Update hooks for pyqtgraph #501
Conversation
Test that colormap data is available in the frozen application.
Collect all data files from `pyqtgraph` package, except for the data files that come with the examples.
Add a test for pyqtgraph's RemoteGraphicsView, which makes use of `pyqtgraph.multiprocess`.
Add a runtime hook that attempts to detect arguments used by `pyqtgraph.multiprocess` to spawn its subprocess worker. Requires the user to call `multiprocessing.freeze_support()` in the entry-point script. Additionally, in onefile builds, the user needs to set (restore) `_MEIPASS2` environment variable to the value of `sys._MEIPASS` to prevent the onefile executable from unpacking itself again in the subprocess (due to it being spawned by `pyqtgraph.multiprocess` via `subprocess.Popen()` or equivalent).
|
Here's oneshot test for the added tests: https://github.com/rokm/pyinstaller-hooks-contrib/actions/runs/3284554698 On linux, the oneshot workflow lacks the apt-installed packages required by |
| # pyqtgraph.multiprocess also uses a subprocess.Popen() to spawn its | ||
| # sub-process, so we need to restore _MEIPASS2 to prevent the executable | ||
| # to unpacking itself again in the subprocess. | ||
| os.environ['_MEIPASS2'] = sys._MEIPASS |
There was a problem hiding this comment.
Any reason why this can't go in a runtime hook?
There was a problem hiding this comment.
Yeah, it would potentially cause issues when trying to run a different PyInstaller-frozen executable from the program, which is the main reason why the environment variable is cleared in the first place.
I think the only way to handle this automatically in the safe way is if we manage to patch subprocess.Popen to detect that it is trying to run sys.executable and restore/set _MEIPASS2 in the subprocess' environment.
There was a problem hiding this comment.
Yeah, it would potentially cause issues when trying to run a different PyInstaller-frozen executable from the program, which is the main reason why the environment variable is cleared in the first place.
Sure, the same problem arises if the user sets the environment variable in their code, but they should know whether their program will also try to spawn a subprocess with a different pyinstaller-frozen executable or not.
Update
pyqtgraphstandard hook to collect all data from the package, including the colormaps and their associated license files, as per #490 (comment).Create a new runtime hook that attempts to handle subprocess spawned by
pyqtgraph.multiprocess. The caveat is that the user still needs to call stdlibmultiprocessing.freeze_supportin their entry-point script (which is something I hope to get around to addressing in PyInstaller'smultiprocessinghook). And for onefile builds, they also need to set_MEIPASS2environment variable to prevent the executable from unpacking itself again in the subprocess (using differentsys._MEIPASS, and thus breaking the path check in the runtime hook). This is becausepyqtgraph.multiprocessusessubprocess.Popento spawn the process; perhaps we could extend oursubprocessruntime hook to detect whensys.executableis spawned and automatically restore_MEIPASS2in the subprocess' environment?Replaces and closes #490.