autodesk.com> Signed-off-by: Éloïse Brosseau <eloise.brosseau@autodesk.com> Co-authored-by: Éloïse Brosseau <54746458+eloisebrosseau@users.noreply.github.com>
…cademySoftwareFoundation#679) ### Feature/Qt6 - Fix issues with bindings and adding missing handrolled ### Linked issues n/a ### Summarize your change. Here's the list of changes: - Some formatting - Fixed some issue with Mu bindingds - Added missing methods in the Mu bindings by adding them in the handrolled section (they will be generated the next time) - Added new classes that were needed ### Describe the reason for the change. Prevent issues with Mu plugins and prevent future issues in the next Mu bindings generation. ### Describe what you have tested and on which operating system. all Signed-off-by: Cédrik Fuoco <cedrik.fuoco@autodesk.com>
…s for clang-format (AcademySoftwareFoundation#684) ### Feature/Qt6 - Add QTimezone to the MuQt5 bindings and ignore templates for clang-format ### Linked issues n/a ### Summarize your change. This PR adds QTimezone to the MuQt5 bindings and added .clang-format-ignore to ignore the MuQt templates. ### Describe the reason for the change. This is needed to fix some Mu plugins. Signed-off-by: Cédrik Fuoco <cedrik.fuoco@autodesk.com>
…AcademySoftwareFoundation#685) ### Feature/Qt6 - Fix issue with QItemSelectionModel::select method in Mu ### Linked issues n/a ### Summarize your change. Added a special case to deal with calls to `QItemSelectionModel::select`. ### Describe the reason for the change. The new implementation of `QMetaMethod:invoke` in Qt 6 is using variadic templates and C++ type deduction (instead of `QGenericArgument`). Therefore, it does not convert the `QItemSelectionModel::SelectionFlags` enum to `int` and an error message appears. ### Describe what you have tested and on which operating system. Rocky 8 and MacOS Signed-off-by: Cédrik Fuoco <cedrik.fuoco@autodesk.com>
### Feature/Qt6 - Fix issue when cloning RV ### Linked issues n/a ### Summarize your change. - Fixed issue with `Clone RV` and `Clone Synced RV` operation and - Fixed issue with a few Mu script in the Qt5 build. - Fixed a warning about an unknown signal named `error`. The `error` signal was renamed to `errorOccurred` in Qt 5.15 and Qt 6+. ### Describe the reason for the change. The Clone RV application was not launching due to the incorrect usage of `startDetached` method. ### Describe what you have tested and on which operating system. Rocky --------- Signed-off-by: Cédrik Fuoco <cedrik.fuoco@autodesk.com>
…#690) ### Feature/Qt6 - Rebase main into feature/qt6 ### Linked issues n/a ### Summarize your change. Rebase main into feature/qt6 --------- Signed-off-by: Bernard Laberge <bernard.laberge@autodesk.com> Signed-off-by: Cédrik Fuoco <cedrik.fuoco@autodesk.com> Signed-off-by: Christopher Lee <christophercarlos.lee@gmail.com> Co-authored-by: Bernard Laberge <117092886+bernie-laberge@users.noreply.github.com> Co-authored-by: Chris <34953610+chrisc-lee@users.noreply.github.com>
…cademySoftwareFoundation#691) ### Feature/Qt6 - Fix issue where Mu Command API Browser is not opening ### Linked issues n/a ### Summarize your change. There were some changes in the Qt 6 QVariant type and it seems like we need to explicitly define a QVariant constructor for QIcon in Qt 6. ### Describe the reason for the change. Mu Command API Browser is not opening in the build using Qt 6. ### Describe what you have tested and on which operating system. Rocky 8 Signed-off-by: Cédrik Fuoco <cedrik.fuoco@autodesk.com>
…sing Qt6Test.dll) (AcademySoftwareFoundation#692) ### Feature/Qt6 - Fix issue that prevent OpenRV to launch on Windows (Missing Qt6Test.dll) ### Linked issues n/a ### Summarize your change. Fix issue that prevent OpenRV to launch on Windows (Missing Qt6Test.dll) ### Describe the reason for the change. OpenRV is not launching on Windows. ### Describe what you have tested and on which operating system. Windows Signed-off-by: Cédrik Fuoco <cedrik.fuoco@autodesk.com>
…ndation#693) ### Feature/Qt6 - Reverting old changes in RvDocument ### Linked issues n/a ### Summarize your change. Reverting changes from a couple weeks ago ### Describe the reason for the change. A couple weeks ago, there was an issue with duplicated menu on MacOS (multiple time the same menu), but the fix broke something in the menu system for all platforms. Therefore, I am reverting the changes and I'll go back to the drawing board for the MacOS issue. --------- Signed-off-by: Cédrik Fuoco <cedrik.fuoco@autodesk.com>
…emySoftwareFoundation#695) ### Feature/qt6 - Fix issue with all menu being duplicated on MacOS ### Linked issues n/a ### Summarize your change. I removed old code that was dealing with a potential memory leaks related to QAction and specific behavior for MacOS that causes issues now. ### Describe the reason for the change. This changes was done to fix an issue where all the menus are duplicated multiple times in the QToolBar. ### Describe what you have tested and on which operating system. Rocky 8 and MacOS Signed-off-by: Cédrik Fuoco <cedrik.fuoco@autodesk.com>
…emySoftwareFoundation#703) ### Summarize your change. When exporting (or, for that matter, calling an external process) was finished, the "finish" callback faied and caused RV to crash. The fix was simply to add a missing parameter at the end of the signature of the callback method called by connect() when QProcess.finish is called, to match what is expected in Qt6 (and Qt5, actually) This crash appeared whenever launching an external process, not just rvio. So other issues we found regarding a crash when launching an external process should also be fixed with this change. ### Describe the reason for the change. Fixes crash at end of export process ### Describe what you have tested and on which operating system. Tested on Linux64 Rocky 9.5. It's not clear why this worked on the Qt5 branch and crashed on Qt6. Indeed, the "finish" signal always took 2 parameters, while the Mu signal only took one. see below : similar signature of QProcess.finish() https://doc.qt.io/qt-5/qprocess.html https://doc.qt.io/qt-6/qprocess.html **Please retest this with a Qt5 build of the qt6 branch.** ### Add a list of changes, and note any that might need special attention during the review. n/a ### If possible, provide screenshots. n/a Signed-off-by: Patrick Bergeron <patrick.bergeron@autodesk.com>
AcademySoftwareFoundation#702) ### Summarize your change. This fix is to prevent QCoreApplication::initLocale() method to complain that an invalid locale is currently set. Before this bugfix, RV's init code explicitly set the current locale to "C". At the same time, this locale is now explicitly unsupported by Qt6, and instead will instead switch to "C.UTF-8" under Linux, or "UTF-8" under Darwin (macOS) For more information about how Qt initializes the locale, look at initLocale() in the file: QT_HOME/Qt6.x.x/Src/qtbase/src/corelib/kernel/qcoreapplication.cpp ### Describe the reason for the change. See above. ### Describe what you have tested and on which operating system. Tested on Rocky 9.5, waiting on Cedrik to test on Darwin. ### Add a list of changes, and note any that might need special attention during the review. n/a ### If possible, provide screenshots. n/a --------- Signed-off-by: Patrick Bergeron <patrick.bergeron@autodesk.com>
