Search also for user fonts on Windows (#12954) #12957

JanSellner · 2018-12-09T18:31:44Z

PR Summary

This PR fixes #12954, i.e. the fonts installed to the user directory are now also found by matplotlib.

tacaswell · 2018-12-09T18:45:03Z

lib/matplotlib/font_manager.py

@@ -180,24 +184,46 @@ def win32InstalledFonts(directory=None, fontext='ttf'):

    items = set()
    for fontdir in MSFontDirectories:
+        # System fonts


The logic in this is complicated enough, can you put it in a helper function that we call twice?

although, I now see there is some subtlety with absolute vs relative paths.

tacaswell · 2018-12-09T18:50:13Z

Thanks for your work on this!

My comments are mostly directed at the long-term maintainability of this code path, we recently went through a few rounds of edits on those try/except blocks interacting with the registry so I am a bit concerned about having 2 copies of that code path (so if we do still have bugs in it we risk only fixing it in one place).

How hard would it be to add a test to appveyor (or to add a test and move the wintests to azure)?

anntzer · 2018-12-09T19:15:37Z

lib/matplotlib/font_manager.py

                    try:
-                        path = Path(directory, direc).resolve()
+                        path = Path(value).resolve()


I think this should still prepend a sensible base direc (which gets dropped anyways if value is absolute) or perhaps better just skip values that are not absolute to start with; otherwise non-absolute values get resolved relative to cwd which is certainly not correct.

anntzer · 2018-12-10T09:00:23Z

lib/matplotlib/font_manager.py

-        directory = win32FontDirectory()
+    Parameters
+    ----------
+    reg_domain : `int`


builtin types should not be backquoted in this position

anntzer · 2018-12-10T09:00:30Z

lib/matplotlib/font_manager.py


-    fontext = ['.' + ext for ext in get_fontext_synonyms(fontext)]
+    base_dir : `string`


JanSellner · 2018-12-10T09:08:12Z

Thanks @tacaswell and @anntzer for your reviews.

I also considered the duplication of the try blocks not optimal but since pathlib.Path does not modify absolute paths, this gets easier. I updated the code which now makes use of a helper function.

In case of the user fonts, I used the user directories in MSUserFontDirectories. When Microsoft decides to store only the font filename in the registry at one day, we are already prepared :P.

Regarding the tests, I cannot say too much about it. If a test can run in a Windows 1809 image with some user fonts installed, it should be easy to compare the result of findSystemFonts() with some ground truth. But you know more about your test infrastructure. However, since there is currently no test for this in the font manager, this is probably best done in a separate issue.

One final question: I guess it is a good idea to increase the version number since the behaviour changed and new fonts may now be found. Should I do this or do you match this number with the version of matplotlib somehow?

anntzer · 2018-12-10T09:14:07Z

You can set the version to e.g. '310', it doesn't have to match exactly the matplotlib version but something approaching it makes sense.
When does the registry get updated? e.g. if one just dumps a font in the user folder, does that immediately update the registry? If yes you could possibly write a test only-to-be-run-on-appveyor/azure, that copies matplotlib's shipped dejavusans to the user fonts directory, and check that it does indeed get picked up.

JanSellner · 2018-12-10T10:00:32Z

Just tested, the registry is not updated when a file is moved to the user folder. It has to be installed.

However, the test idea may still work since the font should also be detected if it is only stored in the folder and not in the registry (via fontpaths update iteration). I will look into it. Is there by any chance an existing test which does something similar you can point me to as a reference (e.g. to set it to run only on appveyor)?

anntzer · 2018-12-10T10:08:12Z

You can lookup os.environ.get("APPVEYOR", False).

anntzer · 2018-12-10T18:58:16Z

lib/matplotlib/tests/test_font_manager.py

+def test_user_fonts():
+    if not os.environ.get('APPVEYOR', False):
+        pytest.xfail('This test does only work on appveyor since user fonts '
+                     'are Windows specific')


Not only that, but you don't even want to run it on a dev's Windows computer to not overwrite their fonts directory.

