Remove deprecations for Pillow 10.0.0

[hugovk]

Remove deprecations scheduled for removal in Pillow 10.0.0:

• Tk/Tcl <= 8.4 support was deprecated in 8.2.0
  Deprecate Tk/Tcl 8.4, to be removed in Pillow 10 (2023-07-01) #5216

• im.category and related Image.NORMAL, Image.SEQUENCE and Image.CONTAINER attributes, deprecated in 8.2.0
  Deprecated categories #5351

• JpegImagePlugin.convert_dict_qtables, deprecated in 8.3.0
  De-zigzag JPEG's DQT when loading; deprecate convert_dict_qtables #4989

• ImagePalette size parameter, deprecated in 8.4.0
  Deprecate ImagePalette size parameter #5641

• ImageShow.Viewer.show_file() file argument, deprecated in 9.1.0
  Deprecated show_file "file" argument in favour of "path" #5959

• Constants deprecated in 9.1.0
  Added enums, deprecate constants #5954

• FitsStubImagePlugin, deprecated in 9.1.0
  Added FITS reading, deprecate FitsStubImagePlugin #6056

• FreeTypeFont.getmask2() fill parameter, deprecated in 9.2.0
  Deprecate FreeTypeFont.getmask2 fill parameter #6220

• PhotoImage.paste() box parameter, deprecated in 9.2.0
  Deprecated PhotoImage.paste() box parameter #6178

• Qt5 support, deprecated in Pillow 9.2.0
  Deprecate support for Qt 5 (PyQt5 and PySide2) #6237

• Image.coerce_e, deprecated in 9.2.0
  Support more affine expression forms in im.point(), deprecate Image.coerce_e #6254

TODO in followup:

• Font size and offset methods
  Deprecate ImageFont.getsize and related functions #6381

• Release notes

[hugovk]

Will remove the other deprecations here too.

[hugovk]

Question about 6369c80 (#7059):

• ImageShow.Viewer.show_file() file argument, deprecated in 9.1.0
  Deprecated show_file "file" argument in favour of "path" #5959

Should we also do this?

-    def show_file(self, path=None, **options):
+    def show_file(self, path, **options):
         """
         Display given file.
         """
-        if path is None:
-            msg = "Missing required argument: 'path'"
-            raise TypeError(msg)
         os.system(self.get_command(path, **options))  # nosec
         return 1

[hugovk]

Question about 4c9f8b4 (#7059):

• Image.coerce_e, deprecated in 9.2.0
  Support more affine expression forms in im.point(), deprecate Image.coerce_e #6254

class _E has this comment:

# _E(scale, offset) represents the affine transformation scale * x + offset.
# The "data" field is named for compatibility with the old implementation,
# and should be renamed once coerce_e is removed.

What should it be renamed to?

[radarhere]

Question about 6369c80 (#7059):

• ImageShow.Viewer.show_file() file argument, deprecated in 9.1.0
  Deprecated show_file "file" argument in favour of "path" #5959

Should we also do this?

-    def show_file(self, path=None, **options):
+    def show_file(self, path, **options):
         """
         Display given file.
         """
-        if path is None:
-            msg = "Missing required argument: 'path'"
-            raise TypeError(msg)
         os.system(self.get_command(path, **options))  # nosec
         return 1

Yes, sounds like good to me. The default value of None was only added in #5959 - my intention would have been just to allow both path and file to work during the deprecation period.

[radarhere]

Question about 4c9f8b4 (#7059):

• Image.coerce_e, deprecated in 9.2.0
  Support more affine expression forms in im.point(), deprecate Image.coerce_e #6254

class _E has this comment:

# _E(scale, offset) represents the affine transformation scale * x + offset.
# The "data" field is named for compatibility with the old implementation,
# and should be renamed once coerce_e is removed.

What should it be renamed to?

From my reading of the comment, the user's plan was for it to be renamed "offset"

[nulano]

My reading of the comment is the following change:

-# _E(scale, offset) represents the affine transformation scale * x + offset.
-# The "data" field is named for compatibility with the old implementation,
-# and should be renamed once coerce_e is removed.
 class _E:
-    def __init__(self, scale, data):
+    def __init__(self, scale, offset):
         self.scale = scale
-        self.data = data
+        self.offset = offset

(which I think is the same as @radarhere's comment, but I didn't understand it until I posted this comment)

[nulano]

I have made a few suggestions for textsize tests: hugovk#100

[nulano]

In #6195 (comment) there was a suggestion to make this function, and possibly others, use keyword-only arguments. I see that I was planning to deprecate positional arguments for other text-related functions, but it seems I didn't get to it in time for removal in Pillow 10.

One option would be to remove support for positional arguments here by replacing fill with *, and to deprecate positional arguments in related functions, so that they will be consistent in Pillow 11.

Removing positional argument support here would simplify updating the API while preserving backwards compatibility, and allow the documentation to be in a logical or alphabetical order rather than version-added order.

However, this does not have to be decided in this PR. Perhaps it might be better to discuss in a separate issue/PR - perhaps I could make a separate deprecation PR once this is merged to discuss.

[hugovk]

Yeah, let's discuss this in another issue or PR.

[nulano]

It looks like this is slightly different when stroke_width != 0. The deprecated function FreeTypeFont.getsize seems to have an additional issue that I had not noticed before: it was adding the stroke width to the "height" twice even though it was returning the bottom coordinate instead of the actual height.

This has the effect of changing multiline spacing due to the change (see the changed image for an example). There are three options on how to proceed, and it is not clear to me whether any one of them is better than the others:

• Do not consider stroke width for line spacing (i.e. remove the parameter). I believe this would be the best visually, I would not expect a varying stroking width to affect the text layout.
• Consider 2x stroke width for line spacing (i.e. add stroke width to the returned value). This would preserve exact backwards compatibility. It would extend the line spacing by the size of stroke both on the top and bottom side.
• Consider 1x stroke width for line spacing (i.e. the current proposed change). This is the simplest option, and avoids mixing ImageDraw and ImageFont behaviour.

An alternative suggestion could be to move line spacing calculations to the font class, which can have other (including non backwards compatible) benefits, but I'm not sure what that API would look like.

[hugovk]

Shall we drop these ImageFont removals from this PR and we can consider all this separately?

The other changes here fix a failure with the new 3.12.0a7 (#7072 (comment)), so it'd be good to get that merged.

[radarhere]

Sure, whatever you prefer.

[hugovk]

Okay, made a copy of this branch and pushed to https://github.com/hugovk/Pillow/tree/rm-qt5a, then dropped the last two commits here (ac23f4f, 7712062), then rebased on main (to cleanup the fixups) and force-pushed.

[hugovk]

@nulano Please could you make a followup for ImageFont? Thank you!

[radarhere]

I've created PR #7080 to help this along - feel free to suggest changes.

[radarhere]

I've created PR #7080 as the followup PR.