[radarhere]

Pillow/src/PIL/ImageFile.py

Lines 462 to 468 in 60dcd68

	def load_prepare(self) -> None:
	# create image memory if necessary
	if self._im is None:
	self.im = Image.core.new(self.mode, self.size)
	# create palette (optional)
	if self.mode == "P":
	Image.Image.load(self)

This "create palette (optional)" code has been there since PIL. The effect would be to call Image.load() and potentially copy the palette from Python to C

Pillow/src/PIL/Image.py

Lines 984 to 987 in 60dcd68

	if self._im is not None and self.palette and self.palette.dirty:
	# realize palette
	mode, arr = self.palette.getdata()
	self.im.putpalette(self.palette.mode, mode, arr)

At minimum, it would be better to change the ImageFile condition to match the self.palette.dirty condition.

However, I don't think the palette needs to be copied here at all. None of our decoders use the C palette. It would be surprising if they read P data only to immediately transform it. A dirty palette can be left for the end of ImageFile's load() to handle.

Pillow/src/PIL/ImageFile.py

Line 460 in 60dcd68

return Image.Image.load(self)

So I suggest removing the call to Image.Image.load(self).