add a hook for qtmodern

[sanzoghenzo]

closes #304

[bwoodsend]

Ughh, I'm seeing Windows+onefile only failures. This looks like a nightmare to debug.

[rokm]

I don't think the proposed import-only test will catch the missing data files, as those are accessed only when qtmodern.windows.ModernWindow is instantiated (resources/frameless.qss) or when qtmodern.style.dark() or light() is called (resources/style.qss). So the test should make use of those, e.g.,

import sys

from PyQt5 import QtCore, QtWidgets
import qtmodern.windows
import qtmodern.styles

app = QtWidgets.QApplication(sys.argv)

window = QtWidgets.QWidget()
window.setWindowTitle("Hello world!")
modern_window = qtmodern.windows.ModernWindow(window)
modern_window.show()

qtmodern.styles.dark(app)

QtCore.QTimer.singleShot(1000, app.quit)

app.exec()

or even without the last two calls, as we do not really need to start the event loop (nor quit it via timer) for this test.

[rokm]

Ughh, I'm seeing Windows+onefile only failures. This looks like a nightmare to debug.

It's not onefile per se that's problematic, but rather the PyQt5-based test that comes second. Which probably means that PyQt5 was first imported within the first test, and the second test somehow lost the PATH modification that happened during the import...

[rokm]

And the "somehow" part boils down to pyi_builder fixture using monkeypatch...

I suppose if we want to use PyQt5 in tests here, we will have to do explicit import outside of the tests, same as we do in main repository's tests...

[bwoodsend]

Something I never understood - where do the PyQt5 DLL paths get added to the master process's PATH? Surely, it should all happen in subproceses?

[rokm]

Something I never understood - where do the PyQt5 DLL paths get added to the master process's PATH? Surely, it should all happen in subproceses?

Yeah, it should...

But what actually happens is that get_module_file_attribute (called from here) is not properly isolated. I.e., if you call it with top-level package name, all is well, but if we call get_module_file_attribute('PyQt5.QtCore'), that ends up importing PyQt5...

>>> from PyInstaller.utils.hooks import get_module_file_attribute
>>> import sys
>>> 'PyQt5' in sys.modules
False
>>> get_module_file_attribute('PyQt5')
'[...]/lib64/python3.9/site-packages/PyQt5/__init__.py'
>>> 'PyQt5' in sys.modules
False
>>> get_module_file_attribute('PyQt5.QtCore')
'[...]/lib64/python3.9/site-packages/PyQt5/QtCore.abi3.so'
>>> 'PyQt5' in sys.modules
True

[bwoodsend]

Ahh so get_module_file_attribute('PyQt5.QtCore') is the leak. In which case it should jump straight to the subprocess behaviour if the module name contains a ..

[rokm]

Ahh so get_module_file_attribute('PyQt5.QtCore') is the leak. In which case it should jump straight to the subprocess behaviour if the module name contains a ..

Would there be too much overhead if we @isolated.decorate the whole function, and removed the exec_statement part (i.e, run the statement's code "directly")?

[rokm]

Ahh so get_module_file_attribute('PyQt5.QtCore') is the leak. In which case it should jump straight to the subprocess behaviour if the module name contains a ..

Would there be too much overhead if we @isolated.decorate the whole function, and removed the exec_statement part (i.e, run the statement's code "directly")?

Although that's applicable only to develop; if we want this also fixed in v4, we should go with the fix you suggested.

[bwoodsend]

@sanzoghenzo I'm afraid you're unlucky in that you've just put your foot into a hornet's nest. For now can you put:

if is_win:
    # This is a hack to prevent monkeypatch from interfering with PyQt5's additional PATH entries. See:
    # https://github.com/pyinstaller/pyinstaller/commit/b66c9021129e9e875ddd138a298ce542483dd6c9
    try:
        import PyQt5
    except ImportError:
        pass

just above your test_qtmodern(). That should fix the failures.

[sanzoghenzo]

@sanzoghenzo I'm afraid you're unlucky in that you've just put your foot into a hornet's nest.

Yeah, usually I use conda to install these packages ad avoid these troubles on windows...
I just re-read the discussion and I get that the issue is not what I was thinking, even though I don't get the error on my conda environment on windows...

I'll fix it ASAP

[bwoodsend]

src/_pyinstaller_hooks_contrib/tests/test_libraries.py:747:9: F401 'PyQt5' imported but unused

Ughh, better make that import PyQt5 # noqa: F401.

[bwoodsend]

Ahh, you're ahead of me...

[bwoodsend]

And here's a CI run. The Linux ones failed but that's just because I hadn't enabled Xfvb virtual screen. They pass locally.

[bwoodsend]

Nice work @sanzoghenzo.