[qodelicious]

Saving an L image as Palm fails right now.

from PIL import Image
Image.linear_gradient("L").save("out.palm", bpp=4)
# ValueError: image has no palette

The L branch switches im._mode to "P" so the P;1, P;2 or P;4 rawmode can be used, and since #8685 the colormap code checks im.mode == "P", so it also runs for these images and calls getpalette() on an image that has none. The same happens when bpp comes from im.info.

Changes proposed in this pull request:

• Only write the custom colormap when saving a real P image (rawmode == "P"). Grayscale Palm images don't use one, which matches the "we ignore the palette here" comment.
• Added tests that check the header and the packed pixel bytes for bpp 1, 2 and 4, and for info["bpp"]
• Documented the bpp save option and which modes can be saved, since it wasn't in the handbook

I also converted the bpp 1, 2 and 4 output of hopper with ImageMagick and it matched the expected gray levels exactly. Palm tests pass and ruff, black and sphinx-lint are happy.

[radarhere]

The L branch switches im._mode to "P"

I actually think that is a problem. Saving an image shouldn't change it like that. I've pushed a commit to keep it in L mode, and fix the rawmode instead.

I'm not a huge fan of testing specific bytes in the saved image. If we're writing specific bytes, and then checking those values specifically again, it doesn't seem like an effective way to know if we did anything incorrect. Instead, I've pushed a commit to check the image mode and similarity to the original image. See what you think.