) ### Feature/Qt6 - Rebase with the main branch ### Linked issues n/a ### Summarize your change. ### Describe the reason for the change. Rebasing the feature branch with the main branch ### Describe what you have tested and on which operating system. See commits list --------- Signed-off-by: Bernard Laberge <bernard.laberge@autodesk.com> Signed-off-by: Cédrik Fuoco <cedrik.fuoco@autodesk.com> Signed-off-by: Éloïse Brosseau <eloise.brosseau@autodesk.com> Co-authored-by: Bernard Laberge <117092886+bernie-laberge@users.noreply.github.com> Co-authored-by: Éloïse Brosseau <54746458+eloisebrosseau@users.noreply.github.com>
…s exporting quicktime video (#35176) (AcademySoftwareFoundation#704) ### Describe the reason for the change. When we clicked or dragged the widget during a quicktime export (or any instance of the ProcessInfo widget for that matter), we (almost) inevitably always crashed. Happened in Qt5 and Qt6 branch, and as far back as RV 2023, apparently. ### Summarize your change. When we clicked on the ProcessInfo widget, the code tries to access the _buttons[] array to see if we are clicking in the (X) button, but simply reading this array crashed RV. It turns out that for the ProcessInfo widget, the render() method recreated the _buttons[] array, which contained the coordinates and process info for the (X) cancel button. This was in a different thread than the event-handling thread. So, while the render method was executing and the _buttons[] array was being recreated, the event-handling thread was trying to read the same array to see if w were clicking in the (X) button. Since the memory had been disposed and replaced in the meantime, this caused a crash . The fix was simply to initialize the array when the ProcessInfo widget is created, and in the render() method we update the x/y/w/h/pid. This does not cause a change in memory space, and the crash no longer happens. Ideally, we'd need a mutex to sync the read/write tof the x/y/w/h/pid values, but, we're really just reading/writing coordinates here, and the pid always remains constant, so even if there was concurrent read/write access it should be completely harmless. Either way, the crash no longer occurs. ### Describe what you have tested and on which operating system . This was 100% reproducible on Linux, and I think also macOS, the problem was no platform-specific but Mu-specifc. ### Add a list of changes, and note any that might need special attention during the review. There is only 1 file to review, so, no big deal. ### If possible, provide screenshots. n/a Signed-off-by: Patrick Bergeron <patrick.bergeron@autodesk.com>
…g the main window (AcademySoftwareFoundation#709) ### Summarize your change. Added a singleshot timer to force a resize event to the main window, in order to fix an image scale issue when first starting RV. ### Describe the reason for the change. On macOS (and, presumably, any platform that supports HDPI) the initial display of the main view's image will not be scaled properly. After much investigation, this appears to be a Qt bug but were being unable to truly identify the root cause (coordinates are ok, pixel scale is ok, pixel ratio is ok, gl surface is ok, gl viewport is ok, etc etc. Oddly enough, the issue disappeared when resizing the view, even 1 pixel, but this has to be done after the view is first shown. As a result, we dimply decided to force two resize events to occur (+1 pixel, and then -1 pixel). This has no visible effect on startup (or, you'd really need eagle eyes) to go: "A-HA!" ### Describe what you have tested and on which operating system. This was reported on macOS, so I tested it on macOS, but it's possible it may occur on other platofrms as well. Unfortunately I did not have HDPI displays where the OS does pixel-doubling by itself like on macOS. ### Add a list of changes, and note any that might need special attention during the review. This is a simple change in RvDocument, where we initially set up the main view. ### If possible, provide screenshots. No need, simply confirm that RV now starts correctly on macOS Signed-off-by: Patrick Bergeron <patrick.bergeron@autodesk.com>
… (eg: NDI) (AcademySoftwareFoundation#707) Fixes #38133 ### Describe the reason for the change. Could not open presentation device after setting Presenation device to NDIDevice, and then doing CTRL-P to enable it. The error message was due to a ";" at the end of an if-statement, but then it caused opengl errors down the line due to the gl context being invalid after calling swapBuffers. Turns out all the swapBuffer management is no longer needed with QOpenGLWidget, explanation in the summary below. ### Summarize your change. Note for Qt6 . QOpenGLWidget implementation: With the new QOpenGLWidget (Qt6 branch), all of the rendering of each widget is done in its own FBOs (aka: off-screen buffers), as opposed to the old Qt5/QGLWidget branch where the rendering was done in each GGLWidget's backbuffer (aka: GL_BACK, an on-screen framebuffer surface). With QOpenGLWidget, it is now the WainWindow's responsibility (more technically, the MainWindow's rendering backend, which happens to be Qt's OpenGL rendering backend when QOpenGLWidget are present in the children tree) to gather and composite all of the off-screen buffers (regardless of which image type they are, eg: cpu-memory images, gpu/opengl images, etc) and finally to call swapBuffers to show the final contents of the mainWindow. As a result, I'm not sure there's a point -- at all -- in calling swapBuffers on the GLView anywhere amnymore. From old comments in the previous version of this tile, it appears that calling swapBuffers was done to force a quicker visual update, or to minimize visual tearing of some sort. This would have worked with the old QGLWidget (becayuse each QGLWidget had its own context, and its rendering target was directly the GL_BACK framebuffer, but, again, with QOpenGLWidget, the rendering target is no longer GL_BACK, it is an FBO that is meant to be used at the end of the application's drawing / visual update pipeline. I am not sure how this would affect (if at all?) the display of the rendered buffer on external video devices. This needs to be tested extensively. ### Describe what you have tested and on which operating system. I have tested on Linux64 and have seen no adverse effects. **Note for QA**: Please test this extensively on all 3 platforms, and with all VideoDevices (NDI, SDI, AJA, BlackMagic), to make sure this causes no problem. ### Add a list of changes, and note any that might need special attention during the review. ### If possible, provide screenshots. n/a --------- Signed-off-by: Patrick Bergeron <patrick.bergeron@autodesk.com> Co-authored-by: Cédrik Fuoco <105517825+cedrik-fuoco-adsk@users.noreply.github.com>
