I believe there is an ambiguity for the case of mode=pad and scale=down.

The spec states:

mode=pad: the image will be evenly padded with whitespace or transparency to become exactly the specified size while maintaining aspect ratio

scale=down: If the constraints are larger than the source image, the resulting image will use the source dimensions instead, foregoing any cropping, padding, or stretching

I am not clear on the desired behavior for the following two test cases given a target size of 800x600:

Source size of 400x300, should output size be:

• 800x600 with padding all around?
• 400x300 with no padding?

Source size of 1200x300, should output size be:

• 800x600, with padding on sides?
• 800x200, with no padding?