ImageOps.expand error with tuple border

[ProgrammerPeter]

What did you do?

Call ImageOps.expand with a tuple border.

What did you expect to happen?

Input image will be expanded.

What actually happened?

It raises a TypeError that says in draw_rectangle function, width requires an integer but got a tuple.

What are your OS, Python and Pillow versions?

• OS: Ubuntu
• Python: 3.7.5
• Pillow: 8.3.1

I find in PR #5552 ImageOps.expand was modified and add two lines of code to draw the output, which are not necessary.
I guess if it is code for debug. If not, please notice that border may be a tuple and can not be passed to draw.rectangle as width.

def expand(image, border=0, fill=0):
    """
    Add border to the image
    :param image: The image to expand.
    :param border: Border width, in pixels.
    :param fill: Pixel fill value (a color value).  Default is 0 (black).
    :return: An image.
    """
    left, top, right, bottom = _border(border)
    width = left + image.size[0] + right
    height = top + image.size[1] + bottom
    color = _color(fill, image.mode)
    if image.mode == "P" and image.palette:
        out = Image.new(image.mode, (width, height))
        out.putpalette(image.palette)
        out.paste(image, (left, top))

        draw = ImageDraw.Draw(out)
        draw.rectangle((0, 0, width, height), outline=color, width=border) # TypeError from here
    else:
        out = Image.new(image.mode, (width, height), color)
        out.paste(image, (left, top))
    return out

[ProgrammerPeter]

@radarhere may know this.

[radarhere]

Pillow/src/PIL/ImageOps.py

Lines 402 to 403 in 696b82e

	draw = ImageDraw.Draw(out)
	draw.rectangle((0, 0, width - 1, height - 1), outline=color, width=border)

It's not for debugging purposes. These lines are there to draw the fill color. If color is an RGB color, then it needs to be looked up in the image palette. This needs to be done on a new palette (to avoid modifying the original image if an additional paletter entry needs to be allocated). Without this draw operation, the border color would just be the zero index.

I've created #5615 to resolve this issue by changing the strategy. Instead, the PR creates a copy of the palette, calculates the palette index from that and uses it for the background color, and then attaches the new palette to the output image.