…reFoundation#700) ### Summarize your change. The dropperSampler tool in RV was not working correctly anymore in the qt6 branch. It had a strange offset (horizontally and vertically) because the OpenGL function glReadPixels was reading from the wrong context and the wrong framebuffer object. in annotate_mode.mu, the method dropperSampler() called an OpenGL method (glReadPixels) while not in a rendering pass. This somehow worked in the old Qt5/GLWidget implementation, because the current OpenGL context and FBO always remained set to the main view widget. But since switching to Qt6/QOpenGLWidget, this no longer holds true - the Qt6/QOpenGLWidget backend has a OpenGL compositor that works differently, and the OpenGL context/fbo of the main view was no longer current after the application was done refreshing. Since we want to query the final rendered view using glReadPixels, we implemented a mu method called framebufferPixelValue(x, y) which internally eventually sets the OpenGL context/fbo to the correct one before calling glReadPixels. (after PR 707, there is no need to restore the previous one) ### Describe the reason for the change. See first paragraph of the "Summarize your change" section ### Describe what you have tested and on which operating system. I have tested the dropperSampler tool, with different layouts and docked windows, but on Linux64 only. Please test in Darwin and Windows. ### Add a list of changes, and note any that might need special attention during the review. Please test and review on all other platforms please; although I am confident this will work just fine. During investigating this bug, the easiest way to see this problem was to dock the annotation bar on the left portion of the application window, thus pushing the main image window a bit to the right. Then, you'd easily notice that the color was wrong when you picked a color, which was horizontally offset by the width of the docked panel. The easiest way to test and repro this, and to notice the fix works, is to start TV with the "smptebars.movieproc" argument, where you can clearly see the differences in color. ### If possible, provide screenshots. n/a --------- Signed-off-by: Patrick Bergeron <patrick.bergeron@autodesk.com> Co-authored-by: Bernard Laberge <117092886+bernie-laberge@users.noreply.github.com> Co-authored-by: Cédrik Fuoco <105517825+cedrik-fuoco-adsk@users.noreply.github.com>
…ndation#710) ### Feature/qt6 - Remove usage of Core5Compat library ### Linked issues n/a ### Summarize your change. I have eliminated all instances of QTextCodec, as this class is no longer supported in Qt 6. Previously, the code relied on the Core5Compat library, which incorporates deprecated and obsolete Qt 5 classes. ### Describe what you have tested and on which operating system. I created a python plugin to execute the httpGet method from the RvWebManager class. Signed-off-by: Cédrik Fuoco <cedrik.fuoco@autodesk.com>
…ndation#719) ### Feature/qt6 - Fix issue with Windows CI (release) ### Linked issues n/a ### Summarize your change. I added the architecture and vfx platform to the key for the dependencies cache on Windows. It is already present for MacOS. ### Describe the reason for the change. Sometimes, the release Windows build was failing because it was using the cache from a VFX2023 build. ### Describe what you have tested and on which operating system. CI Signed-off-by: Cédrik Fuoco <cedrik.fuoco@autodesk.com>
) ### Feature/Qt6 - Fix nuke and rv integration ### Linked issues n/a ### Summarize your change. This PR update the rvNetwork,py to Python 3 (copied from src/plugins/python/network/network/rvNetwork.py) and fix some issues in the Mu binding for the QTextStream class. I will also update the rvNetwork.py file in the main branch in another PR. ### Describe the reason for the change. Nuke and RV integration was broken. An image or video could not be played in RV. ### Describe what you have tested and on which operating system. Rocky Linux 8 Signed-off-by: Cédrik Fuoco <cedrik.fuoco@autodesk.com>
…Foundation#718) ### Feature/qt6 - Support both Qt 5 and Qt 6 in rvcmd.sh ### Linked issues n/a ### Summarize your change. I added checks to detect existing installations of Qt 5 or Qt 6. The search prioritise Qt 6 over Qt 5. Once Qt is found, the script set the right variables for the aliases. If QT_HOME is already set, I added a check to detect if it is Qt 5 or Qt 6. Note that the logic only look for `Qt*` and either `5.15*` or `6.5*` in the path. The script is not looking at the actual libraries files. ### Describe the reason for the change. rvcmd.sh had to be reworked to support both Qt 5 and Qt 6. ### Describe what you have tested and on which operating system. Rocky Linux 8 Signed-off-by: Cédrik Fuoco <cedrik.fuoco@autodesk.com>
…ion#714) This fixes SG-36654 SG-36655 SG-36656 ### Summarize your change. Added a method to filter out some Qt messages that are known/harmless warnings. One such warning had Qt reporting a warning about a scroll event occuring from a HID device whenever it was connected/disconnected/kvm'd. This appears to be fixed with Qt 6.6.3 but we are at 6.5.3, so we shut it up. Another such warning was about a Qt handler for a type conversion already registered (but when initializing PyQt it was tried to be initialized again), etc. ### Describe the reason for the change. Pollutes the console window needlessly with harmless warnings which are Qt bugs that were later fixed. ### Describe what you have tested and on which operating system. macOS and Linux. ### Add a list of changes, and note any that might need special attention during the review. ### If possible, provide screenshots. --------- Signed-off-by: Patrick Bergeron <patrick.bergeron@autodesk.com> Co-authored-by: Cédrik Fuoco <105517825+cedrik-fuoco-adsk@users.noreply.github.com>
…BO (AcademySoftwareFoundation#721) ### Summarize your change. This is an initial code fix while debugging and fixing SG-25296 (can't use 2nd monitor desktop window as presentation device on macOS M computers). This change does NOT Fix the issue (we still get a black screen) but during the port to Qt6, we introduced a change that deleted the current FBO in the current OpenGL context (a GLFBO class destroyed a FBO object that was owned by Qt, not by GLFBO) and made the GLView class unstable. This fix only addresses the incorrect OpenGL FBO destruction that occurs after exiting presentation mode. More concretely: GLFBO has 2 constructors: one where a FBO object is created using glGenFramebuffersEXT (in which case the GLFBO instance owns the FBO) and another one where the FBO handle is passed as an argument to the constructor (in which case the GLFBO instance does not own the FBO). Upon calling the destructor, any non-zero handle was destroyed, but since the non-owned FBOs actually belonged to Qt, we ended up corrupting the current OpenGL context because it no longer had a rendering surface to render onto. To fix this, I have just added a bool member to GLFBO to guide what happens in the GLFBO destructor. Also, while investigating this issue, I noticed that several places missed a call to the de debug macro TWK_GLDEBUG, so I have added them. Finally, I updated the formatting of the error log for gl errors, which was overly verbose. ### Describe the reason for the change. After exiting 2nd-monitor-desktop presentation mode, the OpenGL context got scrapped because an FBO object was mistakenly destroyed, which caused RV to stop displaying content until the view got resize/recreated with new FBOs. ### Describe what you have tested and on which operating system. macOS. Should work just the same on other platforms. ### Add a list of changes, and note any that might need special attention during the review. n/a ### If possible, provide screenshots. n/a --------- Signed-off-by: Patrick Bergeron <patrick.bergeron@autodesk.com>