Yes, that it true. I agree that respecting the user directories of other developers is a thing to keep in mind. Do you want me to change the message or was this only a mere side note?^^

Please do change the note too.

tacaswell · 2018-12-11T20:30:11Z

Thanks @Milania1 !

anntzer · 2019-01-13T17:06:45Z

lib/matplotlib/font_manager.py

-    'afm'.
-    """
-    import winreg
+def win32RegistryFonts(reg_domain, base_dir):


Please make this function private (underscore-prefixed).

anntzer · 2019-01-13T18:36:00Z

Also needs a rebase.

- Change note - Make function private

JanSellner · 2019-01-13T20:31:05Z

Ok, I applied the changes and rebased. Hopefully, everything is correct :-)

anntzer · 2019-01-13T20:41:34Z

It's win32RegistryFonts which should be private, not win32InstalledFonts (which should stay public to avoid API breakage...) :)

and make _win32RegistryFonts private

tacaswell · 2019-01-13T21:01:35Z

@Milania1 I took the liberty of pushing a commit that flips which function is private, hope you do not mind!

lib/matplotlib/font_manager.py

anntzer

anyone can merge after ci

tacaswell · 2019-01-14T04:12:33Z

Thanks @Milania1 ! Congratulations on your first contribution to matplotlib 🎉 Hopefully we will hear from you again!

tacaswell added this to the v3.1 milestone Dec 9, 2018

tacaswell reviewed Dec 9, 2018

View reviewed changes

anntzer reviewed Dec 9, 2018

View reviewed changes

anntzer reviewed Dec 10, 2018

View reviewed changes

lib/matplotlib/font_manager.py Outdated

fontext = ['.' + ext for ext in get_fontext_synonyms(fontext)]

base_dir : `string`

Copy link

Contributor

anntzer Dec 10, 2018

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

str

anntzer reviewed Dec 10, 2018

View reviewed changes

tacaswell approved these changes Dec 11, 2018

View reviewed changes

anntzer reviewed Jan 13, 2019

View reviewed changes

anntzer added the status: needs rebase label Jan 13, 2019

JanSellner added 5 commits January 13, 2019 19:56

Search also for user fonts on Windows (matplotlib#12954)

1fd871f

Use helper function to query the Windows registry for fonts

6d20c0b

Fix docstring types in win32RegistryFonts()

69b68ed

Add a test which checks wheter a font in the user directory is found

b7acfb2

Minor adjustments to font manager implementation

b02905a

- Change note - Make function private

JanSellner force-pushed the win-user-fonts branch from 31c823a to b02905a Compare January 13, 2019 19:06

anntzer removed the status: needs rebase label Jan 13, 2019

MNT: keep win32InstalledFonts public

9cac73a

and make _win32RegistryFonts private

anntzer reviewed Jan 13, 2019

View reviewed changes

lib/matplotlib/font_manager.py Outdated Show resolved Hide resolved

FIX: propagate the name change from previous commit through code

462dd7b

anntzer approved these changes Jan 13, 2019

View reviewed changes

anntzer merged commit a598811 into matplotlib:master Jan 13, 2019

RazrFalcon mentioned this pull request Jun 22, 2020

Load user fonts on Windows 10 linebender/resvg#299

Closed

RazrFalcon mentioned this pull request Feb 7, 2022

Load user fonts on Windows 10 RazrFalcon/fontdb#22

Closed


		fontext = ['.' + ext for ext in get_fontext_synonyms(fontext)]
		base_dir : `string`

Uh oh!

Search also for user fonts on Windows (#12954) #12957

Search also for user fonts on Windows (#12954) #12957

Uh oh!

Conversation

JanSellner commented Dec 9, 2018

PR Summary

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

tacaswell commented Dec 9, 2018

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

JanSellner commented Dec 10, 2018

Uh oh!

anntzer commented Dec 10, 2018

Uh oh!

JanSellner commented Dec 10, 2018

Uh oh!

anntzer commented Dec 10, 2018

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

tacaswell commented Dec 11, 2018

Uh oh!

Choose a reason for hiding this comment

Uh oh!

anntzer commented Jan 13, 2019

Uh oh!

JanSellner commented Jan 13, 2019

Uh oh!

anntzer commented Jan 13, 2019

Uh oh!

tacaswell commented Jan 13, 2019

Uh oh!

Uh oh!

anntzer left a comment

Choose a reason for hiding this comment

Uh oh!

tacaswell commented Jan 14, 2019

Uh oh!

Uh oh!