xref to#24001

I think that this or something similar needs to make it into 3.7 to fix#24641 being partially incorrect; or#24641 needs to be temporarily reverted for 3.7.

____________________________ test_user_fonts_win32 ____________________________[gw1] win32 -- Python 3.10.8 C:\hostedtoolcache\windows\Python\3.10.8\x64\python.exe    @pytest.mark.skipif(sys.platform != 'win32', reason='Windows only')    def test_user_fonts_win32():        if not (os.environ.get('APPVEYOR') or os.environ.get('TF_BUILD')):            pytest.xfail("This test should only run on CI (appveyor or azure) "                         "as the developer's font directory should remain "                         "unchanged.")            font_test_file = 'mpltest.ttf'            # Precondition: the test font should not be available        fonts = findSystemFonts()        if any(font_test_file in font for font in fonts):            pytest.skip(f'{font_test_file} already exists in system fonts')            user_fonts_dir = MSUserFontDirectories[0]            # Make sure that the user font directory exists (this is probably not the        # case on Windows versions < 1809)        os.makedirs(user_fonts_dir)            # Copy the test font to the user font directory        shutil.copy(Path(__file__).parent / font_test_file, user_fonts_dir)            # Now, the font should be available        fonts = findSystemFonts()>       assert any(font_test_file in font for font in fonts)E       assert FalseE        +  where False = any(<generator object test_user_fonts_win32.<locals>.<genexpr> at 0x000002364CB2A8F0>)

https://dev.azure.com/matplotlib/matplotlib/_build/results?buildId=30184&view=logs&j=595c9a33-3ac5-58f0-6ce6-44c63ffdfe86&t=6405dc77-f20a-51d5-22e4-a611ccc41f2e&l=182

The failure is definitely caused by this change

Also, I dunno if this is sensible but I modified the test to run it on my local machine, and I'm now getting the same failure, w/ the same message:

@pytest.mark.skipif(sys.platform!='win32',reason='Windows only')deftest_user_fonts_win32():font_test_file='mpltest.ttf'# Precondition: the test font should not be availablefonts=findSystemFonts()ifany(font_test_fileinfontforfontinfonts):pytest.skip(f'{font_test_file} already exists in system fonts')user_fonts_dir=Path(MSUserFontDirectories[0])# Make sure that the user font directory exists (this is probably not the# case on Windows versions < 1809)user_fonts_dir.mkdir(exist_ok=True)# Copy the test font to the user font directoryshutil.copy(Path(__file__).parent/font_test_file,user_fonts_dir)assert (user_fonts_dir/font_test_file).exists()# Now, the font should be availablefonts=findSystemFonts()assertany(font_test_fileinfontforfontinfonts)    (Path(user_fonts_dir)/font_test_file).unlink()assertnot (Path(user_fonts_dir)/font_test_file).exists()

ETA: if this is sensible, then I'll wrap up "temp copy" in a context manager.

ETA2: I have no idea why the test is failing b/c the directory for writing the tempfile is identical when fontpaths is set or when it's empty

C:\Users\story\AppData\Local\Microsoft\Windows\Fonts\mpltest.ttf
C:\Users\story\AppData\Local\Microsoft\Windows\Fonts\mpltest.ttf

while this is flipped:

findSystemFonts folders: {WindowsPath('C:/Windows/Fonts'), WindowsPath('C:/Users/story/AppData/Local/Microsoft/Windows/Fonts')}
findSystemFonts folders: {WindowsPath('C:/Users/story/AppData/Local/Microsoft/Windows/Fonts'), WindowsPath('C:/Windows/Fonts')}

and I checked and yeah the file is in AppData

Also, I dunno if this is sensible but I modified the test to run it on my local machine, and I'm now getting the same failure, w/ the same message:

Ah, are you saying the test is wrong? I think we maybe should be callingAddFontResourceEx for that test instead of directly copying the file.

Ah, are you saying the test is wrong?

No, just that I'm not sure why I can't run it locally if I create and clean-up the file during the test?

I guess what I would like to confirm is; if you use Explorer to copy a new font there, does it get added to the registry? Does it only do so if you use the Install button on the font viewer? What about if you have a Python script that copies the font there; does the registry update automatically, or do you have to open the directory in Explorer to refresh it, or does nothing update the registry?

if you use Explorer to copy a new font there, does it get added to the registry?

No
Also "'SOFTWARE\Microsoft\Windows\CurrentVersion\Fonts'" doesn't exist at least on my machine

Does it only do so if you use the Install button on the font viewer?

Yup

What about if you have a Python script that copies the font there; does the registry update automatically,

nope

or do you have to open the directory in Explorer to refresh it

that doesn't seem to work either

ETA: as far as I can tell, that test works because it's checking if a file put into a folder is found in the folder, completely bypassing the registry.

Per discussion on this weeks dev call:

• go with windows registry only
• xfail the test
  • defer questions of how to test this to later
  • do we want to change the registry as part of the test?
  • opt in to running locally?
  • flag to only run an CI + windows?
• need a manual test that this works
  • download a new font
  • use windows way to install it
  • remove mpl's font cache
  • verify the new font can be found
• add behavior change note

@story645 Can you please confirm that this works as expected for fonts installed as Windows expects and merge if you are satisfied?

heads up that I won't be able to get to this til tomorrow night/sunday