…Foundation#724) ### Feature/qt6 - Clean up code in QTAudioRenderer class ### Linked issues n/a ### Summarize your change. Removed #ifdef related to RV_VFX_CY2023 since they are not needed in QTAudioRenderer class (QT5AudioRenderer is used for CY2023). ### Describe the reason for the change. Clean up code to make it more readable. ### Describe what you have tested and on which operating system. Rocky Linux 8 --------- Signed-off-by: Cédrik Fuoco <cedrik.fuoco@autodesk.com>
### Feature/Qt6 - Rebase to main ### Linked issues n/a ### Summarize your change. I cherry-picked the missing commits from main. 1. [Fix the failing readthedocs build](AcademySoftwareFoundation@5c54a9c) 2. [Fix network issue with rvnuke plugin](AcademySoftwareFoundation@eeae944) (No changes because the fix was already in the feature/qt6 branch) 3. [712: Adapt build to latest MSYS2 release](AcademySoftwareFoundation@ead6e53) 4. [Fix MMR support in OTIO reader](AcademySoftwareFoundation@1da73a9) ### Describe the reason for the change. Get the latest changes from the main branch. --------- Signed-off-by: Cédrik Fuoco <cedrik.fuoco@autodesk.com> Signed-off-by: Bernard Laberge <bernard.laberge@autodesk.com> Signed-off-by: Cédrik Fuoco <105517825+cedrik-fuoco-adsk@users.noreply.github.com> Co-authored-by: Bernard Laberge <117092886+bernie-laberge@users.noreply.github.com>
### Feature/qt6 - rebase to main ### Linked issues n/a ### Summarize your change. The rebase only include the PNG changes, but I want to get it right away in the feature/qt6 branch since it is affecting performance. ### Describe the reason for the change. Rebase ### Describe what you have tested and on which operating system. ### Add a list of changes, and note any that might need special attention during the review. ### If possible, provide screenshots. Signed-off-by: Cédrik Fuoco <cedrik.fuoco@autodesk.com>
### Feature/qt6 - Rebase ### Linked issues n/a ### Summarize your change. Rebase the branch to main ### Describe the reason for the change. Rebase the branch to main and this is needed to fix a build issue with PR AcademySoftwareFoundation#741 --------- Signed-off-by: Ken McGaugh <ken@mcgaugh.co.uk> Signed-off-by: Éloïse Brosseau <eloise.brosseau@autodesk.com> Signed-off-by: Cédrik Fuoco <cedrik.fuoco@autodesk.com> Signed-off-by: Bernard Laberge <bernard.laberge@autodesk.com> Signed-off-by: kenmcgaugh <ken@mcgaugh.co.uk> Co-authored-by: kenmcgaugh <ken@mcgaugh.co.uk> Co-authored-by: Éloïse Brosseau <54746458+eloisebrosseau@users.noreply.github.com> Co-authored-by: Bernard Laberge <117092886+bernie-laberge@users.noreply.github.com>
…twareFoundation#741) ### Feature/qt6 - Fix issue with multiple_source_media_rep.py ### Linked issues n/a ### Summarize your change. Call QAction() from QtGui. ### Describe the reason for the change. QAction() was moved to QtGui ### Describe what you have tested and on which operating system. Rocky linux 8 Signed-off-by: Cédrik Fuoco <cedrik.fuoco@autodesk.com>
…reFoundation#745) ### Feature/qt6 - Fix issue with icons in file open dialog ### Linked issues n/a ### Summarize your change. For Qt 6. I added more heuristics to find the right icon using the MIME type. ### Describe the reason for the change. All files were using the text-generic icon in the file open dialog. ### Describe what you have tested and on which operating system. ### Add a list of changes, and note any that might need special attention during the review. ### If possible, provide screenshots. **Before:**  **After:**  --------- Signed-off-by: Cédrik Fuoco <cedrik.fuoco@autodesk.com>
### Feature/Qt6 - Rebase to main ### Linked issues n/a ### Summarize your change. Rebase to main b3aa898 81b918f b13dc6d ### Describe the reason for the change. Rebase to main --------- Signed-off-by: Éloïse Brosseau <eloise.brosseau@autodesk.com> Signed-off-by: Cédrik Fuoco <cedrik.fuoco@autodesk.com> Signed-off-by: Ken McGaugh <ken@mcgaugh.co.uk> Co-authored-by: Éloïse Brosseau <54746458+eloisebrosseau@users.noreply.github.com> Co-authored-by: kenmcgaugh <ken@mcgaugh.co.uk>
…on#752) ### Feature/Qt6 - Small tweak of the audio logic ### Linked issues n/a ### Summarize your change. I removed the conditions to check the state of the QAudioSink object because it does not make sense in Qt 6. In Qt 6, when we call the `start `method on the QAudioSink object, The code starts transferring audio data from the device to the system's audio output right away and changes the device state afterwards (while in Qt 5, the state was changed first, and then the audio data was read). Therefore, it does not make sense to check the state in the `readData` callback because even if the state is STOPPED, we want to start reading the audio data. The audio cache/processing seems to behave just like the main branch with those changes. I tested it on Windows and MacOS. ### Describe the reason for the change. Fix audio on Windows and the new logic makes more sense based on how Qt 6 works now. ### Describe what you have tested and on which operating system. Windows and MacOS Signed-off-by: Cédrik Fuoco <cedrik.fuoco@autodesk.com>
…demySoftwareFoundation#754) ### Finalize port of QTabletEvent to Qt6 and fix tablet eraser issue ### Linked issues NA ### Describe the reason for the change. @cedrik-fuoco-adsk had already did the bulk of the work to port the QTTranslator from Qt5 to Qt6 but the QTabletEvent needed some attention because of the Qt6 refactoring with respect to its events handling. ### Summarize your change. **TabletDevice and TabletKind were removed** Rationale: In RV, when handling a table event, we used to record the TabletDevice (ie stylus) and the TabletKind (ie pen or eraser) in the RV's TableEvent class. In Qt6, QTabletEvent::FourDMouse and QTabletEvent::RotationStylus have been removed and replaced with QInputDevice::DeviceType::Puck and QInputDevice::DeviceType::Stylus respectively. So technically we would have had to create Qt version specific enums of TabletDevice and TabletKind. However, they were NOT used anywhere. Note that the important part is the name of the event like for example "stylus-pen--drag". This is what packages such as annotation_mode.mu use to act on tablet events. By getting rid of TabletDevice and TabletKind altogether in the RV codebase, we were able to minimize the differences between Qt5 and Qt6 with respect to Tablet Events. **Added all the Qt6 supported in options for tablet device type and tablet pointer type** Also in this commit: **Fixed a legacy issue when using a tablet's stylus eraser and going pack to the pen:** When using the eraser, the code was correctly switching to the eraser mode but when the user was selecting the pen again, it was not automatically switching to the pen mode, it was still erasing. ### Describe what you have tested and on which operating system. Successfully tested on macOS with a Wacom Intuos tablet ### Add a list of changes, and note any that might need special attention during the review. ### If possible, provide screenshots. Signed-off-by: Bernard Laberge <bernard.laberge@autodesk.com>
### Rebase to main ### Linked issues n/a ### Summarize your change. Rebase to main: 410dae8 --------- Signed-off-by: Éloïse Brosseau <eloise.brosseau@autodesk.com> Signed-off-by: Cédrik Fuoco <cedrik.fuoco@autodesk.com> Co-authored-by: Éloïse Brosseau <54746458+eloisebrosseau@users.noreply.github.com>
### Summarize your change. Note: This is a test PR for internal Autodesk review that needs to undergo testing. This PR aims to dramatically improve the loading time of playlists, especially online playlists, which was quite painful for customers, sometimes having to wait several minutes. For example, on my workstation, loading a playlist of 250 clips went from 3-4 minutes to about 10 seconds, which is pretty nice improvement. Some years ago, RV introduced "progressive source loading" which aimed to achieve a similar result, but caused lots of unintended support issues, like scrips not working well and some random crashes sometimes, and other side effects. This PR is *not* an attempt to improve progressive loading, which had fundamental flaws, instead we use a different technique altogether. Several changes were made in order to achieve this, but first, it's worthwhile to describe the cause of the issue to understand the fix. The reason for the slowness is because after we receive the playlist (the list or URLs) we must create the RV graph, for which we create one FileSourceIPNode per movie, sequentially, in the main thread. For each FileSourceIPNode created sequentially, we must open the movie header (to get the number of frames, framerate, duration, etc) by calling FileSourceIPNode::openMovie(), eventually calling upon libav (ffmpeg), which initiates an http request (with cookies and headers), sends the request, downloads the response and the data, and then can open the movie's header. Unfortunately, doing this over the network, one file at a time, causes RV to block on I/O while the shotgrid server is serving only a single request. To give a rough idea, for a single clip, that network lag is about 0.5-to-1.0 seconds as I measured it, so multiply that by 250 and you've got "several minutes". Obviously, the shotgrid server can handle multiple concurrent requests at once, so we simply decided to create a preloader which loads all 250+ file headers in the background, using up to 32 concurrent threads to issue those http requests. In other words, the idea is to parallelize the I/O waiting time, hopefully by a factor of "up to" 32. So while the preloader is doing all of this work (well, waiting, mostly) in the background, the RV graph gets created in the main thread, just as before. But, by the time FileSourceIPNode::openMovie gets called -- lo-and-behold! -- the movie's header has already been opened (or so we're hoping) so the call completes almost immediately without having to pay that 0.5-to-1.0 second lag time. Unfortunately, there's some post-fixup stuff going on after we load a clip, which causes RV to still need to spend about 0.05 second per node, but still, it's a much better outcome than before. Another performance bug was fixed in this PR: we're no longer paying a O(n^2) (maybe O(n^3)?) loop on trying to determine if the UI name of a clip already exists, by searching for an existing name in the RV graph (which is NOT free), and instead we have an O(1) algorithm to figure out the next UI name for a node in case there's a duplicate. Other little fixes were done here and there, but the crux of the problem is solved using the above approach, with other details provided herein below. ### Describe the reason for the change. Loading long playlists was unusably slow; it was time to improve this for customers especially for the purpose of collaborative review. ### Describe what you have tested and on which operating system. I tested this on my Mac. When logged in shotgrid, I select a playlist of 250 clips, right-click, and then "play in RV". Also, using the Live Review workflow on shotgrid, we join a session in RV, and the clips load way faster than before, although this has made us uncover other issues we need to debug/fix separately. ### Add a list of changes, and note any that might need special attention during the review. **NOTE: For QA testing, please check that this continues to work as before even for static files or image sequences, even though we may not benefit much from preloading speed in some cases.** A few things to mention here. **(1) New methods to MovieReader's public API** The class MovieReader's base public interface was modified to add two new methods, and all MovieReader-derived classes must implement two new functions: - preloadOpen() -- which gets executed in the background to open the movie file. - postPreloadOpen() -- which gets executed in the main thread to perform work done in the FileSourceIPNode's current loading context at node creation. Implementing the above two functions is actually quite easy -- you basically just need to break in two what was previously done in the open() methods. All of this is documented at length in MovieReader.h so if you're interested you may read the comments there. When we did this for MovieFFMpegReader, it turns out that there's a thread-unsafe method that we would have liked to put in preloadOpen(). Indeed, findStreamInfo() calls libav's avformat_find_stream_info() which _should_ be thread safe but isn't, and causes crashes when called concurrently. I haven't debugged this (yet?) but it's a long-time bug that was first reported like 15 years ago, but apparently remains, and I provided a link to one of the places I found online that mentions the issue (as a side note, it might have been a cause for random crashes with progressive loading years ago). In any case, the crashes went away when I moved findStreamInfo() to postPreloadOpen() (which is serialized and executed in the main thread) just like before. **(2) You may not like the way the background thread scheduler works.** It's perhaps possible that the background thread management and reader scheduler is a bit overly complex, although it's still a relatively simple thing overall. Nevertheless, it might be a worthwhile exercise to use a thread pool with each thread in the pool having a queue, instead of a thread scheduler that dispatches one new thread whenever we have pending files to read. While a thread pool has the benefit of reusing threads, all threads remain in existence (but paused) after everything is done loading, consuming some memory and resources. On the other hand, my current approach launches/ends one thread per movie, and the thread ends when the movie's done reading, thus releasing the thread memory after the movie's header is read. We also maybe could have used a thread pool library, like the one provided by boost, but, at this time this custom code works well enough as it is, and it's tweakable to try new things, so why not. **(3) Finding the next UI Name is done is a super simple helper class called UINameCache** It is found in Session.h, and simply maintains a counter to a root name, and every time "nextUIName" is called, the counter increments. While this is a super simple solution to an O(N^2-or-3) problem, it does have slightly different behavior than before, which is also documented in the header file. These differences in behavior were discussed at length with Bernie, and we're ok with the changes. **(4) There's a function called lookupFilenameInMediaLibrary which has moved.** This is a function that not just gets the filename in the media library, but also fetches the cookies and headers used to create the http request (otherwise, fetching the file via http will fail due to not having session cookies). The method used to be in FileSourceIPNode, but it's been moved to a namespace that is accessible by the preloader. It used no special member of FIleSourceIPNode, and mostly static functions at that, so it was easy to move elsewhere; it was mostly copy/pasted from one spot to another. **(5) You may be wondering why some "if()" statements are now like "if (( ))"** This is because of an assignment within the if statement. By adding a () around the assignment, the compiler doesn't complain about the if statement and doesn't issue a warning. I got tired of looking at the same verbose warnings constantly. ### If possible, provide screenshots. No screenshot to provide, but, you may imagine in your head a picture of a coffee mug with more coffee in it than before, by the time the playlist is done loading. --------- Signed-off-by: Patrick Bergeron <patrick.bergeron@autodesk.com>
) ### Feature/qt6 - Merge main into feature/qt6 ### Linked issues n/a ### Summarize your change. Executed a `git merge main` and I fixed some issues in the MuQt6 templates discovered during the merge due to the clang formatting. --------- Signed-off-by: Éloïse Brosseau <eloise.brosseau@autodesk.com> Signed-off-by: Cédrik Fuoco <cedrik.fuoco@autodesk.com> Signed-off-by: Ben Chamberland <becha9260@gmail.com> Signed-off-by: paulbarton90 <28630076+paulbarton90@users.noreply.github.com> Signed-off-by: Bernard Laberge <bernard.laberge@autodesk.com> Signed-off-by: Will Wira <4399853+williamwira@users.noreply.github.com> Signed-off-by: Cédrik Fuoco <105517825+cedrik-fuoco-adsk@users.noreply.github.com> Signed-off-by: Cédrik Fuoco <cedrik.fuoco@autodesk.com> Signed-off-by: mcoms <mastercoms@tuta.io> Signed-off-by: Mirco Tornow <mirco.tornow@outlook.com> Signed-off-by: Christopher Lee <christophercarlos.lee@gmail.com> Signed-off-by: Patrick Bergeron <patrick.bergeron@autodesk.com> Signed-off-by: Ken McGaugh <ken@mcgaugh.co.uk> Signed-off-by: kenmcgaugh <ken@mcgaugh.co.uk> Co-authored-by: Éloïse Brosseau <54746458+eloisebrosseau@users.noreply.github.com> Co-authored-by: Ben Chamberland <85132405+chxmberland@users.noreply.github.com> Co-authored-by: Paul Barton <28630076+paulbarton90@users.noreply.github.com> Co-authored-by: Éloïse Brosseau <eloise.brosseau@autodesk.com> Co-authored-by: Bernard Laberge <117092886+bernie-laberge@users.noreply.github.com> Co-authored-by: williamwira <4399853+williamwira@users.noreply.github.com> Co-authored-by: mcoms <mastercoms@tuta.io> Co-authored-by: Mirco Tornow <mirco.tornow@outlook.com> Co-authored-by: Bernard Laberge <bernard.laberge@autodesk.com> Co-authored-by: Chris <34953610+chrisc-lee@users.noreply.github.com> Co-authored-by: pbergeron-adsk <patrick.bergeron@autodesk.com> Co-authored-by: kenmcgaugh <ken@mcgaugh.co.uk>
…AcademySoftwareFoundation#736) ### Describe the reason for the change. With the port to Qt6/QOpenGLWidget, the DesktopVideoDevice implementation was broken for all platforms (macOS, Linux, Windows). This is a long-time desired fix, since on Apple-Silicon macOS Arm64, DesktopPresentationDevice was broken for over 3 years. This PR aims to restore desktop presentation device functionality, by using the common Qt path to open a full-screen desktop window on the target screen. This was the path taken for Windows and Linux, but for macOS, there was a special CGDesktopVideoDevice, which used Apple's CoreGraphics API to achieve this. ### Summarize your change. This PR contains a surprisingly large amount of code changes to restore this functionality using QOpenGLWidget. The main difference is that where QGLWidget used the default BackBuffer to render to screen (GL FBO id 0), this is no longer allowed using QOpenGLWidget, which exclusively uses off-screen FBOs to render the view. Previous RV code made many, many assumptions about when it was safe to use FBO 0, and this was the root cause of the problem. The other major cause was that during RV's rendering of the session, the temporary FBOs created for the presenation device were created in the wrong OpenGL context. Somehow, this used to "work" in the old implementation (presumably because we were rendering to FBO 0 so it didn't matter) but when we started using actual off-screen FBOs, the FBO for the external desktop view never had any content rendered to it. Thus, the approach taken for this fix are as follows: 0) In RV main for macOS, Added setAttribute to allow accessing GL from different threads. This was enabled for Windows and Linux for the QT path, so we also need it. 1) Since we decided to use the Qt path for the core technology to render to another full-screen display, I merged the relevant (specialized) code from class QTDesktopPresentationDevice back into class DesktopPresentationDevice. 2) The QTDesktopVideoDevice (or was it DesktopVideoDevice?)'s bind() and bind2() methods were removed, which was causing the FBOs to be created in the wrong opengl context. This legacy code caused a very mysterious issue where the FBO to blit to the external screen was always empty. This is because the FBO for the screen was created in the wrong context. Fwiw, no other external video device overrode bind() and bind2() either... not sure why this was ever done in the first place. 3) In DesktopVideoModule, we checked the 3 platforms and added platform-specific code to enumerate and instantiate the proper DesktopVideoDevices. This enumeration/instantiation code was moved the various implementations of the VideoDevice classes (DesktopVideoDevice, CGDesktopVideoDevice - no longer used for now, CGDesktopVideoDeviceArm - no longer used for now, but soon) 4) On macOS, showng the rendered framebuffer on the screen was done using some kind of janky multi-threaded swapbuffer kludge. That kludge was completely removed, and besides, we don't use swapBuffer anymore -- it's not allowed with QOpenGLWidget - we simply force the QOpenGLWidget to repaint itself in the main thread after the transfer function is called. (See SyncBufferThreadData stuff that was removed in GLView) 5) An important, but subtle new implementation note, has to do with changes to GLFBO, which has 2 constructors: The first one, where an FBO is explicitly created at resolution width/height, and with its own pixel buffer (aka render buffer), and the Second, where an GLFBO is created to attach to a video device, which was originally thought to output to a "default" framebuffer (aka: FBO 0, the GL backbuffer directly, which is not allowed anymore). In an early version of the FeatureQt6, the id of the FBO was take from the OpenGL Current context, but, this was incorrect. The real solution is that the id of the GLFBO, when created with constructor 2, is in fact the ID of the FBO owned by the QOpenGLWidget attached to the VideoDevice. It is NOT the id of the current context (which happened to work when there was no DesktopVideoDevice). Therefore, I updated "GLFBO::fboId()", to not return m_id, but to return the id of the QOpenGLWidget's fbo id. It's important to always query the view, because the QOpenGLWidget's fbo id might actually be recycled/changed if the view is changed/resized/etc. Querying the ID of the widget's fbo follows the same patter as asking for the width/height of the fbo when using the constructor 2. 6) At the moment, the DesktopVideoModule initialization method is still a bit janky. You'll see this with some code that checks which platform we're running on, even though we now we always instantiate the DesktopVideoDevice class (the same Qt class). I've left this platform-checking code because It will be required anyway to fix 25296 properly, and I'll clean that a bit better for that PR, see the note below. 7) As part of this work, I had to implement debugging tools to see what was going on with the state of the current openGL context, and the state of the currently bound FBO. In GL.cpp, I've implemented an OpenGL Context Tracker, which is updated every time the TWK_GLDEBUG; macro is invoked. This tracker will query the opengl state and currently bound context, and will report any changes to the currently bound opengl state. I could have destroyed this code, but I thought it would be helpful in the future, so I've left it in GL.cpp but deactivated it through a #define. PLEASE NOTE: This PR is mainly a fix for 38492-Broken-on-all-platforms and thus successfully restores functionality on macOS (25296). However we do lose a bit of functionality on macOS versus the old RV -- in particular the ability to select the target screen's resolution and change the screen's resolution when activating presentation mode (because we use the Qt path instead of the CoreGraphics path). A lot of code in this PR is necessary to fully implement 25296, and since I was working on both at the same time, some of the code serving as groundwork to fully fix 25296 is included in this PR, but I've decided to break it into 2 PRs. (That groundwork code is found in CGDesktopVideoDevice.cpp and CGDesktopVideoDeviceArm.cpp) ### Describe what you have tested and on which operating system. **I have tested this on macOS, but testing on Windows and Linux is needed.** I also suspect that the Stereo or Quad mode remains broken. But since I have no way to test this. I will create another Jira ticket to address this. ### Add a list of changes, and note any that might need special attention during the review. As per above, we need to test this functionality EXTENSIVELY, on all platforms. For MacOS, support for DisplayLink was removed, however DisplayLink may already be transparently taken care of using the Qt implementation of the DesktopVideoDevice. (I have no way to test this). Even if this does not work, Bernard approved its sacrifice. ### If possible, provide screenshots. --------- Signed-off-by: Patrick Bergeron <patrick.bergeron@autodesk.com>
bernie-laberge
pushed a commit
that referenced
this pull request
May 5, 2025
This is the first phase of the Qt 6 port. It was mostly tested on Rocky Linux 8. There are a lot of generated files under `src/lib/mu/MuQt6`, and some modifications but you don't need to go over that folder. There are known issues that will be addressed in future PR: **All platforms:** - Refactor _QtAudioRenderer_ class. Currently, The _pulse_ audio output device is not enabled, and they are loud white noise issues on MacOS. - Remove the usage of _Core5compat_ and refactor what is deprecated. I don't think there is a lot to do here, but it is mostly about refactor the usage of _QTextCodec_. **MacOS:** - The _Open File dialog_ does not open anymore. - Loud white noise audio (Should be related to _QtAudioRenderer_) - The Central widget is rendered in the bottom left corner, but as soon as you resize the windows, it goes back to normal. **Windows:** - Issue with the compilation of Shiboken6 - [ ] Rocky LInux 8 - [ ] MacOS attention during the review. --------- Signed-off-by: Cédrik Fuoco <cedrik.fuoco@autodesk.com>
bernie-laberge
pushed a commit
that referenced
this pull request
Feb 23, 2026
…ftwareFoundation#1122) ### Fix crash calling sourcesAtFrame at session clear. ### Summarize your change. If sourcesAtFrame is called in an event from which the graph is cleared (`graph-node-inputs-changed` in my case, but it could be anything that clears the graph), RV will crash with a stack resembling the below: ``` * thread #1, name = 'RV Main', queue = 'com.apple.main-thread', stop reason = EXC_BAD_ACCESS (code=1, address=0x793bcc0f30f4744a) * frame #0: 0x00000001003e8688 RV`IPCore::AdaptorIPNode::metaEvaluate(this=0x0000000a96080780, c=0x000000016fdf6290, visitor=0x000000016fdf6c40) at AdaptorIPNode.cpp:92:31 frame #1: 0x000000010027b920 RV`IPCore::IPNode::metaEvaluate(this=0x0000000a96080a00, context=0x000000016fdf6290, visitor=0x000000016fdf6c40) at IPNode.cpp:408:23 frame #2: 0x000000010027b920 RV`IPCore::IPNode::metaEvaluate(this=0x0000000a96074000, context=0x000000016fdf6290, visitor=0x000000016fdf6c40) at IPNode.cpp:408:23 frame #3: 0x00000001007d5eb0 RV`IPCore::RetimeIPNode::metaEvaluate(this=0x0000000a96074000, context=0x000000016fdf6378, visitor=0x000000016fdf6c40) at RetimeIPNode.cpp:661:17 frame AcademySoftwareFoundation#4: 0x00000001007e1790 RV`IPCore::StackIPNode::metaEvaluate(this=0x0000000a94a2f100, context=0x000000016fdf67f0, visitor=0x000000016fdf6c40) at StackIPNode.cpp:791:25 frame AcademySoftwareFoundation#5: 0x000000010027b920 RV`IPCore::IPNode::metaEvaluate(this=0x0000000a9320d500, context=0x000000016fdf67f0, visitor=0x000000016fdf6c40) at IPNode.cpp:408:23 frame AcademySoftwareFoundation#6: 0x000000010040cefc RV`IPCore::GroupIPNode::metaEvaluate(this=0x0000000a95bf2080, c=0x000000016fdf67f0, visitor=0x000000016fdf6c40) at GroupIPNode.cpp:114:25 frame AcademySoftwareFoundation#7: 0x00000001003e8690 RV`IPCore::AdaptorIPNode::metaEvaluate(this=0x0000000a94774a00, c=0x000000016fdf67f0, visitor=0x000000016fdf6c40) at AdaptorIPNode.cpp:92:31 frame AcademySoftwareFoundation#8: 0x00000001003e8690 RV`IPCore::AdaptorIPNode::metaEvaluate(this=0x0000000a94cf8c80, c=0x000000016fdf67f0, visitor=0x000000016fdf6c40) at AdaptorIPNode.cpp:92:31 frame AcademySoftwareFoundation#9: 0x000000010040cefc RV`IPCore::GroupIPNode::metaEvaluate(this=0x0000000a95bf0500, c=0x000000016fdf67f0, visitor=0x000000016fdf6c40) at GroupIPNode.cpp:114:25 frame AcademySoftwareFoundation#10: 0x000000010027b920 RV`IPCore::IPNode::metaEvaluate(this=0x0000000a8cd01500, context=0x000000016fdf67f0, visitor=0x000000016fdf6c40) at IPNode.cpp:408:23 frame AcademySoftwareFoundation#11: 0x000000010027b920 RV`IPCore::IPNode::metaEvaluate(this=0x0000000a95bf0280, context=0x000000016fdf67f0, visitor=0x000000016fdf6c40) at IPNode.cpp:408:23 frame AcademySoftwareFoundation#12: 0x000000010040cefc RV`IPCore::GroupIPNode::metaEvaluate(this=0x0000000a95bf0a00, c=0x000000016fdf67f0, visitor=0x000000016fdf6c40) at GroupIPNode.cpp:114:25 frame AcademySoftwareFoundation#13: 0x00000001003e8690 RV`IPCore::AdaptorIPNode::metaEvaluate(this=0x0000000a94cf8780, c=0x000000016fdf67f0, visitor=0x000000016fdf6c40) at AdaptorIPNode.cpp:92:31 frame AcademySoftwareFoundation#14: 0x00000001003f4a40 RV`IPCore::DisplayStereoIPNode::metaEvaluate(this=0x0000000a95437480, context=0x000000016fdf69d0, visitor=0x000000016fdf6c40) at DisplayStereoIPNode.cpp:393:20 frame AcademySoftwareFoundation#15: 0x00000001003e8690 RV`IPCore::AdaptorIPNode::metaEvaluate(this=0x0000000a947bd180, c=0x000000016fdf69d0, visitor=0x000000016fdf6c40) at AdaptorIPNode.cpp:92:31 frame AcademySoftwareFoundation#16: 0x000000010027b920 RV`IPCore::IPNode::metaEvaluate(this=0x0000000a95b96800, context=0x000000016fdf69d0, visitor=0x000000016fdf6c40) at IPNode.cpp:408:23 frame AcademySoftwareFoundation#17: 0x000000010040cefc RV`IPCore::GroupIPNode::metaEvaluate(this=0x0000000a95bf0f00, c=0x000000016fdf69d0, visitor=0x000000016fdf6c40) at GroupIPNode.cpp:114:25 frame AcademySoftwareFoundation#18: 0x00000001003ee27c RV`IPCore::DisplayGroupIPNode::metaEvaluate(this=0x0000000a93e41500, context=0x000000016fdf6bb8, visitor=0x000000016fdf6c40) at DisplayGroupIPNode.cpp:425:21 frame AcademySoftwareFoundation#19: 0x000000010027b920 RV`IPCore::IPNode::metaEvaluate(this=0x0000000a94774780, context=0x000000016fdf6bb8, visitor=0x000000016fdf6c40) at IPNode.cpp:408:23 frame AcademySoftwareFoundation#20: 0x00000001006ce2b4 RV`IPMu::sourcesAtFrame(node_=0x000000013a849440, thread_=0x0000000117748e00) at CommandsModule.cpp:1937:28 frame AcademySoftwareFoundation#21: 0x0000000100e7a890 RV`Mu::DynamicArrayType::nodeEval(this=0x00000001176fcdc0, n=0x000000013a849440, thread=0x0000000117748e00) const at DynamicArrayType.cpp:78:90 frame AcademySoftwareFoundation#22: 0x0000000100076e80 RV`Mu::Node::eval(this=0x000000013a849440, t=0x0000000117748e00) const at Node.cpp:142:62 frame AcademySoftwareFoundation#23: 0x00000001000a92ac RV`Mu::Thread::call(this=0x0000000117748e00, f=0x0000000117e86d20, args=size=1, returnArguments=false) at Thread.cpp:422:23 frame AcademySoftwareFoundation#24: 0x00000001001d2e2c RV`TwkApp::MuSymbol_call(_self=0x000000011b52a7f0, args=0x000000030a0d8bb0, kwds=0x0000000000000000) at PyMuSymbolType.cpp:389:48 frame AcademySoftwareFoundation#25: 0x0000000108d17350 libpython3.11.dylib`_PyObject_MakeTpCall(tstate=0x00000001090a3c30, callable=0x000000011b52a7f0, args=<unavailable>, nargs=<unavailable>, keywords=0x0000000000000000) at call.c:214:18 [opt] frame AcademySoftwareFoundation#26: 0x0000000108d17850 libpython3.11.dylib`PyObject_Vectorcall [inlined] _PyObject_VectorcallTstate(tstate=<unavailable>, callable=<unavailable>, args=<unavailable>, nargsf=<unavailable>, kwnames=<unavailable>) at pycore_call.h:90:16 [opt] [artificial] frame AcademySoftwareFoundation#27: 0x0000000108df4dc0 libpython3.11.dylib`_PyEval_EvalFrameDefault(tstate=<unavailable>, frame=0x000000010a6f01d8, throwflag=<unavailable>) at ceval.c:0 [opt] frame AcademySoftwareFoundation#28: 0x0000000108df05fc libpython3.11.dylib`_PyEval_Vector [inlined] _PyEval_EvalFrame(tstate=0x00000001090a3c30, frame=0x000000010a6f01d8, throwflag=0) at pycore_ceval.h:73:16 [opt] frame AcademySoftwareFoundation#29: 0x0000000108df05ec libpython3.11.dylib`_PyEval_Vector(tstate=0x00000001090a3c30, func=<unavailable>, locals=<unavailable>, args=<unavailable>, argcount=<unavailable>, kwnames=<unavailable>) at ceval.c:6434:24 [opt] frame AcademySoftwareFoundation#30: 0x0000000108d19f24 libpython3.11.dylib`method_vectorcall [inlined] _PyObject_VectorcallTstate(tstate=0x00000001090a3c30, callable=0x000000013792e7a0, args=0x000000016fdf7ab0, nargsf=<unavailable>, kwnames=0x0000000000000000) at pycore_call.h:92:11 [opt] frame AcademySoftwareFoundation#31: 0x0000000108d19efc libpython3.11.dylib`method_vectorcall(method=<unavailable>, args=0x000000030a0d8268, nargsf=<unavailable>, kwnames=0x0000000000000000) at classobject.c:89:18 [opt] frame AcademySoftwareFoundation#32: 0x00000001001cf9d4 RV`TwkApp::PyFunctionAction::execute(this=0x0000000a8fade490, d=0x0000000a912c6000, event=0x000000016fdf7f28) const at PyFunctionAction.cpp:58:17 frame AcademySoftwareFoundation#33: 0x00000001006612a8 RV`TwkApp::Document::executeAction(this=0x0000000a912c6000, event=0x000000016fdf7f28) at Document.cpp:486:20 frame AcademySoftwareFoundation#34: 0x0000000100661b7c RV`TwkApp::Document::receiveEvent(this=0x0000000a912c6000, event=0x000000016fdf7f28) at Document.cpp:613:9 frame AcademySoftwareFoundation#35: 0x000000010066fc38 RV`TwkApp::EventNode::propagateEvent(this=0x0000000a912c6000, event=0x000000016fdf7f28) at EventNode.cpp:71:13 frame AcademySoftwareFoundation#36: 0x00000001001ecd94 RV`IPCore::Session::propagateEvent(this=0x0000000a912c6000, event=0x000000016fdf7f28) at Session.cpp:4644:43 frame AcademySoftwareFoundation#37: 0x000000010066fca8 RV`TwkApp::EventNode::propagateEvent(this=0x0000000a909c9800, event=0x000000016fdf7f28) at EventNode.cpp:82:24 frame AcademySoftwareFoundation#38: 0x000000010066fc00 RV`TwkApp::EventNode::sendEvent(this=0x0000000a909c9800, event=0x000000016fdf7f28) at EventNode.cpp:60:16 frame AcademySoftwareFoundation#39: 0x000000010038b604 RV`IPCore::IPGraph::inputsChanged(this=0x0000000a909c9800, n=0x0000000a8d8b2800) at IPGraph.cpp:3449:13 frame AcademySoftwareFoundation#40: 0x000000010027af8c RV`IPCore::IPNode::setInputs(this=0x0000000a8d8b2800, nodes=size=0) at IPNode.cpp:303:26 frame AcademySoftwareFoundation#41: 0x0000000100279c4c RV`IPCore::IPNode::removeInput(this=0x0000000a8d8b2800, n=0x0000000a94a30500) at IPNode.cpp:166:13 frame AcademySoftwareFoundation#42: 0x000000010027984c RV`IPCore::IPNode::~IPNode(this=0x0000000a94a30500) at IPNode.cpp:57:32 frame AcademySoftwareFoundation#43: 0x000000010040c900 RV`IPCore::GroupIPNode::~GroupIPNode(this=0x0000000a94a30500) at GroupIPNode.cpp:48:5 frame AcademySoftwareFoundation#44: 0x0000000100807490 RV`IPCore::SwitchGroupIPNode::~SwitchGroupIPNode(this=0x0000000a94a30500) at SwitchGroupIPNode.cpp:42:5 ``` Notice the destructors are triggering a sourcesAtFrame call, which then tries to access those same objects being destroyed resulting in a use-after-free crash. To fix this, we'll set a guard when the graph is being cleared, and if sourcesAtFrame is called when the guard is set, we'll return an empty array. This is the as what would happen if you called sourcesAtFrame when there were no sources (ie after the graph was cleared). So we'll just return the eventual result early, which is likely what the users want anyway, and this will protect python events from crashing RV. ### Describe the reason for the change. Prevent crashes ### Describe what you have tested and on which operating system. Mac OS 26.2 --------- Signed-off-by: Roger Nelson <roger.nelson@autodesk.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Linked issues
Summarize your change.
Describe the reason for the change.
Describe what you have tested and on which operating system.
Add a list of changes, and note any that might need special attention during the review.
If possible, provide screenshots.